1

THE STUDY OF
INFORMATION
INTERDISCIPLINARY MESSAGES

Edited by
Fritz Machiup
and
Una Mansfield
 THE CONTRIBUTORS

Michael A. Arbib
Avron Barr
David Batty
Robert D. Beam
Toni Carbo Bearman
Robert C. Berwick
Thomas G. Bever
NUNC COCNOSCO EX PARTE
Margaret A. Boden
Kenneth E. Boulding
Walter Buckley
C. West Churchman
Richard H. Day
Ray C. Dougherty
THOMASJ. BATA LIBRARY
Murray Eden
TRENT UNIVERSITY
Peter Elias
Michael S. Cazzaniga
Saul Gorn
Douglas R. Hofstadter
Samuel Jay Keyser
Manfred Kochen
Richard N. Langlois
Fritz Machlup
Donald M. MacKay
Una Mansfield
Richard Mattessich
Mihajlo D. Mesarovic
George A. Miller
Elliott W. Montroll
Hassan Mortazavian
Joel Moses
Allen Newell
Charls Pearson
Alan J. Perlis
Zenon W. Pylyshyn
W. Boyd Rayward
Jesse FT Shera
Vladimir Slamecka
Myron Tribus
Peter Wegner
Patrick Wilson
Vladimir Zwass
 Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation

https://archive.org/details/studyofinformatiOOOOunse
 THE STUDY OF
INFORMATION
 THE STUDY OF

INFORMATION
Interdisciplinary Messages

Edited by

FRITZ MACHLUP AND UNA MANSFIELD

with a Foreword by

GEORGE A. MILLER

A Wiley-lnterscience Publication

JOHN WILEY & SONS

New York Chichester • Brisbane • Toronto Singapore

Most of the papers included in this volume were obtained
with the support of the National Science Foundation
Grant No. IST-80-08007, A02. However, any opinions,
findings, conclusions, and/or recommendations ex¬
pressed herein are those of the authors and do not
necessarily reflect the views of the NSF.

Copyright © 1983 by John Wiley & Sons, Inc.

All rights reserved. Published simultaneously in Canada.

Reproduction or translation of any part of this work

beyond that permitted by Section 107 or 108 of the
1976 United States Copyright Act without the permission
of the copyright owner is unlawful. Requests for
permission or further information should be addressed to
the Permissions Department, John Wiley & Sons, Inc.

Library of Congress Cataloging in Publication Data

Main entry under title:
The Study of Information.

“A Wiley-Interscience publication.”
Bibliography: p.
Includes index.
1. Information science—Addresses, essays, lectures.
I. Machlup, Fritz, 1902-1983. II. Mansfield, Una.

Z665.S826 1983 020 83-12147

ISBN 0-471-88717-X

Printed in the United States of America

10 987654321
 CONTRIBUTORS

[Editors' Note: The following list does not include past affiliations, honors, and awards or major
publications of contributors. Many of these publications are included in the Cumulative List of
References. Our main objective is to show the major disciplinary commitments of the authors;
we therefore include their past and present offices (but not council memberships) in national
professional associations, and editorial duties and editorial-board memberships for learned
journals and professional periodicals.]

MICHAEL A. ARBIB, Professor of Computer and Information Science,

and Director, Center for Systems Neuroscience, University of Massa¬
chusetts, Amherst; associate editor, Cognition and Brain Theory; mem¬
ber of the editorial boards, International Journal of Man-Machine Stud¬
ies, Journal of Mathematical Biology, Journal of Semantics, Journal of
Social and Biological Structures, and Cybernetics and Systems
AVRON BARR, Research Associate, Department of Computer Science,
Stanford University
DAVID BATTY, President, CDB Enterprises, Inc., Silver Spring, Mary¬
land; Visiting Professor, University of Maryland.
ROBERT D. BEAM, Associate Professor of Economics, and Director of the
Center for Economic Education, University of Wisconsin, Superior,
Wisconsin
TONI CARBO BEARMAN, Executive Director, National Commission on
Libraries and Information Science, Washington, D.C.; former editor,
NFAIS Newsletter; member of the editorial boards, Bulletin of the
American Society for Information Science, The Network Librarian, Re¬
source Sharing and Library Networks, and Serials Librarian
ROBERT C. BERWICK, Assistant Professor of Computer Science and En¬
gineering, M.I.T. Artificial Intelligence Laboratory, Cambridge, Massa¬
chusetts; member of the editorial board, Linguistic Inquiry
THOMAS G. BEVER, Professor of Psychology and Linguistics, Columbia
University, New York; co-founder and associate editor, Cognition
MARGARET A. BODEN, Professor of Philosophy and Psychology, Uni¬
versity of Sussex (U.K.); member of the editorial board. Cognition and
Brain Theory; associate of Behavioral and Brain Sciences; member of
executive committee of Mind Association
VI CONTRIBUTORS

KENNETH E. BOULDING, Distinguished Professor Emeritus of Econom¬

ics, University of Colorado, Boulder; Research Associate and Project
Director, Institute of Behavioral Science, University of Colorado, Boul¬
der; past president, American Economic Association, and American
Association for the Advancement of Science
WALTER BUCKLEY, Professor of Sociology, University of New Hamp¬
shire, Durham
C. WEST CHURCHMAN, Acting Chairman, Center for Research in Man¬
agement, and Professor Emeritus of Business Administration, Univer¬
sity of California, Berkeley; past president, The Institute of Manage¬
ment Sciences; former editor-in-chief, Philosophy of Science, and
Management Science; member of the editorial boards, The Information
Society, Journal of Management Studies, OMEGA, and Research
Policy
RICHARD H. DAY, Professor of Economics, and Director, Modelling Re¬
search Group, University of Southern California, Los Angeles; co¬
editor, The Journal of Economic Behavior and Organization; former
associate editor, Journal of Economic Dynamics and Control
RAY C. DOUGHERTY, Associate Professor of Linguistics, New York Uni¬
versity; former member of the board of editors, Linguistic Analysis
MURRAY EDEN, Chief, Biomedical Engineering and Instrumentation
Branch, National Institutes of Health, Bethesda, Maryland; Professor
of Electrical Engineering, Emeritus, M.I.T.; editor, Information and
Control
PETER ELIAS, Edwin S. Webster Professor of Electrical Engineering,
Massachusetts Institute of Technology; former editor and current mem¬
ber of the editorial board, Information and Control; member of the
editorial board, Proceedings of the IEEE; former member of the edito¬
rial board, SPECTRUM
MICHAEL S. GAZZANIGA, Professor of Psychology in Neurology, and
Director of Cognitive Neuroscience, Cornell University Medical Col¬
lege, New York City
SAUL GORN, Professor, Computer and Information Science Department,
Moore School of Electrical Engineering, School of Engineering and
Applied Science, University of Pennsylvania, Philadelphia
DOUGLAS R. HOFSTADTER, Associate Professor of Computer Science,
Indiana University, Bloomington, Indiana
SAMUEL JAY KEYSER, Professor and Head, Department of Linguistics
and Philosophy, and Director, Center for Cognitive Science, M.I.T.,
Cambridge, Massachusetts; editor, Linguistic Inquiry
MANFRED KOCHEN, Professor of Information Science, Research Mathe¬
matician, and Adjunct Professor of Computer and Information Systems,
University of Michigan, Ann Arbor, Michigan; managing editor, Hu-
CONTRIBUTORS VII

man Systems Management; former associate editor, Journal of the

ACM and Behavioral Science
RICHARD N. LANGLOIS, Assistant Professor of Economics, University
of Connecticut, Storrs
FRITZ MACHLUP (deceased January 1983) former Professor of Econom¬
ics, New York University, and Walker Professor of Economics and
International Finance Emeritus, Princeton University; past president,
American Association of University Professors, American Economic
Association, and International Economic Association; former acting
managing editor and member of the editorial board, American Eco¬
nomic Association; former member of the editorial boards, Knowledge:
Creation, Diffusion, Utilization, Rivista Internazionale di Scienze Eco-
nomiche, and Public Finance
DONALD M. MACKAY, Emeritus Professor, Department of Communica¬
tion and Neuroscience, University of Keele, Staffordshire (U.K.); joint
editor, Experimental Brain Research and Biological Cybernetics
UNA MANSFIELD, Visiting Research Faculty, Department of Economics,
Princeton University, Princeton, New Jersey
RICHARD MATTESSICH, Distinguished Arthur Andersen and Company
Alumni Professor at the University of British Columbia, Vancouver
(Canada); associate editor, Journal of Business Administration; mem¬
ber of the editorial board, Economia Aziendale (Accademia Italiana di
Economia Aziendale)
MIHAJLO D. MESAROVIC, Cady Staley Professor, Department of Sys¬
tems Engineering, Case Western Reserve University, Cleveland, Ohio;
member of the editorial boards, Mathematical Systems Theory and In¬
formation Sciences
GEORGE A. MILLER, James S. McDonnell Distinguished University Pro¬
fessor of Psychology, Princeton University; past president, American
Psychological Association and Eastern Psychological Association; for¬
mer editor, Psychological Bulletin
ELLIOTT W. MONTROLL, Professor at the Institute of Physical Science
and Technology, University of Maryland, College Park; editor, Journal
of Mathematical Physics and Studies in Statistical Mechanics; former
member of the editorial boards, Science and Proceedings of the NAS
[National Academy of Sciences]; current member of the editorial board,
Journal of Physics and Chemistry of Solids
HASSAN MORTAZAVIAN, Research Associate, Department of System
Science, School of Engineering and Applied Science, University of
California, Los Angeles
JOEL MOSES, Professor and Head, Department of Electrical Engineering
and Computer Science, M.I.T., Cambridge, Massachusetts; former
editor, Transactions on Mathematical Software of the Association for
Computing Machinery
viii CONTRIBUTORS

ALLEN NEWELL, University Professor of Computer Science, Carnegie-

Mellon University, Pittsburgh, Pennsylvania; past president, American
Association for Artificial Intelligence
CHARLS PEARSON, President, Catronix Corporation, Atlanta, Georgia;
associate editor for empirical semiotics, Journal of Empirical Esthetics,
and Mind, Brain, and Machine
ALAN J. PERLIS, Eugene Higgins Professor of Computer Science, Yale
University, New Haven, Connecticut; past president, Association for
Computing Machinery; former editor-in-chief, Communications of the
ACM
ZENON W. PYLYSHYN, Professor of Psychology and Computer Science,
and Director of the Centre for Cognitive Science, The University of
Western Ontario, London (Canada); past president, Society for Philoso¬
phy and Psychology; associate editor, Behavioral and Brain Sciences;
member of the editorial boards, Cognitive Science, Cognitive Psychol¬
ogy, Cognition, and Artificial Intelligence
W. BOYD RAYWARD, Professor and Dean, The Graduate Library School,
University of Chicago; former editor and current member of the edito¬
rial board, The Library Quarterly
JESSE H. SHERA (deceased March 1982) former Professor and Dean
Emeritus, Graduate Library School, Case Western Reserve University,
Cleveland, Ohio; past president, Association of American Library
Schools and Beta Phi Mu; former editor, American Documentation,
Information Systems in Documentation, and Documentation in Action;
former associate editor, The Library Quarterly and Journal of Catalog¬
ing and Classification; former member of the editorial board, Library
Science
VLADIMIR SLAMECKA, Professor, School of Information and Computer
Science, Georgia Institute of Technology, Atlanta, Georgia; former
chairman, United States National Committee for FID [Federation for
International Documentation]; associate editor, Information Processing
and Management; member of the editorial boards, Information Sys¬
tems, Library Research, and International Information and Com¬
munication
MYRON TRIBUS, Director, Center for Advanced Engineering Study,
M.I.T., Cambridge, Massachusetts
PETER WEGNER, Professor of Computer Science, Brown University,
Providence, Rhode Island
PATRICK WILSON, Professor, School of Library and Information Studies,
University of California, Berkeley
VLADIMIR ZWASS, Associate Professor of Computer Science, Fairleigh
Dickinson University, Teaneck, New Jersey; editor-in-chief, Journal of
Management Information Systems
 FOREWORD

A nation cannot live merely on victuals, comforts, games, and

weapons; a concern with ideas and values is essential, for without
them life becomes meaningless.

(Fritz Machlup, The Production and Distribution of Knowledge in

the United States, 1962)

Communication has become an American obsession. Those who admire it

justify the obsession as a necessary precondition for democratic govern¬
ment, or for a free economy; others view it less favorably as a defense
against the loneliness of personal independence, or against the insecurity of
competitive co-existence. Whatever the reason, the vast communication
industries that have sprung up in this century now bathe every willing citizen
in a continuous and unprecedented flow of written, spoken, musical, and
pictorial messages. The Earl of Chesterfield to the contrary, knowledge of
the world can now be acquired in a closet. Social and economic changes that
have accompanied this effusion of communication, first in the United States
and now worldwide, do clearly signal a New Industrial Revolution.
The Viennese intellectual and economist Fritz Machlup was among those
who first recognized and tried to characterize the economic consequences of
this revolution. Indeed, his work on the economic role of knowledge may
prove to be Machlup’s most important intellectual contribution. On his visit
to the United States in 1933—a visit that eventually became permanent—he
picked up, through reading and discussions at Harvard, an academic interest
in the theory of monopolistic or imperfect competition. Then one thing led to
another: among the institutions that restrict competition is the patent sys¬
tem; study of the patent system drew his attention to the economic impor¬
tance of research and development; quality research, in turn, depends on
quality education; education leads to the availability of journals and books,
which are but part of the vast communication system of the United States.
As his topic broadened, Machlup decided that what he came to call the
Knowledge Industry deserved comprehensive description and analysis.
In 1959 and 1960 the economic role of knowledge became the subject of a
series of five invited lectures, the first’at Cornell, the next four at Fordham.
Encouraged by their reception, Machlup expanded the lectures into a book

IX
X
FOREWORD

that was published in 1962 by the Princeton University Press: The Produc¬
tion and Distribution of Knowledge in the United States. The two longest
chapters were IV, entitled Education, and V, Research and Development.
The book was immediately recognized as a significant and pioneering
work, the first serious discussion of a basic change in American life. In the
concluding chapter Machlup stated the clear implication: “If employment
opportunities continue to improve for high-level-knowledge-producing labor
and to worsen for unskilled manual labor, the danger of increasing unem¬
ployment among the latter becomes more serious” (p. 397). Needless to say,
the trend has continued: his estimate that knowledge production and distri¬
bution accounted for 29 percent of the adjusted gross national product in
1958 was subsequently updated by Marc Uri Porat to 46 percent in 1967. And
the unemployment of unskilled labor has indeed become more serious.
In 1971, when he retired as Walker Professor of Economics and Interna¬
tional Finance at Princeton, Machlup decided that he had a moral obligation
to update his statistical analysis. In addition to his duties as professor at
New York University and a full agenda of other research, he undertook an
eightfold expansion of the 1962 book: the initial plan was for a series of eight
volumes, roughly one volume for each chapter of the earlier book, the whole
to be entitled Knowledge: Its Creation, Distribution, and Economic
Significance, with the Princeton University Press as publisher.
Volume I, Knowledge and Knowledge Production, was published in 1980.
According to the original plan, the second volume was to survey the
branches of learning and the information sciences, and to analyze the eco¬
nomic concept of human capital. As the work progressed, however, it grew
into three separate books, expanding the planned series from eight to ten
volumes. The first part, The Branches of Learning, appeared as Volume II in
1982. Volume III, The Economics of Information and Human Capital, was
completed only weeks before Machlup died of a heart attack on January 30,
1983, at 80 years of age.
Volume IV was to deal with the information sciences, the remaining part
of the originally planned second volume. In preparation, Machlup persuaded
39 information scientists to write a total of 56 essays on their various special¬
ties so that he, in his role as an editor, could go to school under the experts—
could “see the stir of the great Babel, and not feel the crowd.” The result is
the present book, prepared in collaboration with Una Mansfield. It is not
Volume IV, of course, but some indications of Machlup’s reactions to this
heroic exercise can be gathered from the co-authored Prologue and his own
Epilogue.
Probably nothing less than a Grand Design could have persuaded anyone,
even the enormously energetic Machlup, to organize such an enterprise as
this. The information sciences are highly diverse. Someone who speaks of
information in his own context may know of someone else who uses it in a
different context, yet never find the occasion or the impulse to explore the
similarities and differences. Now Machlup has made that exploration for us,
FOREWORD XI

and made it with a sensitivity to semantics and logic that few could match.
Discourse across these disciplinary boundaries will surely proceed more
smoothly in the future. Experts can join with The Celebrated Intelligent
Layman in their gratitude for the resulting improvement in mutual intelligi¬
bility. The topic is important, sufficiently important that, even if we cannot
speak in unison, we should speak as clearly as possible.
Kenneth Boulding once commented that Machlup had what he called the
anti-Midas touch—even if he were to touch gold, he would turn it to life. All
the contributors to this volume, all who endured his compulsive criticism
and enjoyed his appreciative praise as he tried to bring our essays to life, join
with Una Mansfield and me in dedicating our efforts here to the memory of
our charming friend and mentor, Fritz Machlup.

George A. Miller

Princeton, New Jersey

May 15, 1983
'
 PREFACE

Fritz Machlup’s pioneering work on The Production and Distribution of

Knowledge in the United States [1962] did not deal with the academic disci¬
plines that study information, for the simple reason that at the time Machlup
was researching and writing his book (mid- to late 1950s), few of them
existed. For instance, the concepts of cybernetics and information theory
had been introduced as recently as 1948 by Norbert Wiener and Claude
Shannon, respectively, and were still gaining acceptance among scholars.
The stored-program digital computer pioneered by John von Neumann in the
late 1940s had not yet given rise to the development of computer science or
its subfield artificial intelligence. The movement to develop general system
theory was spearheaded by Ludwig von Bertalanffy in 1954; and Herbert
Simon, a founding father, gives the year of birth of cognitive science as 1956.
The term information science was not used formally until 1958, and the
nature of the study area it denotes is still a matter of debate. Linguistics and
library science did exist, but they were not yet regarded as interrelated by
virtue of their common interest in studying information.
By 1980, however, Machlup wrote in his introduction to the first volume
of the series updating the 1962 work that “a comprehensive work on knowl¬
edge could not reasonably disregard the existence and rapid growth of sev¬
eral young disciplines that concern themselves with systems of information
and communication.” So he planned a volume (Volume IV) on The Disci¬
plines of Information that would ‘‘explore the interrelations among the
numerous disciplines, metadisciplines, interdisciplines, and subdisciplines
that deal with information as their central or peripheral concern.” I had the
privilege of collaborating with him on the research for that volume, which
included the commissioning of papers from experts in the various disciplines
so as to provide a basis for the analysis of interrelations. These papers are
presented here, and the Prologue and Epilogue contain their preliminary
analysis, which was to be expanded in Volume IV. The intentions of the
writers of these pieces will be better understood if I first offer a brief descrip¬
tion of the objectives and procedures adopted for the project. I quote from
the research proposal of Fritz Machlup in his application to the National
Science Foundation in 1979:

This project is designed to examine the interrelations among a number of

disciplines that are now regarded as constituent of, cognate to, or complemen¬
tary with information science. . . . The views expressed by representatives of

xiii
XIV PREFACE

different disciplines are inconsistent and obscure. . . . Specialists in the . . .

disciplines will be invited to prepare papers on the intra- and interdisciplinary
relationships as they see them. . . . The study is not intended as a mere survey
or synthesis of views held in the profession but rather as a serious . . . analysis
of the logical and methodological relations among past and present approaches.
. . . Whether a particular discipline is regarded as autonomous, superior, subor¬
dinate ... to another discipline is far less important than to illuminate the
genuine concerns of each of the fields and the ways in which various idealogical
strands are interwoven. . . . There is no specific method of analyzing logical
and methodological positions. As in the general philosophy of science, the
methodology of particular scientific disciplines calls for an analysis of apparent
consistencies and inconsistencies, of the pertinence and relevance of argu¬
ments, and of the linkages of pragmatic considerations with philosophical
premises ....

Nine lead papers and between three and five discussion papers for each of
the nine were commissioned. The writers of subsidiary papers were asked to
offer critical comments, elaborations, or supplementary observations on the
subject of the lead paper. The author of the lead paper would then come
back with a rejoinder. This procedure was designed to produce a discussion
as lively as a round-table debate and yet as well-thought-out as an argument
presented for publication in a learned journal or collective volume. Al¬
together, 56 contributions from 39 authors, plus an analytical Prologue and
Epilogue by the co-editors, resulted from this undertaking. All responses
and rejoinders were given titles descriptive of the messages delivered, and
the work was divided into nine sections dealing with cognitive science,
computer and information science, artificial intelligence, linguistics, library
and information sciences, cybernetics, information theory, mathematical
system theory, and general system theory. The format of lead paper/
comments/rejoinder worked well, except in Section 3 where the four papers
were written independently, with “endnotes” rather than a rejoinder by the
lead author, and in Section 4, which has no rejoinder.
The interdisciplinary character of the project is evident, whether authors
are listed by departmental affiliation or by professed specialty. As a matter
of fact, diversity of opinions, especially in matters of methodology, is not
confined to diversity of research commitment or of departmental affiliation;
disagreements are most conspicuous, and aired with greatest vehemence,
among those whose research interests are most alike. All in all, however,
more can be learned from the interplay of conflicting ideas, from the argu¬
ments and counterarguments on each issue, than from the best-formulated
but monolithic expositions of the fields.
In view of our interdisciplinary readership, a few editorial conventions
had to be imposed: for instance, technical “jargon” was discouraged and
mathematical notation reduced to the necessary minimum. First names were
used in the text to introduce scholars who though well known in some
disciplines might not be known in others. In citations and quotations, initials
PREFACE xv

of authors with common family names were included for easy reference. The
manuscript was virtually complete at the time of Professor Machlup’s death,
with the exception of his own Epilogue, the Cumulative List of References,
and the Index. The parts missing from the Epilogue have been indicated by
me in an Editor’s Note at the point in the text where his outline had placed
them; the parts that have been published are exactly as Machlup had written
them.
Apart from Fritz Machlup, the information community lost two other
outstanding scholars during the course of this research: Colin Cherry died in
November 1979, about the time Machlup was inviting him to act as an
adviser to the project; and Jesse H. Shera died in March 1982, just a few
months after he had submitted his paper in Section 5. Each has left us a rich
legacy in his writings.
A project of this magnitude involves the cooperation of many organiza¬
tions and individuals. Funding for the research was provided by grants from
the National Science Foundation to New York University, and from the
Spencer Foundation and the Earhart Foundation to Princeton University.
Our thanks to these foundations and institutions for their support. Howard
L. Resnikoff and his successor as director of the NSF Division of Informa¬
tion Science and Technology, Edward C. Weiss, provided encouragement
and guidance to Professor Machlup; and Herbert S. Bailey, Jr., director of
the Princeton University Press, and Robert F. Rich, editor of the journal
Knowledge, were trusted friends whose advice helped shape his plans.
Scholars who served as advisers to the project and nominated writers of
papers and comments included Russell Ackoff, Saul Gorn, Kenneth A.
Klivington, Donald E. Knuth, Lewis Levine, J. C. R. Licklider, Mitchell
Marcus, Mihajlo Mesarovic, George A. Miller, Alan J. Perlis, Claude E.
Shannon, Herbert A. Simon, David Slepian, Don R. Swanson, and Terry
Winograd. Four of them—Gorn, Mesarovic, Miller, and Perlis—contributed
papers of their own, as did the 35 other scholars in the List of Contributors.
We are grateful to all of them, especially the authors of lead papers who
were willing to have their work “critiqued” so intensively.
Professor Machlup benefited from discussing parts of our Prologue and
his Epilogue with several colleagues and family members, among them John
Woodland (Woody) Hastings, Richard C. Jeffrey, Richard N. Langlois,
Stefan Machlup, George A. Miller, Allen Newell, and Kenneth Stigleitz.
He would also want me to express his gratitude to Mary Taylor Huber and
Peter Swire, who provided valuable research assistance in the earliest phase
of the project, and to Rosa Schupbach for secretarial support throughout its
duration. Eliana Covacich and Robert Korchak at New York University
were especially helpful to him; and Michael McGill and Helene Ebenfield at
NSF showed understanding as program and project officials. As chairman of
the economics department, Stephen M. Goldfeld acted as “host” for Profes¬
sor Machlup’s Princeton-based research during his lifetime and bore respon¬
sibility for its successful completion after his death. Two good friends of
XVI PREFACE

Machlup’s, William J. Baumol and Donald W. King, also were generous

with their advice and support in bringing this part of his work to fruition.
The staff at John Wiley & Sons turned a difficult task into a pleasant one.
Special thanks are due to James T. Gaughan, the editor in charge of this
volume; to Carole Schwager and Christina Mikulak, who showed sensitivity
in their management of the copyediting; to Aline Walton for her designing
skill; and to Rose Ann Campise, whose cheerful but firm control kept the
publication on schedule.
I owe a personal word of thanks to Anthony Debons, whose provocative
teaching aroused my own curiosity about the relations among the “informa¬
tion disciplines” and led directly to my collaboration with Machlup. My debt
to Fritz Machlup covers much more than gains in scholarship; his friendship
and that of his wife Mitzi have greatly enriched my life.

Una Mansfield

Princeton University
October 1983
 ACKNOWLEDGMENTS

The authors and editors are indebted to the publishers for their permission to
reprint material from copyrighted works, as follows:

In Section 1

Sloan Foundation, Advisors of the Alfred P., “Cognitive Science 78,”

Report of the State of the Art Committee; passage reprinted by permis¬
sion of the authors and the Sloan Foundation.

In Section 4

Piatelli-Palmarini, Massimo, ed., Language and Learning: The Debate

between Jean Piaget and Noam Chomsky [1980]; passages reprinted
by permission of the Harvard University Press.
Piaget, Jean, Genetic Epistemology [Woodbridge Lectures, 1968],
translated by Eleanor Duckworth, copyright © 1970 Columbia Univer¬
sity Press; passage reprinted by permission of the Columbia University
Press.

In Section 6

Minsky, Marvin L., “Computer Science and the Representation of

Knowledge,” in Dertouzos, Michael L., and Moses, Joel, eds., The
Computer Age: A Twenty-Year View, copyright © 1980 Massachusetts
Institute of Technology; passage reprinted by permission of the MIT
Press.
Wiener, Norbert, Cybernetics, or Control and Communication in the
Animal and the Machine, copyright © 1948 and 1961 Massachusetts
Institute of Technology; passages reprinted by permission of the MIT
Press.
Hall, Thomas S., Ideas of Life and Matter: Studies in the History of
General Physiology 600BC-1900AD, Vol. 1: From Pre-Socratic Times
to the Enlightenment, copyright © 1969 by the University of Chicago;
passages reprinted by permission of the University of Chicago Press.
Waterman, Talbot H., “Systems Theory and Biology—Views of a
Biologist,” in Mesarovic, MihajloD., ed., Systems Theory and Biology
[1968]; passages reprinted by permission of Springer-Verlag, New
York.

XVII
XVIII
ACKNOWLEDGMENTS

Dunlap, Knight, The Elements of Scientific Psychology, St. Louis,

1922, The C. V. Mosby Co.; passage reprinted by permission of the
publisher.
Bruner, Jerome S., “Introduction” to MacDougall, William, Body and
Mind, copyright © by Jerome S. Bruner; passage reprinted by permis¬
sion of Beacon Press.

In Section 7
Tribus, Myron, “Thirty Years of Information Theory,” in Levine,
Raphael D., and Tribus, Myron, eds., The Maximum Entropy Formal¬
ism Conference, MIT, 1978, copyright © by the Massachusetts Insti¬
tute of Technology; paper reprinted by permission of the MIT Press.
 CONTENTS

PROLOGUE

Cultural Diversity in Studies of Information 3

Fritz Machlup and Una Mansfield

Addendum to the Prologue 57

SECTION 1

Information Science, Its Roots and Relations as Viewed From

the Perspective of Cognitive Science 63
Zenon W. Pylyshyn

Cognitive Science: The View from Brain Theory, Michael A. Arbib 81

Cognitive Neuroscience: More Plain Talk, Michael S. Gazzaniga 93

Cognitive Science: A New Patch in the Quilt? Saul Gorn 95

Reflections on the Structure of an Interdiscipline, Allen Newell 99

Informavores, George A. Miller 111

Representation, Computation, and Cognition, Zenon W. Pylyshyn 115

SECTION 2

Informatics (Computer and Information Science): Its Ideology,

Methodology, and Sociology 121
Saul Gorn

Perspectives on Informatics as a Semiotic Discipline,

Charls Pearson and Vladimir Slamecka 141

The Role of Information in Computer Science, Alan J. Perlis 149

The Province of Computer Science, Vladimir Zwass 151

xix
XX CONTENTS

Computer Science as the Science of Discrete Man-Made Systems,

Joel Moses 157

Paradigms of Information Engineering, Peter Wegner 163

A Pragmatist Replies, Saul Gorn 177

SECTION 3

Intellectual Issues in the History of Artificial Intelligence 187

Allen Newell

Methodological Links between Artificial Intelligence and Other

Disciplines, Margaret A. Boden 229

Artificial Intelligence: Cognition as Computation, Avron Barr 237

Artificial Intelligence: Subcognition as Computation,

Douglas R. Hofstadter 263

Endnotes to the Papers on Artificial Intelligence, Allen Newell 287

SECTION 4

Linguistics and its Relations to Other Disciplines 297

Thomas G. Bever

Thought, Language, and Communication, George A. Miller 319

Current Views of Language and Grammar, Ray C. Dougherty 321

Rules and Principles in Phonology and Syntax, Samuel Jay Keyser 327

Computation Does Matter to Linguistics, Robert C. Berwick 335

SECTION 5

Library and Information Sciences: Disciplinary Differentiation,

Competition, and Convergence 343
W. Boyd Rayward

Knowledge and Practice in Library and Information Services,

David Batty and Toni Carbo Bearman 365
CONTENTS xxi

Library Science and Information Science: Broad or Narrow?

Manfred Kochen 371

Librarianship and Information Science, Jesse H. Shera 379

Bibliographical R & D, Patrick Wilson 389

Librarianship and Information Research: Together or Apart?

W. Boyd Rayward 399

SECTION 6

Cybernetics 409
Murray Eden

Cybernetics: Past and Present, East and West, Peter Elias 441

Cybernetics and System Theory: A Search for Identity,

Richard Mattessich 445

Cybernetics in the Information Disciplines, Manfred Kochen 453

Cybernetics: The View from Brain Theory, Michael A. Arbib 459

Cybernetics: Closing the Loop, Murray Eden 467

SECTION 7

Thirty Years of Information Theory 475

Myron Tribus

The Wider Scope of Information Theory, Donald M. MacKay 485

Information Theory in Psychology, George A. Miller 493

Entropy and the Measure of Information, Peter Elias 497

The Entropy Function in Complex Systems, Elliott W. Montroll 503

Entropy, Probability, and Communication, Myron Tribus 513

 CONTENTS
XXII

SECTION 8

On System Theory and Its Relevance to Problems

in Information Science 517
Hassan Mortazavian

System Theory, Mathematics, and Quantification,

Kenneth E. Boulding 547

Can Mathematical Systems Be Concrete? Richard N. Langlois 551

System Theory and Information Science—Further Considerations,

Richard Mattessich 555

Problems of Systems Theory, C. West Churchman 561

Mathematical Systems Theory and Information Sciences,

Mihajlo D. Mesarovic 567

System Theory versus System Philosophy, Hassan Mortazavian 573

SECTION 9

System Theory, Knowledge, and the Social Sciences 581

Richard N. Langlois

Signals, Meaning, and Control in Social Systems, Walter Buckley 601

Toward a System-Based Unified Social Science, Robert D. Beam 607

Systems, Information, and Economics, Richard H. Day 619

Information Economics, Team Theory, and Agency Theory,

Richard Mattessich 625

On the Reception of Noise: A Rejoinder, Richard N. Langlois 631

EPILOGUE

Semantic Quirks in Studies of Information 641

Fritz Machlup

Cumulative List of References 673

Name Index 731

 THE STUDY OF

INFORMATION
PROLOGUE
 CULTURAL DIVERSITY IN
STUDIES OF INFORMATION
Fritz Machiup and Una Mansfield

The idea of two cultures in the academic world has been most forcefully
presented by C. P. Snow. According to him, a deep intellectual gulf divides
the mathematically minded laboratory-dwellers, engaged in the natural sci¬
ences, from the book-loving denizens of the library stacks, the literary intel¬
lectuals. [Snow, 1959.] The two cultures, sharing a mutual incomprehension
and disrespect of each other, are what now (in violation of semantic tradi¬
tion) are called science and the humanities. [Machiup, 1980, pp. 62-84.]
Snow later became aware of the existence of other fields. He conceded that
recognition of the social sciences would present us with “something like a
third culture,” but he denied the good sense of considering all possible
branches and twigs of learning that might give us “two thousand and two”
cultures. (Snow, 1964, pp. 66, 70.) We have chosen to deal here with only
thirty or forty cultures, a limitation explained by the fact that we are not
going to deal with the entire universe of learning but with only that part of it
that is characterized by the keyword information.

THE PROJECT

Our project is to analyze the logical (or methodological) and pragmatic rela¬
tions among the disciplines and subject areas that are centered on informa¬
tion. Disciplines (sciences, academic areas of research and teaching) are
orderly arrangements (metaphorically called bodies) of coherent thoughts,
formulated as propositions, about things (sense-objects or thought-objects)
deemed worthy of being known (i.e., being believed with some degree of
confidence) and being passed on. In other words, disciplines are what a
number of people, respected for having read widely and for being read by
other widely read people, have claimed to be disciplines. (We shall return
later to the question of how this pronouncement can be made operational.)
Relations among different disciplines are therefore relations among the ex¬
pressed thoughts of selected scholars (scientists). Logical relations among

3
4 FRITZ MACHLUP AND UNA MANSFIELD

propositions formulated by different people cannot easily be examined un¬

less the same words used by them denote the same meanings and equal
meanings are denoted by the same words. Whether this is actually the case
in any particular instance cannot simply be assumed but has to be estab¬
lished by scrutinizing the contexts of their messages. If such scrutiny reveals
the presence of misleading homonymity and/or expected synonymity, trans¬
lation becomes a prior task. Translation of discourses by different speakers
and writers is not easy and will succeed only if the translators have studied
the other fields sufficiently to have become “multilingual”—even though all
is being said in English. Few of those brought up in one of our thirty or forty
“cultures” have become bilingual or trilingual in our sense, and only very
exceptional scholars have succeeded in understanding the communications
emanating from all the disciplines involved.
It will now be understood why we have chosen the procedure of inviting
representatives of a variety of disciplines to set forth how they see rela¬
tions between their own specialty and various other disciplines, metadisci¬
plines, interdisciplines, or subdisciplines. Their specialty may be a con¬
stituent of a larger discipline; it may share principles in common with
cognate specialties or disciplines; it may be complementary with other
fields; but whatever interdisciplinary relations our authors see, they should
describe. To have such “interdisciplinary messages” discussed by others
who are working in the same field or a neighboring one, would allow us and
our readers to reach at least a fractional understanding of the logical and
pragmatic relations among the disciplines involved.

Interdisciplinary Explorations

When we attempted to explain our project of interdisciplinary explorations

to various curious people, we received some strange reactions. Many ot our
listeners could not believe that there were all that many disciplines con¬
cerned with information; when we offered a sample of the -ics and -logies,
. theories, and . . . sciences that had been unknown to them, their faces
expressed the mixture of awe and contempt that is characteristic ot those
suffering from an inferiority complex effectively suppressed by a strong
armor of intellectual superiority: “cybernetics? informatics? cognitive sci¬
ence? semiotics? robotics? artificial intelligence? Never heard ot it!
Others had “heard of it” and showed some appreciation for our efforts to
find out about the fields of inquiry that seem to form a class of academic
subjects dealing with the same object—or at least with something that has
the same name, information. One easy discovery of the explorer is the fact
that he or she finds here a prize example of homonymity: Information is not
just one thing. It means different things to those who expound its character¬
istics, properties, elements, techniques, functions, dimensions, and connec¬
tions. Evidently, there should be something that all the things called infor¬
mation have in common, but it surely is not easy to find out whether it is
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 5

much more than the name. If we have failed and are still at sea, it may be our
fault: Explorers do not always succeed in learning the language of the na¬
tives and their habits of thought.
One scholar who is at home in several of the different areas of study—
Herbert Simon, spiritus motor in many scientific endeavors—likened our
project to an anthropological exploration: We go into areas whose inhabi¬
tants speak foreign tongues (with many words sounding like words in our
own language but having very different meanings); we try to find some
guides to help us learn the meanings of these strange sounds; and we try to
make sense of what we see and hear, yet we probably misunderstand much
and are bewildered by even much more. Another scholar refused to compare
our work to anthropological research, chiefly because anthropologists often
have taken vows of intellectual chastity, so that they can embark on their
explorations untroubled by prior knowledge, immaculately free from pre¬
conceptions, and innocent of philosophical liaisons. What is really needed
for a successful inquiry is “an interdisciplinary epistemology . . . integrated
with philosophical understanding and with psychological and biological
knowledge.” We tend to agree with this judgment expressed by Margaret
Boden but should quickly admit that the statement, made in her paper in this
volume, was meant to refer chiefly to computational insights. We find it
equally valid for all interdisciplinary inquiries in the fields concerned with
information.
The intellectual requirement to break down the barriers between disci¬
plines has long been recognized. In virtually all branches of learning, we can
observe an ongoing conflict between specialization and integration, separat¬
ism and unificationism, isolationism and scholarly cosmopolitanism, or, in
brief, fission and fusion. The warning that we ought not to become
Fachidioten—the German pejorative word for excessively specialized ex¬
perts ignorant of things outside their narrow fields—has been sounded in all
areas of knowledge, but it is noteworthy that some of the founders of the
disciplines of information have been particularly concerned about the schol¬
arly seclusion of their fellow researchers.
Although a call for “mere intellectual communication across the bound¬
aries of these several disciplines is not enough”—to quote again from
Boden’s paper—listening to interdisciplinary messages is better than plug¬
ging one’s ears. International and interdisciplinary symposia were organized
soon after Norbert Wiener came out with his Cybernetics [1948] and Claude
Shannon with his paper on “A Mathematical Theory of Communication”
[1948]. The First International Symposium on Information Theory was held
in London in the summer of 1950. Among the participants were “mathema¬
ticians, physicists, engineers, linguists, physiologists, geneticists—in fact
folk from almost every branch of science.” (MacKay, 1969, p. 9.) All
branches of learning mentioned in this quotation are traditional disciplines,
represented in the conventional structure of every university. When similar
interdisciplinary symposia are held nowadays—32 years after that first
 FRITZ MACHLUP AND UNA MANSFIELD
6

London symposium—many of the participants regard themselves as repre¬

sentatives of new disciplines, specialties, or interdisciplines that did not
exist in the early 1950s or at least had not yet been recognized as academic
subjects of instruction, not even for courses offered by any of the existing
departments or programs.

A Cluster of Disciplines and Specialties

Although an enumeration of these new fields of inquiry and instruction may

be a rather unnecessary task for most readers of this volume, some readers
may be list-fanciers and may care to see the catalogue. In offering one, we
must warn, however, that some of the listed disciplines, hybrid disciplines,
or specialties, are challenged by some critics or opponents as illegitimate (in
one sense or another); that boundary disputes are unresolved; and that even
the major concerns of some of the subjects are controversial. The main
purpose of the listing is to impress the semi-informed reader with the lack of
order in a set of “bodies of ordered knowledge.’’ After all, it was the lack of
order, the jungle of coexisting, apparently interrelated subjects of study, that
prompted the undertaking of our inquiry.
We begin our list with designations that contain the letters i nj o r m:
information theory, information science, and informatics.1 We continue our
list with a similarly primitive scheme of classification: first the disciplines
with names ending in ics, then those ending in logy, followed by those called
science, analysis, theory, or research: bibliometrics, cybernetics, linguis¬
tics, phonetics, psycholinguistics, robotics, scientometrics, semantics,
semiotics, systemics; cognitive psychology, lexicology, neurophysiology,
psychobiology; brain science, cognitive science, cognitive neuroscience,
computer science, computing science, communication sciences, library sci¬
ence, management science, speech science, systems science; systems analy¬
sis; automata theory, communication theory, control theory, decision
theory, game theory, general system theory; artificial-intelligence research,
genetic-information research, living-systems research, operations research,
pattern-recognition research, telecommunications research. We may add
two fields, one no longer fashionable (largely because of its less flashy desig¬
nation) but once quite respected—documentation; and the other often over¬
looked (because of its association with classified intelligence-gathering)-
cryptography. Having grouped the disciplines or specialties in our list
according to the endings of their names, ics and logy (favored for scientific
subjects) or according to appellations like science, analysis, and so forth, we

1 We limited ourselves in the preceding to the three best-known designations but may add here
another six of lesser currency: informantics, informatistics, informatology, informology, infor-
metrics, and infometrics. We are indebted to Alvin M. Schrader, who not only compiled a long
list of designations but also knows the names of the terminologists responsible for coining them.
[Schrader, 1984.]
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 7

hasten to say that no hierarchical differences should be inferred. Indeed,

some who profess a field designated as a science are wont to renounce its
claim to this (apparently honorific) generic term, while others who hold a
general or empirical theory insist on its being really a science. Discussions of
such hierarchical distinctions are usually quite unproductive.
The more than thirty fields of inquiry and instruction included in the
preceding catalogue do not exhaust the class of studies of information and
knowledge. The field of communication in the sense of mass media has been
getting an increasing share of attention. The two keywords, information and
knowledge, go together in a large range of fields. This is true for most
disciplines dealing with cognition; moreover, almost all social sciences in¬
clude specialties devoted to the role of information and knowledge in soci¬
ety. Sociology of knowledge has long been a specialty in good standing, and
sociology of information has emerged more recently. Anthropology and poli¬
tics have begun to inquire into knowledge and information as increasingly
relevant subject matters. The economics of knowledge and information has
accumulated a bibliography of far over 20,000 published titles. Social psy¬
chology has its own literature on information and knowledge. And we should
not miss mentioning social phenomenology of knowledge. [Schutz and
Luckmann, 1973, especially chaps. 3, 4.]
This is a forbidding array of disciplines, metadisciplines, interdisciplines,
and specialties, far too big to be covered by the interdisciplinary messages
prepared for this volume. A few major disciplines of information had to be
selected for special emphasis, but the contributors were asked to examine
and discuss the methodological and pragmatic interrelations among as many
areas in the study of information as they cared to link to their major research
interests. They did, and thereby did much to show that disciplinary isola¬
tionism is an unsound attitude.

Fences around the Fields

Several analogies have been used to characterize isolationist or parochial

attitudes of specialists uninterested in cognate or complementary fields of
inquiry. For example, they erect fences around their fields—like unsociable
property owners inhospitable to their neighbors. The impediments dividing
the universe of discourse need not perpetuate isolation of the specialists;
fences can be taken down, or jumped over, and in any case, one may look
over the fence to observe what the neighbors are doing. A good many of the
originators and innovators of disciplines of information have urged their
fellow researchers to promote mutual understanding and collaboration.

Norbert Wiener deplored that there are

fields of scientific work . . . which have been explored from the different sides
of pure mathematics, statistics, electrical engineering, and neurophysiology, in
8 FRITZ MACHLUP AND UNA MANSFIELD

which every single notion receives a separate name from each group, and in
which important work has been triplicated or quadruplicated, while still other
important work is delayed by the unavailability in one field of results that may
have already become classical in the next field.
It is these boundary regions of science which offer the richest opportunities to
the qualified investigator. (Wiener, 1948, p. 2.)

In the same vein, Ludwig von Bertalanffy, the biologist and one of the
pioneers of general system theory, wrote that

science is split into innumerable disciplines continually generating new subdis¬

ciplines. In consequence, the physicist, the biologist, the psychologist, and the
social scientist are, so to speak, encapsulated in their private universes, and it
is difficult to get word from one cocoon to the other. (Bertalanffy, 1968, p. 30.)

And, to quote one more champion of mutual understanding among the

representatives of different fields, Kenneth Boulding lamented that

the Republic of Learning is breaking up into isolated subcultures with only

tenuous lines of communication between them ... an assemblage of walled-in
hermits, each mumbling to himself words in a private language that only he can
understand. (Boulding, 1956b, p. 198.)

However, in his plea for interdisciplinary collaboration, Boulding warned

that “it is all too easy for the interdisciplinary to degenerate into the undisci¬
plined.” (Ibid., p. 13.)
Some of these pleas go much farther than the plea that specialists learn to
know at least what their neighbors are doing. The interdisciplinary messages
in this volume are all from disciplines concerned with information, com¬
munication, cognition, or knowledge. The hope that representatives of dif¬
ferent fields of study may begin to learn each other’s vocabularies, and try to
comprehend what others are saying, should not be deemed overly ambi¬
tious. When the same word, information, is used in a dozen different mean¬
ings in different areas of inquiry, one should expect all users of that word to
be sufficiently curious to find out what “the others” mean by it. Moreover,
when users of the word have come to comprehend that the use of homonyms
may make their own communications unintelligible, or at least misleading, to
others, they should attempt to guard against misunderstandings, perhaps by
careful specification of their terms and by explicit efforts to exclude un¬
intended meanings.
We do not wish to give the impression that most specialists in the fields in
question favor isolation and oppose closer interdisciplinary relations. Some,
however, want to draw a line between research on information and research
on knowledge. Although they admit that mutual understanding and some
joint work may be helpful to specialists researching information-processing
and other operations concerned with information, they want to call a halt
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 9

when it is suggested that the cooperation be extended to the study of knowl¬

edge. This limited broadmindedness, which excludes the study of knowl¬
edge—although, as a rule, information is designed to effect or affect
knowledge—is not much less parochial and isolationist than the decision
of an information scientist to learn nothing but how to measure storage
capacity or compare the costs of alternative data-retrieval systems.
To be sure, the relative number of researchers who reject a wide lens for
their camera and insist on a sharp but narrow focus has been declining—
largely because people have become ashamed of their narrowness. Still,
many teachers in academic departments, divisions, and schools hold to the
view that the narrow focus is science and the wide lens is speculation.
Strong support for greater breadth has come from the small team of cogni¬
tive scientists, some of whom have progressed from the study of informa¬
tion-processing to the study of knowledge-structures. Indeed, at least one of
them has defined cognition as “the activity of knowing: the acquisition,
organization, and use of knowledge.” (Neisser, 1976, p. 1.) To impose a
fixed boundary line between the study of information and the study of
knowledge is an unreasonable restriction on the progress of both.

SPLITS AND MERGERS CHANGING THE MAP OF DISCIPLINES

We have enumerated almost forty fields in which information plays a

strategic role. Most of these fields made their first appearance on the
academic scene during the last three or four decades. Their emergence
is remarkable because of the extraordinary bunching of new arrivals
of manifestly cognate disciplines (metadisciplines, interdisciplines, and sub¬
disciplines). The evolution of new species of academic subjects is largely a
matter of splits or mergers of existing ones and a subsequent process of
selection on intellectual and quasi-intellectual (political) grounds. The no¬
tions of splits, mergers, and selection call for elaboration.

Fission and Fusion in Academe

Splitting, or fission, of academic disciplines has been observed for over two
thousand years. Chemistry, for example, was an early spin-off from physics,
both being exhibited as separate sciences in philosophical discussions and
classifications of natural philosophy. After various subdivisions had oc¬
curred, a merger of parts of physics and chemistry proved expedient: It
resulted in physical chemistry. Several splits and mergers—fissions and
fusions—have occurred in the life sciences, with biology being divided into
several subdisciplines and merged with parts of physics and chemistry. This
process led to biophysics and biochemistry, with the latter splitting again
into chemical biochemistry and biological biochemistry. [Machlup, 1982, pp.
101, 152.] Similar sequences of splits and mergers have occurred in moral
10 FRITZ MACHLUP AND UNA MANSFIELD

philosophy. Fission is already indicated in Aristotle’s work and becomes

quite explicit in the writings of the Christian church fathers. In the depart¬
mental organization of modern universities, the separations were made in
the nineteenth and twentieth centuries. The mergers were usually made
through the creation of interdepartmental programs, but, in many instances,
political forces have led to the creation of interdisciplinary departments.
Thus, moral philosophy, or social science, was first divided, in several con¬
secutive steps taken during a period of fifty to a hundred years, into cultural
anthropology, demography, economics, human geography, political science,
social psychology, and sociology. Then all these and several more disci¬
plines were again brought together for area studies, ethnic studies, urban
studies, women’s studies, and several more problem areas. For example, all
social sciences had to be called on for a program or department of Latin
American studies, sometimes supplemented by history (cultural, political,
and social), languages (Spanish and Portuguese), and law.
As a rule, the reasons for splits and mergers are associated with such
philosophical and methodological issues as those encapsuled in pairs of
opposites like analysis versus synthesis, abstract versus concrete, and basic
versus applied. (Note that we did not include deduction versus induction,
those Pandora boxes of confusion.) Analysis, increasing abstraction, and
emphasis on basic research dictate splits of disciplines; whereas synthesis,
attention to the concrete, and interest in application to real-life problems
dictate collaboration and mergers of disciplines. However, to explain the
historical facts of fission and fusion of academic disciplines on grounds of
sound methodological principles does not mean that actual splits and merg¬
ers were always motivated entirely by sound principles; personality clashes,
campus politics, institutional funding, and similar less-principled influences
have often caused or triggered decisions by faculties and administrations of
our universities.

Creation or Emergence of New Fields

These decisions do not, however, create new disciplines—either by splitting

existing ones or joining parts of existing ones into new amalgams—but
merely ratify a development that has already sufficiently progressed in
learned journals, colloquia and symposia, and the invisible college of oral,
written, and prepublished communications. This is the place for us to come
back to the question of whether a discipline becomes a discipline just be¬
cause some people say that it is a discipline. We qualified this statement by
an essential proviso: that people who say, “Let there be a new science of
Aplopathoscopics!”2 are respected scholars who have read widely and are

2 Aplopathoscopic neumatology was one of the subdivisions of psychology proposed by Jeremy

Bentham. [Bentham, 1816; Machlup, 1982, p. 63.]
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 11

being read widely by respected scholars who have read widely and are being
read widely by respected . . . and so forth—though the circle of scholars
need not be large. There are operational tests for the existence of such a
consensus among recognized scholars. The tests include such things as ref¬
erences to certain keywords characteristic of the new discipline (or spe¬
cialty) in subject indexes of recent textbooks and in titles of articles and
books; the appearance of survey articles summarizing the contents of publi¬
cations in the new field; the extension of the classification scheme of a larger
discipline in order to provide one or more special headings for listing publi¬
cations in the new discipline (or specialty); the appearance of books of
readings and volumes of conference proceedings focusing on the new disci¬
pline (or specialty); and, finally, courses offering instruction in the new
subject in universities, first to graduate students, later, perhaps, also to
undergraduates.3
A novel approach to the identification of specialties that are presumably
“ready to become recognized as having arrived” relies on citation analysis.
The analyst finds clusters of articles and books linked either by cocitation or
by bibliographic coupling. In cocitation, different earlier publications are
linked by being cited together in the same new pieces; in bibliographic
coupling, different new publications are linked as they cite the same earlier
pieces. [Garfield, Malin, and Small, 1978, p. 185.] A similar approach can
establish mutual-citation societies by compiling a citation matrix: In a new
or emerging field, whom did the writers cite and by whom were they cited?
If, however, a small group of people cite one another but cite hardly anyone
else, must one not suspect the existence of a club, a cult, or a conspiracy?
Perhaps so; yet one should not on this ground deny the group recognition as
a bona fide field of study, especially not if some of its members are scholars
also respected outside the group.

Reapportionments and Multiple Assignments

That a new discipline is put together entirely by merging selected portions

from two or more existing disciplines is no reason for denying it the designa¬
tion of a science. Portions from several fields may be required for the study
of particular phenomena or particular problems. If such multidisciplinary
efforts are needed not only temporarily but for prolonged periods of time, an
interdiscipline will be established, without any presumption regarding a
higher or lower scientific level. If the new interdiscipline is largely con¬
cerned with special or concrete problems, its orientation is likely to be more
toward applied research.
Some scientists have been taught (or rather mistaught) to restrict the

3These tests have actually been carried out to ascertain that the economics of knowledge and
information has become an established specialty within the field of economics. [Machlup, 1984,
chap. 9.] The number of titles published in this specialty exceeds 20,000.
12 FRITZ MACHLUP AND UNA MANSFIELD

appellation science to basic research and deny it to applied research. Such

snobbism or bigotry is not compatible with the fact that currently at least 95
per cent of all scientific effort is mission-oriented, that is, designed to pro¬
duce practically useful knowledge. Let us submit, first, that the merger of
parts of several sciences creates a compound discipline no less scientific
than the donor disciplines; second, that the parts taken over from existing
disciplines and incorporated into new ones are not thereby removed from
their original or conventional discipline but belong to both, the old and the
new; and thirdly, that any new findings altering the structure of our knowl¬
edge or adding to it may enrich both the old and the new fields.
In the title of this section, we have suggested that splits and mergers are
changing the map of academic disciplines. This suggestion may be mislead¬
ing in some respects. It may induce us to think of a process like redistrict¬
ing—changing the boundaries of election districts in a state—and this would
be fallacious. For what is added to one district is taken away from another—
something we have just denied with regard to mapping academic disciplines
or areas of research.
Fortunately, the philosophers of science who have discussed the bound¬
aries of various disciplines have not drawn maps showing how certain phe¬
nomena, problems, concepts, laws, or theories should be appropriately as¬
signed to various disciplines without undue duplication. Any degree of
overlapping is acceptable in reapportioning the universe of learning. Multi¬
ple assignments of the same tasks (for research or instruction) may be highly
productive. And the fact that a discipline imports most of its materials and/or
methods from other disciplines does not reduce its significance and should
not reduce its scientific respectability.
We have refrained in this section from drawing on Allen Newell’s instruc¬
tive examples of and observations on the emergence of new interdisciplines.
Instead of quoting and paraphrasing, we want to call Newell's paper to the
reader’s special attention. What Newell tells us in his comments on
Pylyshyn’s lead paper and in his own lead paper on artificial intelligence is
most illuminating on the issues we have discussed here.4

Science or Nonscience, Is That the Question?

We have discouraged some of our contributors from using much space for
discussing whether a particular discipline is true science or something else.
Several of them decided, even without our advice, not to expatiate on this
sterile topic. We did not, however, wield a red pencil to delete all comments
on whether something was a science, or “only” philosophy, speculation,
technology, practical art, professional training, and so forth. We knew that

4Among the most intriguing is Newell’s story of neuroscience, “also an interdiscipline—the

neuro parts of anatomy, physiology, pharmacology, and biochemistry—just working itself up to
disciplinary status,” but participating “in the central paradigm of modern biology.” (See
Newell’s paper in Section 1 of this volume.)
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 13

many writers in information science had guilt feelings about the fact that this
discipline had neither discovered new laws nor invented new theories and
therefore did not deserve recognition as a science. Such an inferiority com¬
plex is the result of indoctrination with an outmoded philosophy of science,
with persuasive (propagandist) definitions of science and scientific method.5
The restrictive meaning of science, taught in courses and texts in experi¬
mental natural sciences and designed to exclude other academic disciplines
from the class of sciences, occurs only in English. Neither French nor Ger¬
man—nor any other language, to our knowledge—has words to express the
narrow concept of science or any expressions equivalent to scientific
method. [Machlup, 1980, pp. 67-69.] If the restriction of the term to eligible
disciplines could serve any scientific purpose, the French, Germans, Rus¬
sians, Japanese, and others would surely have coined words to express what
their English-speaking colleagues had in mind. Still, some contributors to
this volume do allude to conflicts between science and philosophy, science
and unverified belief, science and art, science and technology, science and
profession. These issues are therefore briefly addressed in the Epilogue by
Fritz Machlup.
For present purposes, we may confine ourselves to the statement that we
do not care, and no one else need care, whether information science, library
science, computer science, decision science, system science, or any of the
disciplines discussed in this volume, are genuine sciences. Fittle depends on
the decision to deny this honorific designation to a particular discipline.6 If
little or nothing follows from the distinction between science and nonsci¬
ence, it is not worth investing much time making or defending it. Too bad
that even a little time has to be spent showing that the distinction is mis¬
chievous.

SOME TRANSDISCI PLINARY ISSUES

The titles of the papers contributed to this volume contain the names of
twenty disciplines or subdisciplines. The texts of the papers contain refer-

5“. . . the attempt to define scientific method or to analyze science is a search for a persuasive
definition . . . the term ‘science’ has no definite and unambiguous application. (Black, 1949, p.
69.) “Neither observation, nor generalization, nor the hypothetic-deductive use of assump¬
tions, nor measurement, nor the use of instruments, nor mathematical construction nor all of
them together—can be regarded as essential to science.’’ (Ibid., pp. 80-81.)
6If someone points to the differences in salaries of librarians, physicists, and computer special¬
ists, the explanation lies in supply and demand, not in discriminatory designations of their
fields. However, it may be possible that agencies and foundations supporting research are
persuaded to favor grant applications for supposedly more scientific projects: The charge of
being ‘unscientific’ is not mere namecalling; it is a charge with financial consequences. It has to
do with what research gets funded and therefore with what research gets carried out. (Lakoff,
1978, pp. 267-268.) We suspect, however, that it is not the magic of the word scientific that
charms the grantors but rather the lure of research leading to practical findings.
14 FRITZ MACHLUP AND UNA MANSFIELD

ences to or discussions of innumerable disciplines or subdisciplines; in forty

of these, the term information, in one of its several meanings, plays a key
role. We shall not undertake a survey of the whole lot, but we feel an
intellectual obligation to select a few of the fields for commentary in this
introductory chapter, because representatives of these fields have expressed
conflicting views on their methodological nature, their scope, and pragmatic
significance. Our interpretations and suggestions may differ—in some in¬
stances quite drastically—from some of the pronouncements of our con¬
tributors, notwithstanding the fact that it is thanks to their statements that
we have reached our present positions. Indeed, before we studied their
papers, we had seen some of the interconnections and relations among the
fields in question quite differently from the way we do now. This is espe¬
cially true with regard to general system theory and should not surprise
readers of the papers dealing with this field: They will find that intradisci-
plinary conflict and controversy are rampant. Intradisciplinary dissension
becomes particularly pronounced when representatives of a subject try to
clarify its relations to other disciplines.
Before offering our commentaries on selected areas of research and in¬
struction, we shall say a few words on transdisciplinary and interdisciplinary
issues.

The Balance between Research and Instruction or Training

Some of the conflicts about the scientific character of a field commonly

called science are due to prejudice with respect to a scientist’s proper activi¬
ties. The high priests of the scientific establishment have created a pecking
order giving highest rank to the successful researcher, second highest to the
plodding researcher (with some publications to his credit), third rank to the
teacher (without meritorious publications), and lowest rank to the profes¬
sional who only practices some of the things (know-how or skill) included in
the curriculum of the school, department, or academic program that dissemi¬
nates the knowledge embodied in the discipline. A fifth category is usually
disregarded: those who have studied and mastered the discipline but are not
active as researchers, teachers, or practitioners.
Like many or most hierarchical distinctions, ranking the level of science
according to the activities in which those trained in the field are engaged is
largely a matter of snobbism and arrogance. Should the question of how
much of a science a particular discipline is be answered by calculating the
proportions of basic researchers, applied researchers, teachers, consultants,
and practitioners who have acquired advanced degrees in the discipline? Is
biophysics really more of a science than biochemistry just because the pro¬
portion of degree holders employed by universities is greater? Does molecu¬
lar genetics become less of a science when many of its professors move into
industry? Is computer science or one of its specialties, such as artificial-
intelligence research, downgraded because increasing numbers in the field or
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 15

specialty are yielding to the pecuniary inducements of nonacademic employ¬

ment? For many decades, the majority of degree holders in chemistry and
geology have been employed in other than academic positions, but have they
for that reason regarded themselves as nonscientists? All these questions are
rhetorical, inviting, in our opinion, negative replies.7 If we are right and the
status of a science is not determined by what those who have studied it are
doing for a living, then some of the self-doubts of professors of practical
sciences could be removed. Professors in schools of information science,
computer science, and/or library science need not forego the self-esteem
usually cultivated by “true scientists.”

Formal and Empirical Sciences

A more detailed discussion of the distinction between formal (or rational)

and empirical sciences will be deferred to the last chapter, but a few words
are needed here to introduce our characterization of some of the disciplines
discussed in this volume. That logic and pure mathematics are the proto¬
types of formal science, convey truth by resolution (a priori) and therefore
cannot be tested and disproved by experience (a posteriori), and can be
applied to propositions about the real world in attempts to test these proposi¬
tions for consistency and formal validity—all these are verities with which
most educated people are familiar. Less well known is the fact that new
developments in scientific reasoning can extend the scope of logic as well as
mathematics and that some novel disciplines, or some of their parts, belong
to the domain of formal science. When purely formal statements (axioms,
heuristic postulates, analytical conventions, paradigms, rigid laws, and

7 The issue of practical versus pure sciences is perhaps most easily clarified by contrasting
medical science with medical practice. Clearly, medical research, teaching medicine, a.n& prac¬
ticing medicine are three different activities. No matter whether students of medicine plan to
become researchers, professors, or practitioners, they will have to be educated in several
medical and cognate sciences—anatomy, physiology, chemistry, biochemistry, pharmacology,
cardiology, ophthalmology, otolaryngology, gynecology, psychiatry, and several others. The
majority of medical students plan to take a degree of doctor of medicine and obtain a license to
practice medicine. In their capacity as practitioners, doctors are not scientists, though they
have mastered a good many sciences. Those engaged in medical research and/or in teaching
medicine may spend part of their time as practicing physicians; their scientific work consists of
research, publication, and teaching; but many medical doctors double as scientists and prac¬
titioners. None of the sciences taught in medical schools is less of a science due to the fact that
the ratio of those who have studied it and are “only” practitioners is much larger than in other
fields. An enthusiastic teacher-scholar may take pride in the fact that a larger percentage of
students choose to engage in research rather than in medical practice, but this does not reflect
on the discipline professed. Likewise, a medical school that produces relatively large numbers
of medical researchers will have a much better reputation than one that produces only prac¬
titioners. But this has nothing to do with the scientific character of disciplines taught in medical
schools. It is possible, of course, that the way the discipline is presented to students is affected
by the knowledge of the vocational-career orientation of most students, and, thus, the scientific
level of the presentation might be lowered.
16 FRITZ MACHLUP AND UNA MANSFIELD

scientific research programs) are applied to empirical propositions (based on

observations or generalizations), expositors are often misled in their discern¬
ment of the analytical or synthetic character, not only of propositions and
general theories, but also of entire specialities and disciplines. Whether
these areas of inquiries should be assigned to the category of formal sciences
or to the category of empirical sciences is sometimes hard to decide. It
happens that such doubts prevail regarding the nature of several of the
disciplines dealing with information.
Information science is not involved in any such methodological con¬
troversy, chiefly because no agreement exists about its object or objects. By
and large, information scientists deal with practical matters and, therefore,
with the world of experience. Library science is clearly empirical in all its
aspects: bibliographical work, cataloging, indexing, reference services,
management, organization, acquisition, circulation, and all the rest; every
phase of research in this field is practical-empirical. The case of computer
science is not quite so obvious, because there is some formalism behind a
few of the research agenda and some innovational projects; still, on closer
examination, sound judgment will assign computer science to the category
of empirical disciplines. This holds also for its semiautonomous subdiscip¬
line, artificial intelligence.
Cognitive science, a recently developed metadiscipline comprising almost
a dozen interdisciplines, gives us difficulties, because not all of these subdo¬
mains are equally easy to characterize. Among the traditional disciplines
that are partly engaged in interdisciplinary bonding is philosophy; it is joined
with psychology on the one hand and with linguistics on the other. Should
the philosophy of psychology and the philosophy of language be regarded as
empirical if only one partner in the joint venture is in the epistemic domain,
in the domain of experience and observation? We are inclined to answer
affirmatively. Nonempirical reasoning, although not subject to testing and
falsification, does not transport the synthetic propositions about observed or
experienced phenomena and processes into the domain of pure construction
or evaluation. The empirical strain in the pair dominates the nonempirical,
which, thus, is the recessive one.
One other subdomain of cognitive science open to question about its
methodological status is the interdiscipline formed by the cross-connections
between computer science and neuroscience, both of which are clearly em¬
pirical fields. Why, then, should any question arise regarding the empirical
nature of the joint research efforts of two empirical disciplines? It is the
designation given to the interdiscipline that can make us suspicious: The
name cybernetics—the theory of communication and control—can raise
doubts (which will become clear presently when we discuss cybernetics).
Our tentative conclusion, at this juncture, is that the name does not fit the
interdiscipline in question. Joint research in neuroscience and computer
science cannot be anything but empirical. Regarding other subdomains of
cognitive science, no doubt about their empirical nature can be entertained.
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 17

Semiotics and linguistics, the science of signs and the science of those
signs that make up language, respectively, deal with observable objects.
Although a good deal of the reasoning in these disciplines is formal and,
thus, in the domain of mental construction, the subject matter is no doubt
empirical.
We now come to disciplines of a rather different character, and we want
to state that we have approached them with a minimum of preconceptions
regarding their methodological status. After careful study of the papers in
this volume and of some of the earlier literature on cybernetics and general
system theory (including mathematical system theory), we have come to the
following conclusions about the relation between these fields and their
places in the universe of learning.
Cybernetics is a method of examining interactive parts of any system of
communication and control, with particular emphasis on feedback and
homeostasis. It is part and parcel of general system theory, which is con¬
cerned with these and also other relations among parts of a whole and
between the whole and its environment. General system theory, which thus
comprises cybernetics, is a method of organizing one’s thinking in terms of
interrelated elements in closed or open sets. It is part of the general
methodology of nomological disciplines. General methodology, which thus
comprises general system theory, is “the second part of logic” (sometimes
called material logic in contrast to formal logic). (Windelband, 1913, p. 22.)
To summarize, cybernetics is a part of general system theory, which is a
part of methodology, which is a part of logic. All of these may be given
the designation of disciplines or sciences, though not empirical sciences.
Whenever formal sciences are applied to the interpretation of phenomena,
or of relations among phenomena, the resulting propositions are parts of the
discipline or disciplines that are concerned with the phenomena or relations
in question. To offer examples, mathematics and methodology applied to
molecular biology yield molecular biology; system theory applied to macro¬
economics yields macroeconomics; and cybernetics applied to neuro¬
physiology yields neurophysiology.
More will be said later, both about cybernetics and general system theory
in special sections of this introductory essay. Our next question for the
present discussion of formal versus empirical science concerns the status of
information theory in the narrow form of the mathematical theory of com¬
munication. With its progenitor, Claude Shannon, and his precursors, Harry
Nyquist and Ralph Hartley, all of them researchers at Bell Laboratories
doing mission-oriented research on electrical circuitry and signal transmis¬
sion, one would hardly have any serious doubts about this kind of work
being empirical. Yet, we shall argue that the strategic propositions of Shan¬
non’s theory are purely formal uses of general probability theory applied to
assumed states of assumed objects; they are analytical relations derived
from definitions and other resolutions.
This statement may sound strange to ears not attuned to methodological
18 FRITZ MACHLUP AND UNA MANSFIELD

dissonances. Can one reasonably conclude that purely formal propositions,

a purely analytical framework, may have a proper place in a practical-
empirical discipline? The answer is affirmative; yes, indeed. However, the
use of formal systems and abstract models in a discipline is not conclusive
evidence that the discipline is formal or empirical. We reserve judgment on
the methodological status of information theory, especially since informa¬
tion theory comes in numerous dresses and guises, manifestations and incar¬
nations.

PROFILES OF SOME SELECTED DISCIPLINES

If we were to attend to the traditional tasks of introducing a collective

volume, we should have to say a few words about each of the papers. We
prefer to think that the papers need no introduction. Perhaps we should have
offered introductory remarks to each of the nine sections in which the
volume is organized; we have decided against this scheme, because the
interdisciplinary character of many of our authors’ messages makes it more
desirable to bring the editors’ comments “up front.’’ We shall offer brief
sketches, or profiles, of information science, computer science, artificial
intelligence, cognitive science, semiotics and linguistics, cybernetics, sys¬
tem theory, and information theory. The sketches are not intended to sum¬
marize the messages conveyed in the papers, still less to criticize or take
sides. They may, we hope, be of service to readers who are quite unfamiliar
with the subject matter.

Information Science

We have detected at least four main uses of the term information science in
the literature: (1) In its broadest sense, it stands for the systematic study of
information and may include all or any combination of the academic disci¬
plines discussed in this volume; (2) when included in the phrase computer
and information science, information science denotes the study of the phe¬
nomena of interest to those who deal with computers as processors of infor¬
mation; (3) in library and information science, it indicates a concern with the
application of new tasks and new technology to the traditional practices of
librarianship; and (4) in its narrow sense, information science is used as the
name for a new area of study that is evolving from the intersection of the
other three mentioned areas, with perhaps a special interest in improved
communication of scientific and technological information and in the appli¬
cation of well-tested research methods to the study of information systems
and services.
The earliest formal use of the term information science in the United
States seems to have been in the description of a program in computer and
information science at the Moore School of Electrical Engineering, Univer-
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 19

sity of Pennsylvania, in 1959. [Wellisch, 1972, p. 164.] However, an Institute

of Information Scientists had been formed in the United Kingdom in May
1958, although the use of the term information scientist may have been
intended to differentiate information scientists from laboratory scientists,
since the main concern of members was with the management of scientific
and technological information. [Farradane, 1970, p. 143.] The level of sub¬
ject expertise needed to handle this information dictated that it be done by
scientists fully qualified in the disciplines concerned, rather than by librari¬
ans with a knowledge of only general science terminology.

Cultural Diversity and the Melting Pot

The early work in the broad area of information science is well characterized
by the subheading chosen for these paragraphs. Scientists and scholars from
a variety of disciplines were attracted to the new area of research and teach¬
ing; they brought with them their own research methods and terminology
resulting in a Babel of sorts. Librarians found it difficult to interact with
mathematicians; cognitive psychologists had problems in dealing with physi¬
cists; and social epistemologists considered engineers to be a breed apart.
This variety of interests is reflected even today in the membership of the
only society in the United States that seeks to cater to all information profes¬
sionals—The American Society for Information Science. [King, Krauser,
and Sague, 1980.]
The great melting pot of information science included, among other
things, electrical engineering (e.g., research on signal transmission over noisy
channels), computer technology (e.g., research on information-processing
by machines), biological sciences (e.g., research on information-processing
in living systems), behavioral sciences (e.g., research on cognitive pro¬
cesses), and social sciences (e.g., research on the sociology of knowledge).
Not that there were many who engaged in all of these areas of inquiry, but
there were some who concentrated on one or two and did so under the
banner of information science. However, assembling under a banner is not
the same thing as eliminating cultural diversities in the processes ascribed to
the melting pot: The diversities remain and no unified science emerges.

The Power of the Plural S

We submit that most of the confusion caused by the use of the term informa¬
tion science in its broadest sense could be avoided by the addition of the
plural 5. The information sciences could then take their place alongside the
natural sciences, the social sciences, and other umbrella terms that indicate
a grouping of disciplines and fields of study that share a common characteris¬
tic. The bond among the information sciences is, of course, their focus on
information as the object of study, though it is important to bear in mind that
the word information is interpreted very differently by various groups of
20 FRITZ MACHLUP AND UNA MANSFIELD

researchers. [For a discussion of “What They Mean by Information,” see

Machlup’s Epilogue.]
The assemblage of different fields under the collective designation would
have solved a good many problems or pseudoproblems. Like the natural
sciences and the social sciences, the information sciences need no single
paradigm, no overarching scientific research program, no common funda¬
mental postulates and axioms, no unified conceptual framework.8

Computer and Information Science

The omission of the plural 5 in the combination of computer and information

science has also proved misleading. Was it intended to bring two sciences
into the same organizational unit or to forge one science out of two? In the
latter case, hyphens would have been needed to show that the word com¬
puter shared with information the function of a modifier of science: com-
puter-and-information science. On the other hand, if no complete union, but
only cohabitation, was intended, the continued coexistence of the two sci¬
ences should have been recognized by a plural 5. This had evidently been in
the thoughts of earlier concerned scientists; thus, the Curriculum Committee
on Computer Science of the Association for Computing Machinery told us in
their report in 1968 that although they had decided to use the term computer
science, others, “wishing perhaps to take in a broader scope and to empha¬
size the information being processed, advocate calling this discipline ‘infor¬
mation science’ or, as a compromise, ‘the computer and information sci¬
ences’.” (ACM Curriculum Committee, 1968, p. 153.) Note that the plural 5
here is intended to indicate two separate sciences—computer science and
information science—and not the information sciences in the sense of an
umbrella term. Incidentally, although the use of the compound computer
and information science (with or without the plural s) is more usual, some
authors do use the terms computer science and information science inter¬
changeably. [See, for example, Pylyshyn, 1970, p. 61, and Wegner in this
volume.]
There was, of course, a good reason for getting studies of the computer
and information under one roof. The interest of computer scientists in infor¬
mation tends to be confined to its role in computer systems and to involve
signs, symbols, and so forth (the semiotic approach), and their processors
(the informatic approach). [See papers by Gorn, and Pearson and Slamecka
in this volume.]

Library and Information Science

What has been said about missing hyphens and the plural t holds also for the
combination between the study of the library and the study of information.

8Most schools and even some universities offer courses called science and social science,
evidently presenting surveys of several disciplines under each title. We consider the omission of
the plural s in these instances as a benighted practice.
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 21

One needs to decide whether two separate but cognate subjects ought to be
housed under one roof or whether they should be fused into one science with
the research effort fully integrated. Most writers and teachers have not made
up their minds, and by omitting both hyphens and the plural y, they put an
unstructured string of words on their letterheads.
The type of information science that grew out of the documentation
movement of the 1960s, and is taught mainly in library schools, has a focus
different from that of the computer scientists. The emphasis is on better
techniques for managing recorded information in whatever medium it re¬
sides. There is concern with how the records will be used and, therefore,
with their intellectual content. Much of the information-science activity in
this field centers on the application of new technology to traditional library
functions of acquiring, storing, retrieving, displaying, and disseminating
records—in other words, on technically improved librarianship. Thus far,
there is little evidence that the new technology has been exploited to pro¬
duce radically new ways of handling the store of knowledge recorded in
library and other materials. But there is no doubt that new tasks have been
assigned to librarians, for example, the management of information centers
concerned with nonbibliographic information, participation in resource¬
sharing networks made necessary by the increasing volume of recorded
knowledge and dwindling library budgets, retrieval from electronic online
databases made possible by the new computer and communications technol¬
ogy, and so on.
The eagerness with which library schools have moved to incorporate the
word information into their titles is proof that their deans and faculties view
the new technological developments in information handling as vital for their
growth or even their survival. A recent article pointed out that “ ‘Informa¬
tion’ is now contained in the names of fully 37 of the 70 schools appearing in
the most recently revised list of programs accredited by the American Li¬
brary Association, although only one (Syracuse’s School of Information
Studies) has forsaken the word ‘library’ altogether.” (Harter, 1982, p. 40.)
But although library schools are quick to enlarge their curricula to include
aspects of information studies, there is a question whether this is always a
substantive change in disciplinary orientation. Often courses in information
science consist merely of teaching students to use a new tool, the com¬
puter—something that will eventually be done at earlier stages in the school¬
ing process, at the secondary, if not elementary, level. (Learning to use a
computer will soon be regarded as a basic skill.) On a somewhat higher level
is teaching data retrieval from online electronic databases, which is now
taught and may well remain in the curricula of library schools. Still, this sort
of training cannot reasonably be regarded as information science. To qualify
for that designation, the respective studies should involve designing, build¬
ing, and programming the online retrieval systems.
The discipline of library science, or librarianship, onto which this new
type of information science is being grafted has a long and proud tradition of
service. Its origins and development, and the disciplinary dilemma it faces in
22 FRITZ MACHLUP AND UNA MANSFIELD

this era of information technology, are eloquently described by Boyd Ray-

ward and the commentators in Section 5 of this volume and need not be
profiled separately here. [See also Buckland, 1983.]

Narrowly Focused Information Science

We have characterized information science in the broadest sense as a rather

shapeless assemblage of chunks picked from a variety of disciplines that
happen to talk about information in one of its many meanings. We have then
looked at information science oriented toward the needs or interests of
computer scientists. The third kind of information science was that which
would have an intimate liaison with library science. We shall now ask our¬
selves whether there can be an independent information science with a
narrower focus—the problem of information linked neither to computer
science nor to library science and also avoiding the vagueness associated
with information science in its broadest sense.
Perhaps such a discipline will evolve, but we doubt that it now exists.
Among research projects reported in this narrow field of information science
are studies of patterns of communication among scientists and scholars
(e.g., cocitation analysis); studies of improved methods of classifying infor¬
mation (e.g., computer-based cataloging of documents); statistical studies of
the growth and distribution of the literature (e.g., the area known as bib-
liometrics); novel methods of information exchange (e.g., electronic infor¬
mation networks, teleconferencing, etc.); control of access to information
(e.g., governmental regulation of information transfer, international com¬
munications conventions, etc.); modeling and computer simulation of infor¬
mation systems and networks; studies of the characteristics and behavior of
users of information systems and services; studies of human factors in¬
volved in the design of man/machine systems; and so on.
It is difficult to say whether these types of research constitute an indepen¬
dent discipline or are merely developments that belong in either computer
science or library science. There is no doubt, however, that information
service and knowledge analysis are growing branches in most developing
economies and that there is a need for trained information professionals.9
But received wisdom suggests that an academic discipline without unique
and substantial research objectives would lose its internal dynamism. To
generate such a research program, we need professional schools of informa¬
tion sciences (with a plural 5). Several of these cooperative sciences offer
promising opportunities for research. If schools persist in using the singular
information science, it is our belief that as subsets of research activities in

9For a discussion of the concept of the information professional, and the results of a statistical
survey of information professionals employed in organizations in the United States, see The
Information Professional: Survey of an Emerging Field. [Debons, King, Mansfield, and Shirey,
1981.]
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 23

the area make significant progress, they will adopt identifying titles of their
own, as did the area of artificial intelligence within computer science.

Computer Science

Our first question about computer science is whether its subject is the com¬
puter itself, a highly complicated machine, or rather what is being done with
computers, namely, the processing of all sorts of information. This is contro¬
versial, even among those who teach computer science, with those who
favor the information focus recommending that the name of the discipline be
computer and information science or, better still, following the continental
European practice, informatics. [See Gorn’s lead paper in this volume.]
In their preliminary recommendations for an undergraduate program in
computer science, the Curriculum Committee of the Association for Com¬
puting Machinery supported the second position:

Computer Science is concerned with information in much the same sense that
physics is concerned with energy; it is devoted to the representation, storage,
manipulation and presentation of information in an environment permitting
automatic information systems. As physics uses energy transforming devices,
computer science uses information transforming devices. (ACM Curriculum
Committee, 1965, p. 544.)

The first position has been taken by Anthony Ralston. He insisted that
“information is no more uniquely the province of computer science than
energy is of physics. . . . The unique aspect of computer science is the
computer.” (Ralston, 1971, p. 1.) Many authoritative writers in the field
concur with this view: “Phenomena breed sciences. There are computers.
Ergo, computer science is the study of computers.” (Newell, Perlis, and
Simon, 1967, p. 1373.)
The machine in question is the general purpose automatic electronic digi¬
tal computer. Digital computers deal with discrete (numeric and nonnu¬
meric) quantities, as opposed to analogue computers, which deal with con¬
tinuous flows.10 Design of both types of computers was a specialty of the
field of electrical engineering, and operating and maintaining analogue com¬
puters has remained a subspecialty within that discipline. However, the
enormously complex task of designing, programming, operating, and main¬
taining the all-purpose digital computer gave rise to the development of a
separate discipline, computer science.

10 "Digital computers operate arithmetically, or according to other logical rules, on strings of

digits, generally on the BINARY SCALE. ... As they work by manipulating symbols and not
by direct analogues of the quantities represented, their accuracy in numerical work is limited
only by the size of their STORE and the nature of the ALGORITHM used, and they are equally
capable of operation on nonnumerical information.” (Strachey, 1977, p. 122.)
 FRITZ MACHLUP AND UNA MANSFIELD
24

The choice of the name computer for the machine in question has proved
to be somewhat misleading. To be sure, the initial applications of these
machines were to processing numbers. Digital computers, however, can and
do manipulate entities that are symbolic representations of other things. We
have no estimate, but no doubt a very large portion of their use now is for
processing nonnumeric contents (e.g., texts, images, and graphics); there¬
fore, we agree with those who have suggested that symbol manipulator
would have been a more appropriate name for the machine. [For example,
Ralston, 1971.]

A Mathematical or an Engineering Discipline?

There is no agreement among computer scientists as to whether theirs is a

mathematical or an engineering discipline. Some of the protagonists of the
thesis that computer science is essentially mathematical may have been
motivated by the age-old prejudice that gives scientists superior ranking
over engineers. If this is what is behind the controversy, we submit that it is
an irrelevancy. Mathematics is no doubt essential in the training and work of
computer scientists. The same thing, however, can be said for a large num¬
ber of other disciplines, in the natural sciences as well as the social sciences.
Neither the fact that mathematics is indispensable to the discipline, nor the
fact that most of the literature in the discipline makes heavy use of mathe¬
matical symbols and equations, is a sufficient criterion for the discipline
being essentially mathematical. Certainly, it would not be sufficient for re¬
garding it as a branch of mathematics. The use of mathematics in any field
makes the field no more a part of mathematics than the use of the English
language by the expositors makes the discipline part of English as an aca¬
demic field.
Having said this, however, we want to stress that certain specialties
within computer science, such as algorithms, automata theory, numerical
analysis, and symbolic manipulation,11 may well be called essentially mathe¬
matical. In the case of algorithms, Alan Perlis has provided us with a nice
example of where mathematical activity gives way to computer-science con¬
cerns:

The algorithms of computer programming are enormously complex and more

specialized than it is the custom of mathematics to treat. It could be argued that
the construction and studies of such specialized algorithms is out ot place in the

"Symbolic manipulation is a field “concerned with manipulation of algebraic formulas used in

algebra or calculus.” (Moses in this volume.) It should not be confused with symbol manipula¬
tion, an area that studies physical symbol systems and their processing by computers. The
applications ... of computers as nonnumeric symbol manipulators are reason enough in them¬
selves to emphasize the symbol-manipulation point of view. Another reason, however, is that
this point of view suggests an analogy with human thought processes. The human brain may
also be accurately considered to be a symbol manipulator.” (Ralston, 1971, p. 4.)
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 25

mathematics department; the labor of synthesis being too great for the insight
into theory each such study provides. Yet these algorithms . . . deliver the
power of the computer to the most demanding applications of our technological
society. (Perlis, 1968, p. 71.)

Perhaps we should leave the last word on this subject with Herbert
Simon, who has said that the “highly abstractive quality of computers
makes it easy to introduce mathematics into the study of their theory—and
has led some to the erroneous conclusion that, as a computer science
emerges, it will necessarily be a mathematical rather than an empirical
science.” (Simon, 1969 and 1981, pp. 18-19.)
There is little doubt about the engineering nature of computer science: It
designs and builds artifacts for all to see. The structure of a computer system
consists of hardware (physical components) and software (specialized pro¬
grams for managing the hardware). Hardware comprises electronic, electro¬
mechanical, and other devices for switching, storing, and communicating
signs and signals (in the form of electrical impulses) within the system.
Software consists of “a class of computer programs . . . which are used to
aid in the production, debugging, maintenance, and orderly running of the
other programs.” 12 It may be “more important than computer hardware in
determining the productivity of computer installations.” (Rosen, 1976, p.
1283.) The production of software for computers has become such a critical
and costly activity that in recent times attempts have been made to apply
engineering principles to it, with a view to standardization. This has led to
the evolution of a new specialty within computer science called software
engineering. [Wegner in this volume.]
Closely related to engineering is the question of design. In his widely read
monograph, The Sciences of the Artificial, Herbert Simon argues that “a
science of design not only is possible but is actually emerging at the present
time. It has begun to penetrate the engineering schools, particularly through
programs in computer science . . . .” (Simon, 1969 and 1981, p. 58.) Saul
Amarel points to a “fundamental reason for a close coupling between com¬
puter science and a science of design. It comes from . . . concern . . . with
the information processes of problem solving and goal-directed decision
making, which are at the core of design.” (Amarel, 1976, p. 314.)

12The layman often has difficulty in distinguishing between the hardware, firmware, software,
and programs of computer systems. One measure of distinction among them is their degree of
modifiability. Hardware refers to the material parts of the computer system; it cannot be
modified by a user. Firmware refers to computer programs (sets of instructions) that have been
embodied in a physical device that can form part of a computing machine; it also cannot be
modified by a user. Software refers to various programming aids that facilitate a user’s efficient
operation of the computer equipment (examples are assemblers, generators, subroutine librar¬
ies, compilers, operating systems, etc.); it is subject to some modification by a user. Programs
refer to sets of instructions that tell the computer exactly how to handle a problem; they are
highly modifiable and are generally tailored to fit specific user needs.
26 FRITZ MACHLUP AND UNA MANSFIELD

The Theoretical and the Innovational

The main endeavors of computer science are theoretical and innovational.

The former builds conceptual frameworks for understanding the workings of
the computer. The innovational objective is to explore new computer sys¬
tems and applications in the light of new insights. Different computer scien¬
tists have different priorities. Saul Gorn emphasizes the distinction between
knowledge orientation and action orientation, but he is convinced that the
interaction between the two will prove most productive. [Gorn in his lead
paper in this volume.]
Theoretical work in computer science over the past twenty-five years has
developed slowly. The phenomenal growth of the field has drawn heavily on
“fundamental research in solid-state physics, electronics, classical numeri¬
cal analysis, and mathematical logic.” (Perlis, 1979, p. 423.) However, there
is evidence that theories such as complexity theory (the study of resource
requirements of abstract algorithms) have come into existence because of
the computer and the variety of tasks it performs, and not in response to
problems posed purely in terms of mathematical logic. [Ibid., p. 424.]
Amarel lists the following areas of theoretical work in computer science as
“concentrating on comprehensive analysis of specific classes of phenomena
for which formal models exist . . . theory of computation, automata theory,
theory of formal languages, and switching theory” and adds that work in the
“new area of analysis of algorithms (which includes important approaches to
the study of computational complexity) promises to contribute significant
theoretical insights into problems that are in the mainstream of the computer
field.” (Amarel, 1976, p. 316.)
On another level, Newell and Simon have cited their physical symbol
system hypothesis to show that “computer science is a scientific enterprise
in the usual meaning of that term: that it develops scientific hypotheses
which it then seeks to verify by empirical inquiry.” (Newell and Simon,
1976a, p. 125.)
One of the most active areas of computer research has been the develop¬
ment of programming languages and systems. Programming has been de¬
scribed by Donald Knuth as an art rather than a science. [Knuth, 1974a.]
Programming languages are the media through which humans instruct com¬
puters to execute particular tasks. Languages operate on various levels, and
versions have been developed for particular groups of users, particular areas
of study, and so on. The two best-known computer programming languages
are FORTRAN (for scientific applications) and COBOL (for business data-
processing). But while new languages continue to be developed, “language
invention no longer plays a major role in computer science research.” (Per¬
lis, 1979, p. 426.)
Of greater concern now is the development of programming systems,
which has led to a theoretical approach to the problems of program develop¬
ment that is based on structured programming and the use of mathematical
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 27

techniques for verification and proof in connection with the production of

programs. “The aim is to produce programs that have been proved to be
correct before they are tested on a computer and thereby to eliminate much
of the program-testing activity.” (Rosen, 1976, p. 1285.)
Another major area of innovational research in computer science is infor¬
mation storage and retrieval. This involves the study of efficient methods of
storing and processing large quantities of data in a computer and methods
of searching for and retrieving these data. It includes the very active field
of database management.
These, then, are a few of the most publicized areas of theoretical and
innovational research in computer science. But perhaps the most challeng¬
ing—and the most controversial—research area of all is that known as
artificial intelligence, which is concerned with the means by which com¬
puters may perform tasks that would “require intelligence if done by men.”
[Minsky, 1968.] This area uses most of the traditional research methods of
computer science, while adding a few special approaches of its own. (We
shall devote a separate section of this chapter to this subdiscipline and no
less than five papers in this volume.)

Relations to Other Disciplines

Although computer science is now developing its own research identity, its
boundaries are by no means clearly defined, for it overlaps with many estab¬
lished disciplines. Zenon Pylyshyn has said that

Computer scientists find themselves involved in widely diverse academic

areas. They conduct research and publish papers in engineering, mathematics,
economics, sociology, psychology, linguistics, philosophy, library science, the
biological sciences, business, law, and the humanities, as well as in other
cross-discipline areas such as communications and information theory, control
theory, and general systems theory. (Pylyshyn, 1970, p. 61.)

Pylyshyn was referring to the disciplines with which computer science

shares research interests, and not just to those in which the computer has
found fruitful application as a tool. The widespread use of digital computers
in virtually all academic disciplines presents a different sort of challenge to
computer scientists: The representation of knowledge and problems of the
other discipline in forms that are acceptable to computers and the develop¬
ment of computer methods for effectively handling these problems. As com¬
puter science becomes more and more involved in the business of other
disciplines, a certain amount of transfer occurs. In their paper in this
volume, Pearson and Slamecka have enumerated some of these migrations:
“information systems and database areas into management science, com¬
puter-aided design into mechanical engineering and architecture, . . . and
very-large-scale integration (VLSI) onto a new turf shared by physics, elec¬
tronics, and materials science and engineering.”
28 FRITZ MACHLUP AND UNA MANSFIELD

This leads Pearson and Slamecka to ponder on what will remain in the
core of the computer-science curriculum, and they speculate that “the
knowledge-oriented component of this core will be concerned principally
with the theory of signs, sign structures, sign processes, and algorithms. The
disturbing thought is that, if this scenario is correct, the educational pro¬
grams in today’s departments of computer science . . . are not very rele¬
vant.” (Pearson and Slamecka in this volume.)
Which brings us back to the question with which we started: Whether
information and its processing, or the computer as a configuration of hard¬
ware, firmware, software, and programs, is the central object of study in
computer science.

Artificial Intelligence

There seems to be general agreement that artificial intelligence (AI) is a part,

although an isolated part, of computer science.13 It has been described as an
“audacious effort to duplicate in an artifact what we humans consider to be
our most important, our identifying property—our intelligence.” (McCor-
duck, 1979, p. xi.) Marvin Minsky, one of the pioneering researchers in this
field, suggests that it shares its goals with other disciplines:

With computer science we try to understand ways in which information-using

processes act and interact. With philosophy we share problems about mind,
thought, reason, and feeling. With linguistics we are concerned with relations
among objects, symbols, words, and meanings. And with psychology we have
to deal not only with perception, memory, and such matters but also with
theories of ego structure and personality coherence. (Minsky, 1979, p. 400.)

External Influences

Artificial intelligence benefited from a rich intellectual heritage. Work in

mathematical logic (mainly by Frege, Whitehead and Russell, Tarski, and
Church) had demonstrated that some aspects of reasoning could be for¬
malized. Work on computation by Turing and others linked the formaliza¬
tion of reasoning to the computing machines that were ripe for development.
And the abstract conception of computation as symbol processing had been
firmly established: Even before the first computers were designed, Turing
had seen that “numbers were an inessential aspect of computation—they
were just one way of interpreting the internal states of the machine.” (Barr
and Feigenbaum, 1981, p. 4.)
The cyberneticists also exerted considerable influence on research in

13 The term artificial intelligence was first used formally in the title of the Dartmouth Summer
Research Project on Artificial Intelligence, held during the summer of 1956. It is attributed to
John McCarthy, an AI pioneer and the principal organizer of that conference. [McCorduck,
1979, p. 96.]
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 29

artificial intelligence: They linked ideas about the workings of the nervous
system with logic and computation. To quote Minsky again,

AI grew out of the cybernetics of the 1940s with the hope that the limitations of
that methodology might be overcome by the new ideas coming from computa¬
tion—particularly the recognition that programs themselves, rather than their
applications, were objects of scientific interest. . . . The era of cybernetics was
a premature anticipation of the richness of computer science. The cybernetic
period seems to me to have been a search for simple, powerful, general princi¬
ples upon which to base a theory of intelligence. (Minsky, 1979, p. 401.)

Computation and Intelligence

Such questions as can computers think and are computers intelligent can be
answered only on the basis of agreed definitions of thinking and intelligence.
According to Newell and Simon, who have done ground-breaking work on
intelligent systems, “there is no ‘intelligence principle,’ just as there is no
‘vital principle’ that conveys by its very nature the essence of life.” Rather,
they point to structural requirements for intelligence, one of which is the
ability to store and manipulate symbols. “Symbols lie at the root of intelli¬
gent action, and one of the fundamental contributions to knowledge by
computer science has been to explain, at a rather basic level, what symbols
are.” (Newell and Simon, 1976a, p. 114.)
Artificial intelligence has been described as “the study of intelligence as
computation.” (Hayes, 1973, p. 40.) But there is some controversy as to
whether computation may be applied to the cognitive or subcognitive level.
[See the papers by Barr and Hofstadter in this volume.]
A distinction has been drawn between artificial intelligence and computa¬
tional psychology (simulation of cognitive processes). Margaret Boden ad¬
mits that there is “a difference in emphasis between workers who try to
make a machine do something, irrespective of how humans do it, and those
who aim to write a program that is functionally equivalent to a psychological
theory.” But she adds that

most programs depend, at least in part, on the programmer’s intuitive notions

(implicit theories) about how people function in comparable circumstances
[and] the distinction between these two categories of machine research is be¬
coming less clear and less relevant with increasing appreciation of the
difficulties involved in ‘powerful’ programming, that is, programming that en¬
ables the machine to function intelligently. (Boden, 1977, p. 5.)

Programs as Objects of Scientific Interest

Artificial intelligence research is centered on the art and science of computer

programming. Indeed, two factors that served to isolate AI within computer
science were (1) its choice of heuristic programming techniques, as distinct
30 FRITZ MACHLUP AND UNA MANSFIELD

from the algorithms favored by computer scientists; and (2) its development
of list-processing program languages, when the rest of computer science was
moving toward the use of compilers.14 Allen Newell has pointed to these
isolating factors, and added a third, which became apparent as soon as the
discipline of computer science identified itself with digital computers, rel¬
egating analogue computers to a subsection of electrical engineering: (3) The
digital-computer people initially regarded computers as machines that mani¬
pulated numbers, whereas the group interested in artificial intelligence saw
computers as machines that manipulated symbols. [Newell in his lead paper
in this volume.]
According to Newell, the choice by AI workers of programming systems
as the class of systems to describe intelligent behavior, “led to psychologi¬
cally revealing tasks” and caused the AI group to turn to the discipline of
psychology for inspiration. It was a two-way partnership, with AI exerting
an influence on psychology, principally in the areas of problem-solving and
concept formation. “However, when psychology opted to concentrate on
memory structure, psychology and AI went fundamentally separate ways.
Psychologically relevant work on memory by AI researchers did exist, but
moved out of AI into psychology.” (Newell in his lead paper in this volume.)

Connections with Brain Theory

The decision by AI researchers to turn to psychology, rather than neuro¬

physiology, for inspiration was the subject of comment by Michael Arbib:

Virtually no eminent workers in AI feel it important to relate their research to

actual brain mechanisms. Some, like Newell and Simon, are interested in using
AI to shed new light on human problem-solving, but such studies are psycho¬
logical rather than neurophysiological. The brain theorist, on the other hand,
studies neural networks, be they representative of actual brain structures, of a
regular geometry conducive to simulation or the proving of theorems, or of
interest mainly for the way in which their connections change over time. Most
of the functions dominating AI work simply are at too high a level for their
expression in neural networks to be ripe for study. (Arbib, 1975a, p. 270.)

So the cyberneticists and others work independently of AI researchers in

modeling the human brain. Arbib gives us a glimpse of their work in the field
of brain theory in his two papers in this volume.

14“List-processing languages are computer languages that facilitate the processing of data
organized in the form of lists. . . . LISP (short for LISt Processing). . . one of the most popular
of such languages . . . was developed by John McCarthy and his associates at M.I.T. during the
late 1950s and early 1960s.” (Slagle, Dixon, and Jones, 1978, pp. 778-779.) Compiling tech¬
niques for computer languages appeared with the advent of algebraic languages in the mid-
1950s. They were used originally for numerical computation; almost all of the major computer
languages are compiled (COBOL, FORTRAN, PASCAL, etc.).
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 31

The Internal Organization of Al

Within AI itself, several subfields evolved, each with its own specific inter¬
ests, research techniques, and terminology. In addition to work on special
programming languages and tools, these included research on natural-
language understanding, vision, problem-solving, game-playing, theorem¬
proving, and so forth. It soon became clear that the main AI programs all
used the same fundamental technique—heuristic search. [Newell and Ernst,
1965.] However, “as the scope of AI programs seemed to narrow, there
arose a belief by some AI scientists that the essence of intelligence lay not in
search, but in large amounts of highly specific knowledge, or expertise. . . .
The subfield called ‘expert systems’ . . . emerged in the mid-1970s in part as
a result of this emphasis.” (Newell in his lead paper in this volume.)
As heuristic search with little knowledge of the task domain gave way to
knowledge-intensive programs, the area of knowledge engineering (com¬
puter-based management of given assemblages of “knowledge”) was born.
This has led to the development of so-called expert systems that can match
the diagnostic skills of a human expert in a relatively limited domain. How¬
ever, it is important to remember that knowledge is here being used in a very
restricted sense. To quote from The Handbook of Artificial Intelligence,
“we often talk of list-and-pointer data structures in an AI database as knowl¬
edge per se, when we really mean that they represent facts or rules when
used by a certain program to behave in a knowledgeable way.” (Barr and
Feigenbaum, 1981, p. 143.)
With the development of expert systems on computers, many workers in
AI believed that the time had come for them to move away from work on toy
tasks (small illustrative tasks, such as manipulation of blocks by a “seeing”
computer) and onto work on real tasks (e.g., construction of expert systems
for medical diagnosis). Douglas Hofstadter in his paper in this volume sug¬
gests that this move is counterproductive for the basic scientific interests of
the field.

Robotics

One area of AI involvement in real tasks does not appear to be in dispute.

Bertram Raphael has stated that “interest [by AI researchers] in the devel¬
opment of . . . robot-like devices is both a natural consequence of past
developments, and a necessary stimulant to future research in the evolution
of artificial intelligence and non-numerical problem solving.” (Raphael,
1970, p. 455.) He defines a robot as “a computer-controlled mechanism that
can interact with its real-world environment in an autonomous, reasonably
intelligent manner.” (Ibid.) This is in contrast to the relatively simple de¬
vices known as industrial robots that can be programmed to perform some
repetitive task in a fixed environment. “The initial growth of industrial
robots took place largely outside of AI as a strictly engineering endeavor.
32 FRITZ MACHLUP AND UNA MANSFIELD

. . . [it] tended to minimize the intelligence involved, e.g., the sensory-

motor coordination.” (Newell in his lead paper in this volume.)
On the other hand, during the late 1960s, three major AI research groups
(at Massachusetts Institute of Technology, Stanford Research Institute, and
Stanford University) pioneered work on intelligent robots. Vision was
coupled with arms and motion and (at one center) with speech, so that
“these heuristic robot systems have exhibited behavior more complex than
that of any other robot systems in existence.” (Raphael, 1970, p. 456.) The
renewed association of AI with robotics takes advantage of the continued
advance in vision research in AI during the 1970s.
Allen Newell points to the unresolved question of whether robotics is “a
central part of AI or only an applied domain .... Do graduate students in AI
have to understand the underlying science of mechanics and generalized
coordinate systems that are inherent in understanding manipulation and mo¬
tion? Or is that irrelevant to intelligence? Cases can be made either way.”
(Newell in his lead paper in this volume.)
This unresolved question reminds us that AI does indeed share its goals
with other disciplines, as Minksy said. Clearly, a high degree of cooperation
among disciplines will be needed to achieve the ambitious goal of automating
intelligence. One area that promises to be of special interest to AI research¬
ers is the emerging discipline—or metadiscipline—of cognitive science.

Cognitive Science

This is the youngest of the disciplines featured in this volume. One of its
protagonists is not sure whether this new science, or metadiscipline, is still
only “a goal rather than a reality.” (Pylyshyn in his paper in this volume.)
Very real, however, is the fact that a number of distinguished researchers
call themselves cognitive scientists, meet in conferences, publish books, and
have established a journal named Cognitive Science; its first volume ap¬
peared in 1977. Yet, this may have been a delayed christening. One of the
founding fathers took 1956 “as the year of the birth of cognitive science.”
(Simon, 1980, p. 34.) A brief description of the new discipline is in order.
Such a description may stress the subject matter, the major problems ad¬
dressed, the techniques of analysis employed, the theoretical formalisms
adopted, and a genealogical map showing from which disciplines the new
one has descended. [Collins, 1977.]

The Scope and Nature of the New Science

Cognitive science is “the domain of inquiry that seeks to understand intelli¬

gent systems and the nature of intelligence. ... At the root of intelligence are
symbols, with their denotative power and their susceptibility to manipula¬
tion.” (Simon, 1980, p. 35.) Or, more briefly, cognitive science is “the
analysis of the human mind in terms of information process.” (Ibid., p. 34.)
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 33

The first of these definitions refers to intelligent systems, the latter to human
mind; the idea is that both human intelligence and computer intelligence are
to be studied. (Ibid.) Alternative definitions seem to avoid the terms mind
and intelligence and instead refer to cognitive systems and mental faculties.
(Sloan Foundation, 1978, p. viii.)
The resumption of studies of the mind constitutes a decisive break with
behaviorism in psychology and neopositivism in general methodology, the
two isms that had banished all nonobservables from the domain of science.
The new area of research is the result of collaboration of members of several
disciplines—of six wider fields (represented by separate departments in
most universities) and eleven interdisciplines (or specialties) that are the
outcome of pairwise bonding between several of the major six. As shown in
the figure reproduced in the appendix to Pylyshyn’s paper, psychology,
philosophy, linguistics, anthropology, neuroscience, and computer science
are the six fields of departmental rank; computational psychology (or simula¬
tion of cognitive processes), neuropsychology, philosophy of psychology,
philosophy of language, psycholinguistics, neurolinguistics, computational
linguistics, anthropological linguistics, cognitive anthropology, evolution of
the brain, and cybernetics are the eleven subdomains of cognitive science.
Not that they are all equally active or equally productive; indeed, some of
the subdomains live only a token existence.15
“The most immediate problem areas” cultivated by cognitive science are
“representation of knowledge, language understanding, image understand¬
ing, question answering, inference, learning, problem solving, and plan¬
ning.” (Collins, 1977, p. 1.) The favored techniques of analysis “include
such things as protocol analysis, discourse analysis, and a variety of experi¬
mental techniques developed by cognitive psychologists in recent years.”
(Ibid.) Somewhat questionable is the following pronouncement about the
methodological commitments of the new discipline; “Unlike psychology or
linguistics),] which are analytic sciences[,] and artificial intelligence),] which
is a synthetic science, cognitive science strives for balance between analysis
and synthesis.” (Ibid., p. 2.) All four disciplines compared and contrasted in
this quotation are clearly empirical sciences and as such cannot do without
either analysis or synthesis. Even more puzzling is the remark by the editor
of Cognitive Science that “this discipline might have been called ‘applied
epistemology’.” Epistemology is generally regarded as a major part of
metaphysics and hence nonempirical, purely analytical; and “applied” epis¬
temology could in a way fit any branch of knowledge. Fortunately, Pylyshyn
in his paper in this volume settles the question of the methodological status
of cognitive science when he declares it to be “an empirical natural sci¬
ence.”

15 With all these subdomains and contributing disciplines, one hardly expects further additions
to the list. Yet some major researchers in the field have proposed additional shareholders in the
new enterprise: epistemology (perhaps a stand-in for philosophy) and economics [Simon, 1980,
p. 33]; and education [D. G. Bobrow, 1975.]
34 FRITZ MACHLUP AND UNA MANSFIELD

That cognitive science is considered a discipline with its focus on informa¬

tion and knowledge needs hardly any documentation; but we can profit from
quoting some of the fundamental questions on its agenda: “How is informa¬
tion about environments gathered, classified, and remembered? How is such
information represented mentally, and how are the resulting mental repre¬
sentations used as a basis for action? How is action coordinated by com¬
munication? How are action and communication guided by reason?” (Sloan
Foundation, 1978, p. v. This part of the report to the Foundation was drafted
by George A. Miller.)

"Representation," a Keyword

One of the keywords in cognitive science is representation. It is not always

clear just what is being represented and what is representing.16 One could
imagine our knowledge representing something observed or assumed in the
external world or our knowledge being represented by something going on or
retained in our brain or nervous system; or our knowledge being represented
by some expressions (visual, auditory, and tactile), artifacts (signs, signals,
symbols, and codes), or various kinds of action (meaningful and communica¬
ble to others). Assuming that the stingy economizers of prepositions mean
not representation by knowledge but representation of knowledge, we rule
out the first of the three possible meanings; considering that most users
much of the time do their research using computers, we rule out the second
meaning. Thus, we conclude that the representations in question are largely
in terms of computer programs.17
One can get an idea of the role of representation in research and analysis
in cognitive science when one realizes the inordinately large number of
adjectives employed to modify the noun: In three papers in one volume,
we encountered visual representation, multiple representation, exhaustive
representation, selective representation, formal representation, seman¬
tic-network representation, predicate-calculus representation, procedural
representation, declarative representation, analogical representation, propo¬
sitional representation, semantic representation, syntactic representation,
intensional representation, universal-knowledge representation, and ad hoc-
knowledge representation. [D. G. Bobrow, 1975; Woods, 1975; R. J. Bobrow
and Brown, 1975.]

16 “Through some mapping M, a representation (call it knowledge-state 1) is created which

corresponds to world-state 1.” Ftere, knowledge is representing a particular (observed? in¬
ferred? imagined?) world state. A little later, the same author speaks of the problematic “sim¬
plicity of representing particular knowledge.” Ftere, it is knowledge that is being represented.
(D. G. Bobrow, 1975, pp. 2-3.)
17 “Clearly, the best representation for a body of knowledge depends on how that knowledge is
to be used by the program . . . .” (R. J. Bobrow and Brown, 1975, p. 104.)
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 35

Adaptive Systems, Learning, and Evolution

A delicate philosophical problem in cognitive science is created by the fact

that intelligent systems are essentially adaptive “in the face of . . . changing
environments [which] creates a subtle problem in defining empirical in¬
variants for them.” (Simon, 1980, p. 36.) Simon examines malleability and
adaptation on three time scales: the shortest, in which systems “change
their behavior in the course of solving each problem situation they en¬
counter”; a somewhat longer one, in which they learn, that is, “make adap¬
tations that are preserved and remain available for meeting new situations
successfully”; and the longest, in which “intelligent systems evolve,” an
evolution that can be biological or social or both. (Ibid., pp. 36-37.)
Realization of the adaptability of the cognitive system, especially through
learning, which may be deliberately designed, caused Simon to call cognitive
science “a science of the artificial.” The term, although it conveys impor¬
tant characteristics of the systems in question, is not self-explanatory and
has led many a student astray. Yet, we have to concede the main point of the
argument: Cognitive science “is concerned with phenomena that could be
otherwise than they are . . . and which will be altered continually as they
adapt to the demands of their environments.” (Simon, 1980, p. 45.) Inciden¬
tally, the fact that learning can be designed and, as a matter of fact, is
designed in several ways, blurs, or perhaps destroys, the methodological
boundary line between the positive and the normative. Simon is not worried
about this (actually inevitable) development. Long before cognitive science,
several other areas of study have experienced this boundary conflict; it has
been endemic in economics, with its optimization (or economic principle); in
psychology, with its positive and prescriptive learning theory; and in lin¬
guistics, with its goals of competence and performance. [Ibid., p. 43.]

Motivational and Emotional Processes

The small group of researchers that constitutes the community of research¬

ers in cognitive science consists of dynamic and creative scholars active in
this new discipline because they believe in its achievements and its promise.
There may be a few outsiders who are critical or skeptical, but we cannot yet
find any vociferous dissenters within the group. This makes it difficult to
present a profile of this young metadiscipline. We do, however, have a few
statements from leaders in the field, where they admit that little progress has
been made on (or beyond) some of its frontiers. One of these relates to the
understanding of motivational and emotional processes and their interaction
with purely cognitive processes. [Norman, 1980.]
The lack of progress on this front is explained by the fact that “the
communication codes appear to be so radically different in the affective and
cognitive subsystems. Information in the cognitive system (or at least a
considerable part of it) is encoded symbolically, while the signaling systems
36 FRITZ MACHLUP AND UNA MANSFIELD

for motivation and emotion appear much more analogical and continuous.”
(Simon, 1979, pp. 383-384.) Since computer simulation has been the major
research technique in cognitive science but seems unsuitable to research on
affective processes, the slow advance in the study of motivation and emo¬
tion is not surprising.

Semiotics and Linguistics

Of the eleven interdisciplines listed as subdomains of cognitive science, no

less than five are areas cognate to linguistics, or rather interdisciplines of
which linguistics is one of the partners: psycholinguistics, neurolinguistics,
computational linguistics, anthropological linguistics, and philosophy of lan¬
guage. Five out of eleven sounds like a large share; however, the subdo¬
mains’ eligibility for listing was based on some research going on in the
subdomains, not on great activity or productivity. It seems, moreover, that
the linguists were, more or less, ‘‘sleeping partners” in the various under¬
takings, or that the joint research was done largely by noncertified or visiting
linguists, that is, by members of the cooperating discipline doubling as lin¬
guists. If one asks some academic linguists about cognitive science or any
other of the studies of information, there may be a blank look in the eyes of
some and a disclaimer of interest voiced by others.
In addition to the five interdisciplines that link linguistics with studies of
information and knowledge, there is the metadiscipline of semiotics, the
science of signs, which is ordinarily counted among the subjects within the
wide domain of information science. Indeed, Gorn, and especially Pearson
and Slamecka in their papers in Section 2 of this volume agree that “infor¬
matics is a semiotic discipline.” Yet, many (perhaps most) traditional lin¬
guists reject semiotics as a legitimate member of their family, treating it as a
stepchild, or stepfather, if at all;18 some of the papers included in Section 4
reflect this attitude.
Perhaps one should understand it as a part of usual academic-political
practice if members of an old established discipline do not easily accept
collaboration with a new fledgling interdiscipline. This cannot explain lin¬
guists’ lack of enthusiasm for a close link with semiotics, but it may be
pertinent to their attitude toward the subdomains of cognitive science. The

18To avoid a confusion of historical timing, we should note that, whereas some of the interdisci¬
plines involving linguistics have only recently arrived on the scene, semiotics is by no means a
young subject. “Signs,” as distinguished from “Things” and “Actions,” were among “the
three great Provinces of the intellectual World” mapped by John Locke in his famous Essay.
[John Locke, 1690.] Locke took the term semiotics from the Greek Stoics; modern semiotics
began with Charles Sanders Peirce. [Peirce, 1867.] The Swiss linguist Ferdinand de Saussure is
credited with important developments. [Saussure, 1916 and 1959.] Most widely quoted is the
exposition by Charles W. Morris. [Morris, 1939.] The studies by Roman Jakobson are cited with
special respect. [Jakobson, 1962; 1971.] A recent survey of the state of the art was published by
Thomas A. Sebeok. [Sebeok, 1976.]
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 37

affection of scholars not brought up as traditional linguists is not requited by

the proper linguists. This coolness of linguists vis-a-vis their sometimes
ardent suitors seeking active collaboration is well described by Allen Newell
in the case of psycholinguistics. [See Newell’s lead paper on artificial intelli¬
gence in Section 3.]
Another gray area within this many-faceted discipline is mathematical
linguistics. Perhaps we may quote here an apt comment by Vasilii V.
Nalimov, the Russian mathematician and all-round scientist:

Linguistics, one of the most ancient sciences, is also losing its humanistic
appearance. Mathematical linguistics has been created, which, in the manner
of Bar-Hillel, can be divided into statistical linguistics dealing with the fre¬
quency analysis of symbol systems and structural linguistics dealing with con¬
structing abstract models of language. If the first branch can be regarded as a
result of the mathematization of science, the second results from the humaniza¬
tion of mathematics: problems emerging within the humanities are formulated
in the frame of mathematics. At any rate, Chomsky’s theory of contextfree
languages is clearly a mathematical subject generated by linguistic problems.

Linguistics has also acquired some purely engineering aspects. The problems
of machine translation, working out languages for computers, and especially
the problem of “man-computer dialogue’’ have added engineering features
even to such a purely humanistic field as semantics, though the principal prob¬
lems of semantics have retained their humanistic core. (Nalimov, 1981, pp.
204-205.)

Semiotics

Realizing that the science of signs includes the science of those signs that
make up language, one cannot help seeing in semiotics the wider field that
embraces linguistics. This does not imply, however, that every semioticist
must be a linguist, nor that every linguist must be a semioticist. Division of
labor may be strongly indicated, but one expects that the two groups know
what in general the others are talking about and that both avoid using the
same terms in different meanings. The second expectation is not fulfilled, in
that one of the keywords, syntactic, is ambiguous.
In semiotics, syntactic is the level of sign theory at which the observer
finds that he or she has reason to recognize certain things as intended signs
but does not understand what they signify. In linguistics, syntactic refers to
rules or regularities in grammar regarding the coordination of words and
their functions in the structure of a sentence (like subject and predicate).
Syntactic analysis in linguistics thus represents a higher level of sign recog¬
nition than in semiotics.
Semiotics distinguishes three levels: syntactics, semantics, and pragmat¬
ics, dealing with nine kinds of signs (such as tokens, types, and tones on the
syntactic level; indexes, icons, and symbols on the semantic level). A pre¬
syntactic level (or dimension) has been proposed to cover instances where
38 FRITZ MACHLUP AND UNA MANSFIELD

the observer sees some objects, such as ink marks on paper, patterns of
chiseled grooves on rocks or stones, or tree stumps in the woods, but does
not know whether they are signs; that is, artifacts that signify something, at
least to the person or people who produced them and wanted to communi¬
cate to others. [Seiffert, 1968, pp. 80-81.] One speaks of the syntactic di¬
mension when one clearly recognizes that the observed objects are signs but
does not know what they signify. When what is spoken, written, or printed is
in a foreign language that the observer does not understand, the signs are
still in the syntactic dimension. The borderline of the semantic dimension is
blurred when communication is in the recipient’s own language but on a
technical level not comprehensible to him, like an argument in the jargon of
another scientific discipline. The reader or listener may understand almost
every word, and even its grammatical function in each sentence, but still not
grasp the meaning of the sentences. To be at the truly semantic level, those
at the receiving end have to be somewhat familiar with the subject matter
and/or with the linguistic peculiarities in which it is treated. For example,
native Germans with twelve or fourteen years of schooling may fail to under¬
stand Kant’s Critique of Pure Reason if they have had no background in
philosophy. While these readers are on the semantic level with regard to the
words of the text, they are only on the syntactic level as far as Kant’s total
message is concerned. [Seiffert, 1968, pp. 84-85.] In the pragmatic dimen¬
sion, semantic comprehension is associated with action, as when the mes¬
sage is an invoice requiring payment or a summons to appear in court.
What contributions has semiotic research made to either the study of
natural language or to studies of information? This question is debated.
Views expressed in various papers in this volume show disagreement. Pear¬
son and Slamecka hold that semiotics has promoted progress in informatics.
According to a skeptical view, “the accomplishments of the research
program in general semiotics . . . have not been spectacular, unless one
classifies the achievements of cybernetics under this heading.” (Weinreich,
1968, p. 169.) As to linguistics, the same source has this to say: “Although
no sign system equals language in the variety and overlapping of semiotic
devices employed, it has been instructive to embed the study of natural
language in a broader investigation of sign phenomena of all kinds, including
substitutes for language (e.g., flag codes) and extension of language (gesture
patterns, chemical formalism, etc.).” (Ibid.)

Grammar, Inherited or Learned?

The old controversy about heredity versus environment (nature versus nur¬
ture) as major determinant of human intelligence reappeared in the field of
linguistics: Was the understanding of, or feeling for, grammar entirely a
result of learning or was there a predisposition for it due to genetic evolu¬
tion? The terms chosen for these notions are strangely different from those
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 39

cusLomary in other areas of research and debate: Linguists like to speak of

“nativism” (for the genetic source) and empiricism (for milieu and learning).
The scholars associated with the two sides of the controversy are Jean
Piaget, the psychologist who taught us about learning stages in children, and
Noam Chomsky, the linguist who developed the theory that an understand¬
ing for grammar is an innate characteristic of humans. [Piatelli-Palmarini,
1980.] The controversy is still going on, and some of the papers in Section 4
of this volume include observations on the theme.

Cybernetics

More than any other discipline, cybernetics sprang from the head of one
founding father, Norbert Wiener.19 [Wiener, 1948 and 1961.] He regarded
cybernetics as “a new scientific subject,” which at the time of the second
edition was “an existing science.” (Ibid., p. vii.) The question is only
whether the scope of that science coincides with that of Wiener's book, in
the sense that cybernetics comprises all that is in Cybernetics, but not more.
Wiener himself held that “the center of interest” in a subject that has “real
vitality” is apt to “shift in the course of years.” (Ibid.) Hence, cybernetics
should not be limited to what Wiener put down in his book. On the other
hand, not all that is covered in the book can reasonably be said to be
cybernetics, since he devoted many pages and several chapters to applica¬
tions of cybernetics to various empirical subjects, such as computing ma¬
chines, psychology, psychopathology, social organization, learning theory,
electrophysiology, and what not.
A recent leaflet by the American Society for Cybernetics contains a mot¬
ley collection of definitions of cybernetics. After giving Wiener’s formula¬
tion (from the title of his book) and Ampere’s obsolete suggestion of the
science of government, the leaflet supplies four definitions ranging from
epistemology via organization theory and the science of form and pattern to
the art of manipulating defensible metaphors."0

19It is easy to show that some of the fundamental ideas presented in Wiener’s Cybernetics—
say, the notion of (though not the term) feedback—had been known to earlier writers about
reflexes and related physiological phenomena. Also, Bertalanffy’s general system theory, pre¬
sented in German articles, anticipated Wiener’s publication. Still, the exposition of these ideas
under the name cybernetics is unquestionably Wiener’s.
20To quote in full, “For philosopher Warren McCulloch, cybernetics was an experimental
epistemology concerned with the communication within an observer and between the observer
and his environment. Stafford Beer, a management consultant, defined cybernetics as the
science of effective organization. Anthropologist Gregory Bateson noted that whereas previous
sciences dealt with matter and energy, the new science of cybernetics focuses on form and
pattern. For educational theorist Gordon Pask, cybernetics is the art of manipulating defensible
metaphors, showing how they may be constructed and what can be inferred as a result of their
existence.” (From a leaflet published by The American Society for Cybernetics, 2131 G Street
NW, Washington, D.C., 20052.)
40 FRITZ MACHLUP AND UNA MANSFIELD

Murray Eden in his paper in the present volume considers two related
insights as the most important findings of cybernetics: First, that there is an
essential unity in the set of problems in communication, control, and statisti¬
cal mechanics (noisy phenomena), whether they are to be studied in the
machine or in living tissue. Second, that “the computing machine must
represent almost an ideal model of the problems arising in the nervous
system.” (Eden in this volume.)

Connections or Applications?

Eden has been able to show connections, including some close ones, be¬
tween cybernetics, either as a science or (more often) as a viewpoint, and
some thirty other disciplines or specialties. This many-sidedness alone
should suggest that the connections are not empirical conjunctures of phe¬
nomena that are common subjects of observation and explanation by the
different disciplines but, instead, that cybernetics offers a technique of
analysis applicable to a large variety of phenomena, natural, social, and
man-made. Without hinting at the possibility of cybernetics being a
methodological research program of wide applicability, Eden comes to this
conclusion: “The notions of cybernetics have permeated many disciplines—
computer science, information theory, control theory, pattern recognition,
neurophysiology, psychophysics, perceptual psychology, robotics, and the
like. Having been integrated into them, cybernetics has performed the func¬
tion for which it was proposed.” (Eden in this volume.)
This conclusion sounds as if cybernetics has fulfilled its mission and can
be buried—which would not be the appropriate thing to do in the case of a
successful and fertile, methodologically sound scientific research program.
Perhaps another interpretation offered by Eden as well as by Elias, Arbib,
and Mattessich explains the virtual disappearance of cybernetics from the
Western academic establishment. According to that interpretation, cyber¬
netics is a part of general system theory and has been completely absorbed
by this much broader scientific research program.
When Arbib holds that “the history of cybernetics is, in large part, also
the history of cognitive science,” one might infer that he sees cybernetics as
an empirical discipline like cognitive science. The same impression is con¬
veyed when he reports that members of his research group used “cybernet¬
ics to refer to a conjoined study of brain theory and AI”—both clearly
subjects of empirical research. On the other hand, Arbib speaks of modern
system theory as “the descendant of Wiener’s control theory”—and this
agrees with our conception.

A Part of System Theory?

Richard Mattessich (in his paper in Section 6) supplies a neat enumeration of

what he regards as the basic notions of cybernetics: organization and emer-
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 41

gent properties; structure, hierarchy, and evolution; function and goal orien¬
tation; information, control, and feedback; environment and its influence;
system laws and mathematical homologies. According to Mattessich, cyber¬
netics is most appropriately regarded “as a subset of systems research.’’-1
We endorse this view.
The tendency to link cybernetics with general system theory and, even
further, with information theory began in Europe almost as soon as Norbert
Wiener’s and Claude Shannon’s works were reported. Thus, in a series of
conferences held in Paris in the spring of 1950 on Cybernetique: theories du
signal et de l’information, the first speaker defined cybernetics as “the sci¬
ence of relations, controls, and transmissions of information.’’ (de Broglie,
1951, p. v.) This definition announces the unification of the triad: The
science of relations is general system theory, the science of controls is
cybernetics proper, and the science of transmissions of information is infor¬
mation theory. The speaker subsumed them all under the name cybernetics.
This merger of the three disciplines under a single name can be seen on
the organization charts for the academic enterprise in Europe, but not in the
United States. We doubt that most American representatives of general
system theory and information theory would be prepared to accept such a
take-over offer. On the other hand, a take-over of cybernetics by system
theory seems to be more or less a fait accompli.

System Theory

We have long been baffled by the variants we encountered in the literature

on systems. Were system theory and systems theory two different fields of
study? Was system theory different from general system theory? What dis¬
tinguished system theory from systems analysis, systems research, the sys¬
tems approach, and systems science? All these appellations occur not only
in the earlier literature but also in papers contributed to this volume. And
there is systemics too.

Distinctions and Differences

Some of our contributors hold that genuine differences, either in methods or

methodological positions, account for the distinctions. We have concluded
that hypotheses about intended meanings of singular versus plural are inter-

21 Some writers go farther and make the two disciplines one and the same: “Cybernetics is the
science of general system theory.” (Paul E. Martin, 1981, p. 11.) This, however, is the view of a
writer based in Central Europe, who thinks that cybernetics serves chiefly to devise systems of
“improved, simplified and less costly information handling [by management] in business and
government,” and that Stafford Beer, who wanted to design the socialist planning system of
President Allende in Chile, is “one of the greatest living cyberneticians of our time.” (Ibid., p.
14.) As Eden and Elias mention in their papers in this volume, the meaning of cybernetics in the
Soviet Union is wider still in that it comprises mathematical economics and econometrics.
42 FRITZ MACHLUP AND UNA MANSFIELD

esting but not convincing. For we established that the same writers alter¬
nated in casual inconsistency between using plural and singular interchange¬
ably in the same papers, chapters, pages, and even paragraphs. (Why were
they not queried by their editors? We, too, decided to grant our contributors
freedom to be inconsistent in this matter.)22
The question which noun is modified by general in general system theory
proved even more bewildering. Was it the general theory of systems or the
theory of general systems? At one point, we became convinced that the
absence of a hyphen between the first two words was not deliberate but mere
sloppiness. Our evidence seemed strong: The yearbooks of the Society for
General Systems Research have carried the title General Systems, and au¬
thors wrote of general systems as contrasted with special systems. We were
disabused of this conclusion when we saw foreign-language renditions of
general system theory—all emphasizing the generality of the theory, not of
the systems.23 We should not be surprised, however, if a parishioner of one
of the various denominations of system theory were to castigate us for
downgrading the theory of general systems. The ferocity with which differ¬
ent system theorists attack one another’s methods and methodological posi¬
tions has shown us that whatever is said about systems will make some of
the partisans very angry.
There is no agreement about definitions of either system or system
theory.24 This need not bother us: Scholars (scientists) may disagree on how
to define a term and yet agree on its meaning. They are too proud to imitate
the layman’s excuse, “You know what I mean,” when he is aware of failing
to express himself clearly. And scholars are usually able to convey clear
meanings without formulating explicit definitions; not so when they speak
about systems. Still, three or four notions are mentioned as fundamental by

"One of the founding fathers of this subject matter, Ludwig von Bertalanffy, in his book on
General System (singular) Theory wrote of the systems (plural) approach, and he reported that
the proposed name of the Society for General System (singular) Theory was changed to Society
for General Systems (plural) Research. [Bertalanffy, 1968, pp. 4, 15 ] In the foreword to his
book, General System (singular) Theory, he used Systems (plural) Theory in the second sen¬
tence and in the next sentence, stated that “Systems (plural) Science, or one of its many
synonyms, is rapidly becoming part of the established university curriculum.” (p. vii.) We can
easily confirm this last statement: The system theorist George Klir professes his discipline in a
Department of Systems (plural) Science. In support of those who favor the singular, we may
mention that the Theory of Games (plural) is always rendered as Game (singular) Theory;
likewise, the Theory of Numbers (plural) is transformed into Number (singular) Theory. On the
other hand, one speaks of a numbers (plural) game, and economists talk about commodity
markets and commodities markets without awareness of a possible distinction.
23Five years before his article “An Outline of General System Theory” [Bertalanffy, 1950],
Bertalanffy published in German a paper “Zu einer allgemeinen Systemlehre.” [Bertalanffy,
1945.] Similarly, we find “La theorie generale des systemes” in French and “La teoria generale
dei sistemi” in Italian.
24 “Definitions of general systems theory differ. . . . The divergence begins at the very start
when attempts are made to define a system.” (Rapoport, 1976, p. 11.)
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 43

most system theorists: a whole (or assemblage, set, group, etc.), parts (or
elements, objects, constituent entities, members, etc.), relations (among the
parts), and sometimes, as an afterthought, environment (outside the bound¬
aries of the system or also within it for each entity).

Interactions or Just Any Interrelations?

The biggest question arises with regard to the relations. Some writers
specify interactions and thereby restrict the class of possibly relevant rela¬
tions; others say interdependence, which restricts it less, but some speak of
interrelations, which includes much more than interdependence and interac¬
tions. Relations among the entities may be of many kinds—for example,
logical (contrariety, transitivity, etc.), spatial (next to, below, etc.), com¬
parative (larger, faster, etc.), attitudinal (friendly, uncooperative, etc.), hier¬
archical (foreman, boss, etc.), kinship (son, sister, etc.)—without necessar¬
ily implying particular influences, stimuli, responses, actions, or reactions.
Most of the concrete systems and special systems that are presented in the
literature (as illustrations of the nature and significance of general systems)
feature mutual influences, attractions, communications, and interactions.
Exceptions, however, are not infrequent, and when the argument is pre¬
sented in mathematical language, with any one element changing as a func¬
tion of changes of all others, one can no longer make out the nature of the
interrelated parts and the nature of their interdependence. There may be
humans (acting on command, request, advice, or in spite; or in order to gain
or to survive); cells (in a living body); neurons (firing at neighboring ones);
species of animals (multiplying or dying out or evolving physiological, ana¬
tomical, or behavioral traits); molecules in different arrangements (say, in
chemical substances); electrical impulses (transmitted over a channel). But
the parts of the system may also be shapes and colors (on a canvas or
screen); tones in various intervals (in melodic or harmonic consonances and
dissonances); ink spots on pieces of paper; all sorts of patterns and
configurations, arrangements, and organizations. Indeed, it is the very idea
of a general theory of systems in general to say something about all kinds of
entities that are interrelated members of a set (which may be a subset of
another set) or of a whole (which may be part of another whole).
The question is whether much of great importance can be said about all
possible systems.25 Important in this context may mean helpful to inves¬
tigators in several disciplines in their research, analysis, and interpretation;
or it may mean helpful to organizers and managers in the practical world ot
engineering, business, and government; or helpful to philosophically minded
scholars in their abstract reflections about the scientific enterprise in
general.

25 “.. . all we can say about practically everything is almost nothing.” (Boulding, 1956b, p. 197.)
44 FRITZ MACHLUP AND UNA MANSFIELD

Most writers on system theory see the value of their formal theory in its
aid to various other disciplines, formal or empirical. Particularly, general
mathematical systems theory is said to provide foundations for other disci¬
plines, to “offer both a conceptual framework and a working methodology.”
(Mesarovic, 1968b, p. 60.) According to one authority, the logical connec¬
tions among phenomena, model, and system are expressed in these two
statements: “(a) A theory of any real-life phenomena ... is always based on
an image, termed a model, (b) Without introducing any constraints what¬
soever the formal, invariant, aspects of that model can be represented as a
mathematical relation. This relation will be termed a system.” (Ibid.)

General, Special, and Concrete Systems

We have mentioned general, special, and concrete systems; perhaps we

should state what, in our opinion, these adjectives may mean. A general
system is constituted by the interrelated (interdependent, interactive) ele¬
ments a, b, c, . . . in an ensemble S. We are not told what the elements are
and what the ensemble is—they may be just anything. A special system is
less abstract, in that we are told that it represents, for example, an economy
in which individuals, firms, and perhaps also a government are interacting;
or a human body, in which genes, nerves, blood vessels, and so forth,
receive stimuli and transmit responses; or a solar system, in which stars,
planets, and satellites attract one another; or a machine, in which different
parts are doing what the successful designer has contrived for them to do.
Specifications about interrelations among the elements of such systems will
evidently take account of what the theorists who invented (or discovered?)
the system believe to be plausible. The system theorist may still write a, b,
c, . . . for the elements of a special system, but he or she will tell the reader
whether these letters symbolize people or cells or celestial bodies or ma¬
chine parts. A concrete system refers to presumably observable things, for
which data can be procured. Thus, a concrete system is linked to happenings
at a particular time in a particular place or at least to so-called records,
genuine or invented by statisticians, of so-called observations, actual or
inferred. For example, a concrete system of the economy may have real
numbers for the money supply in the United States at certain dates and for
the gross national product within certain periods. Most of the numbers in
economic models are even more fictitious than the statistical aggregates
mentioned, so much so that West Churchman in his paper in this volume
remarks that “systems information does not correspond to empirical infor¬
mation at all.” Such an admission is, of course, in flagrant contradiction of
the confidence of mathematical system theorists who believe that there can
be real data and exact and complete data that uniquely determine a model
(though not in economics).
In any case, concrete systems, with numerical values substituted for all
elements and relations, are incorporated into the body of the disciplines that
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 45

are charged with responsibility for investigating the phenomena or records in

question. As we said in an earlier section, system theory provides the princi¬
ples for the organization of the data. In a similar vein, Talbot Waterman, the
biologist, said that, because of their “multivariable and highly intercon¬
nected organization, living things require for their effective study some over¬
all strategy like systems analysis.” (T. H. Waterman, 1968, p. I.)26 Charac¬
terizing systems analysis (or system theory) as an overall strategy for
studying complex structures is precisely what we meant when we alluded to
analytical conventions, scientific research programs, and a few other
phrases hinting at the rational, nonempirical nature of system theory and
similar formal sciences, particularly methodology.

System Theory and Information Studies

In what sense can system theory or general system theory be regarded as

one of the disciplines that study information or as based on such studies or
as contributing to them to such an extent that it belongs to that class or
family? Several partly overlapping reasons can be culled from the literature
or derived from statements not necessarily intended to answer our question.

1. System theory is said to contain (more conspicuously than other dis¬

ciplines) automata theory, communication theory (in one or more of its
meanings), control theory, cybernetics, decision theory, game theory, infor¬
mation theory (in several of its meanings), and operations research, all of
which are unquestioned members of the family. [Mesarovic, 1968b.]
2. Systems composed of discrete components are at the core of com¬
puter science. Indeed, the goals of the “systems view of computer science
are analogous to the goals of general systems theory.” (Moses in his paper in
this volume.)
3. With respect to systems of which humans are the essential members
and that are designed to aid the study of (either spontaneously developed or
deliberately designed) societal organizations, the interrelations are largely in
the form of (direct or indirect) communications through meaningful signs,
ordinarily through (spoken, written, or printed) verbal messages yielding
information (in the semantic sense).
4. Where systems are collections or societies of animals, the interrela¬
tions are usually in the form of communications through signals that are
interpreted as meaningful messages yielding (nonverbal but still semantic)
information.

26 Waterman went on to say that “[djepending on the available data and the purpose of the
analysis any, or more usually several, of a wide range of specific techniques may be employed
ranging from the use of information theory and cybernetics to computer simulation or multivari¬
able statistical analysis.” (T. H. Waterman, 1968, p. 4.)
46 FRITZ MACHLUP AND UNA MANSFIELD

5. In systems whose members are neither humans nor animals, relevant

interrelations among the inanimate parts are often characterized as com¬
munications that yield information (in a metaphoric, nonsemantic sense).
6. In open systems, where inputs are received from the outside-
environment, these inputs are often treated as matter-energy-information¬
negative entropy. A closed system would gradually run down to a state of
maximum entropy; only an open system, receiving inputs of “negentropy,”
can be maintained (and may even develop). The equivalence (in some re¬
spects) of energy and information is the subject of the mathematical theory
of communication, widely known under the alias information theory.
7. Among the many kinds of special systems are information systems,
designed to optimize the operations of collecting pertinent data, storing
them, retrieving them at appropriate moments, processing them in appropri¬
ate ways, encoding them, transmitting them over appropriate channels, re¬
ceiving and decoding them—all this in a fashion that maximizes the surplus
of benefits over costs. Systems theory will guide the designing of informa¬
tion systems and many of the activities of system engineers and information
professionals.
8. General system theory develops “a framework of general theory to
enable one specialist to catch relevant communications from others” and
allow knowledge to grow “by the receipt of meaningful information—that is,
by the intake of messages by a knower which are capable of reorganizing his
knowledge.” (Boulding, 1956a, pp. 11-12.)

This is not supposed to be a complete statement of connections between

system theory and information. Moreover the various connections should be
questioned regarding the compatibility of the terms employed: The meanings
of system varied somewhat and the meanings of information varied in essen¬
tial respects, indeed so much that one may raise the charge of equivocation
unless one charitably accepts a plea of innocent deception. Still, the inclu¬
sion of general system theory in the class of disciplines concerned with
information seems legitimized.
This should not mean, however, that general system theory has no place
outside that class. We would not want the celestial system to be treated as a
discipline concerned with information. If general system theory is seen as a
part of general methodology, it would be wrong to limit its relevance to
studies of information, the empirical phenomenon.

Information Theory

Those who have read ten or more different expositions of this field of re¬
search (and have not been influenced by a charismatic teacher) cannot help
being uncertain about the field’s real subject, it scope, and its name. Infor¬
mation theory is only one of several alternative names of this discipline;
among other designations are mathematical theory of communication, com-
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 47

munication theory, coding theory, signal-transmission theory, and mathe¬

matical theory of information measurement. Is this discipline chiefly about
information or about communication or about signals? Is it largely abstract
mathematics and mathematical statistics, a branch of probability theory? Is
it perhaps a specialty of electrical engineering with emphasis on channel
capacity? Is it a part of (or closely allied with) thermodynamics and statisti¬
cal mechanics, with its focus on physical entropy, or is its major concern
telecommunication, with the focus on systems design—or is it general
enough to apply to problems of biology, psychology, and the social sci¬
ences? Every one of these claims has been made.27

Three Senses or Many More

One reason for the diversity of claims is that “ ‘Information theory’ is used
in at least three senses. In the narrowest of these senses, it denotes a class of
problems concerning the generation, storage, transmission and processing of
information [in a very special sense], in which a particular measure of infor¬
mation is used.” (Elias, 1959 and 1968, p. 253. Emphasis added.) This is
essentially Shannon’s theory.28 [Shannon, 1948.] “In a broader sense, infor-

27Some of these positions may be exemplified here by direct quotation: "Unlike Newton’s laws
of motion and Maxwell’s equations, which are strongly physical in that they deal with certain
classes of physical phenomena, communication theory is abstract in that it applies to many sorts
of communication, written, acoustical, or electrical. . . . Communication theory proceeds from
clear and definite assumptions to theorems concerning information sources and communication
channels. In this it is essentially mathematical, and in order to understand it we must under¬
stand the idea of a theorem as a statement . . . which must be shown to be the necessary
consequence of a set of initial assumptions. This is an idea which is the very heart of mathemat¬
ics as mathematicians understand it.” (Pierce, 1961 and 1965, p. 18.)
On the other hand, one of the contributors to this volume contends that “Shannon’s theorem
about the existence of codes that make it possible to transmit information at rates that come
arbitrarily close to a specified upper limit with arbitrarily low probabilities of undetected error
rooted what others called ‘information theory’ firmly as an engineering discipline.” (Kochen in
his paper in Section 5.) Strangely enough, the same author a few years earlier had characterized
information theory (of the Shannon type) as a mathematical specialty.
In contradistinction to this view, a Russian mathematician stated that “Information Theory
is one of the youngest branches of applied probability theory.” (Khinchin, 1956; English trans¬
lation, 1957, p. 30.)
A less apodictic position is taken by another of the contributors to this volume. He holds
that after “the popularity of the information-theoretical approach, especially in the 1950s, [the]
later disillusionment with that approach, especially among psychologists, stems from an initial
lack of public clarity as to the scope and limits of Shannon’s theory . . . , and especially from a
widespread failure to distinguish the concept of information per se from various measures of its
‘amount’. It should be added that some of the most generally useful qualitative concepts of
information engineering . . . , such as feedback and feedforward, have quite independent origins
and owe nothing to the mathematical theory of information measurement.” (MacKay in his
paper in this volume.)
28 “This area is also called ‘coding theory’ and . . . ‘the mathematical theory of communica¬
tion’.” (Elias, 1968, p. 253.)
48 FRITZ MACHLUP AND UNA MANSFIELD

mation theory has been taken to include any analysis of communications

problems including statistical problems of the detection of signals in the
presence of noise, that make no use of an information measure.” (Elias,
1959 and 1968, p. 253. Emphasis added.) “In a still broader sense, informa¬
tion theory is used as a synonym for the term ‘cybernetics’ ” and thus
includes the theories of servomechanisms, automata, communication, con¬
trol, “and other kinds of behavior in organisms and machines.” (Ibid.)
One can easily extend this list, especially if one includes the meanings
intended by users of the name information theory who are not committed to
the technical writings on the discipline in the three senses just distinguished.
Many who respect the meaning of the word information in common par¬
lance, in business, public, social, and private affairs, affirm that any schol¬
arly study of those social, linguistic, and psychological processes and
phenomena can reasonably be called information theory. For example, Bar-
Hillel and Carnap, concerned exclusively with semantic information, re¬
garded themselves as information theorists. [Bar-Hillel and Carnap, 1953b.]
Likewise, Helmut Seiffert, a German philosopher who wrote one of the most
sensible books on information, views information theory through a very
wide lens. [Seiffert, 1968.] Dean Jamison distinguishes “six alternative theo¬
ries of information,” all of them semantic, unrelated to electronic processes.
[Jamison, 1970, p. 29.] However, such authors are usually not regarded as
authentic representatives of information theory, since that designation has
been assigned by the academic community to the specialists concerned with
what we would prefer to call the mathematical theory of signal transmission.
That the designation information theory has been used in so many senses
is, of course, closely connected with the multifarious meanings in which the
word information has been used. Although Machlup devotes a special sec¬
tion of the Epilogue to that theme, a few comments on the equivocal use of
this word may not be amiss at the present juncture.

Strange Uses of Common Words

Virtually all prominent representatives of the field under discussion have

expressed regret about the name under which it has become known and the
use of the word information for something that is, at best, related to certain
aspects of specific techniques of signal transmission. The main protagonist,
Claude Shannon, avoided the name information theory in his publications;
he called it mathematical theory of communication. [Shannon, 1948; 1949.]
He also had misgivings about giving a special and uncommon meaning to the
word information. Myron Tribus in his paper in the present volume recounts
that Shannon had consulted John von Neumann about this nomenclature
and von Neumann advised him to call the new thing entropy, which few
people could confuse with anything with which they thought they were
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 49

familiar.29 Warren Weaver, in his attempt to interpret Shannon’s theory,

warned his readers that the word information was used here in a ‘’rather
strange way, [in] a special sense which . . . must not be confused at all with
meaning.” (Weaver, 1955, pp. 100, 104.)
Shannon was not the first to use the term in a strange way. Ralph Hartley,
one of Shannon’s predecessors at Bell Laboratories, had said information
when he spoke of the successive selection of signs from a given list, regard¬
less of meanings. [Hartley, 1928.] Some authors would like to involve the
physicists and insinuate that early writers on thermodynamics had spoken of
entropy in terms of information. To be sure, the formula developed by
Boltzmann was an early exemplar of the species developed by Shannon, and
Boltzmann’s subject was entropy; and, of course, he spoke of ordered and
not-ordered movements, mixtures, and states. But he did not use the word
information in the context. [Boltzmann, 1872; 1877; 1886 and 1905a; 1904
and 19056.] Perhaps one could say that entropy and uncertainty were logi¬
cally correlated and that uncertainty and information could be logically re¬
lated by the use of appropriate (special, arbitrary, or contrived) definitions.
Still, one cannot in fairness blame physicists for the confusion between
signal transmission and information, though some may be blamed for failing
to distinguish between information and observation. In any case, the roles of
entropy and uncertainty in information theory are complex and controver¬
sial, and they involve not only semantic but also methodological considera¬
tions.
Communication, one of the possible alternatives for the term information,
is also a word with multiple meanings, some of which are concerned with
making oneself understood. (For example, the statement that a speaker does
or does not communicate well refers to his or her capacity to convey to the
audience the meaning of what he or she intends to communicate.) Some use
communication to indicate bilateral or two-way information. [Boulding,
1955, p. 111.] (The idea in this case is not only a meaningful message from a
speaker to a listener but a meaningful dialogue between discussion partners.)
Yet, there are also senses of the word communication that do not imply
meanings intended by a speaker or writer and understanding by a recipient;
communication of heat is one example. There is also communication of
liquids or gases between connected vessels; finally, there is the communica¬
tion of diseases. Thus, unlike information, the term communication has for
centuries been used without any connotation of transmission of ideas to an
understanding recipient.30 We conclude that Shannon was right in calling his

29That von Neumann continued to have misgivings about speaking of information and informa¬
tion theory may be inferred from the fact that he placed quotation marks around information
theory” when he first referred to it in the introduction of his posthumously published booklet
The Computer and the Brain, [von Neumann, 1958, p. 1.]
30“Communicating, from an engineering point of view, means simply moving electronic traffic
50 FRITZ MACHLUP AND UNA MANSFIELD

results a theory of communication and he should not have allowed his fol¬
lowers to call it information theory.
Regret about the misleading name given to the theory and its keyword is
shared by many eminent writers on the subject. It is a pity, said Colin
Cherry, “that the mathematical concepts stemming from Hartley have been
called ‘information’ at all.” (Cherry, 1957 and 1966, p. 50.) What is really
involved in the mathematical theory of communication are signs and signals
and their transmission over channels of communication.

Signal Transmission and Channel Capacity

Communication theory was formulated to serve engineers who design tele¬

communication systems. “Engineers are concerned primarily with the cor¬
rect transmission of signals, or (electric) representations of messages; they
are not commonly interested, professionally, with the purposes of mes¬
sages” or their meanings. “A message is regarded as the ‘selections from the
alphabet,’ which is then put into physical form (signals) as sound, light,
electricity, et cetera, for transmission. . . . The signals reaching the receiver
represent instructions to select .... All communicable messages (i.e., ex¬
pressible by signs) may be coded into . . . binary 1, 0 sequences.” (Cherry,
1957 and 1966, pp. 168-170.) For example, if the ensemble of messages
available for selection contains eight possible messages, three signals suffice
to instruct the receiver to select the message that was selected by the
sender.31 An ensemble of 32 messages (letters or signs) requires a se¬
quence of five binary signals, since 32 — 25. To select the correct word
from a dictionary with approximately 32,000 words, a communication sys¬
tem would require only 15 binary signals, since 32,768 = 215.
For telegraphic communication, sequential discrete signals are transmit¬
ted; for telephonic communication, continuous wave forms are transmitted.
These wave forms are distinguished by amplitude (measured by the ordi¬
nate) and frequency (cycles per second). The “maximum (sinusoidal . . . )
frequency which the signals are considered to contain [is called] band¬
width.” (Ibid., p. 303.) As a rule, “the longer the duration of a signal, T, the
narrower its spectral bandwidth, F, . . . and vice versa.” (Ibid., p. 139.)
Under certain conditions, signal duration multiplied by effective bandwidth
may be treated as constant, which makes for certain upper limits in channel
capacity. The apparently continuous wave forms are broken down in some

from one place to another. It matters little if the signal represents random noise or a Shake¬
spearean sonnet.” (Branscomb, 1979, p. 143.) We wish to record dissent from this statement by
a professor of communication engineering: He emphasized that his discipline is very much
concerned with the distinction between signals and noise.
The first signal reduces the possible choices from 8 to 4; the second signal, from 4 to 2; and the
third signal commands the final selection. That three binary (or yes/no) signals suffice to select
one message from an alphabet of eight is mathematically expressed by the fact that 8 = 23.
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 51

systems into discrete, equally spaced ordinates, and only samples of succes¬
sive amplitudes are transmitted to the receiver. The composite signal re¬
ceived is, within limits, the same as the unsampled signal supposedly sent.
[Ibid., p. 142.] (Do we risk ridicule from experts if we suggest as an analogy
the sequence of still photographs in film strips showing apparently continu¬
ous movement?)
According to some specialists, the greatest contribution to statistical com¬
munication theory was Shannon’s capacity theorem, also called the coding
theorem, regarding noisy channels. It states that within certain limits, de¬
pending on specified channel constraints, it is possible “by using codes of
sufficient length to set the input rates arbitrarily close to [the channel’s
capacity] and yet maintain [despite the noisy channel] the probability of
reception error as small as desired.” (Feinstein, 1958, p. 43.) Although elec¬
trical engineers had known this before Shannon’s equations, they agree that
the formal and rigorous determination of capacity limits is of fundamental
importance.
In these few paragraphs on elements of communication theory—signal
transmission and channel capacity—we have avoided using the word infor¬
mation, but it is time to explain what the engineers mean by it. They do not
mean the what that is to be communicated, but, instead, they mean the
instruction that the sender, by signals, conveys to the receiver, commanding
it (not him or her, but it) to select a particular message from the given
ensemble of possible messages. Thus, this information in the engineering
sense is an instruction to select.32 And the rate of information is the rate at
which such instructions can be transmitted per signal or per second.
[Cherry, 1957 and 1966, pp. 170-180.] However, even the word instruction
is misleading, in that it may produce an association with verbal instructions
given to a person equipped with a mind and supposed to engage in a cogni¬
tive process. The word command has the same semantic defect. Activation
would be a more appropriate term: Activating impulses are transmitted
through the channel to the receiver. Thus, activation rate might be a suitable
substitute for information rate.

The Urge to Measure and the Choice of Words

One may wonder why people other than communication engineers were
willing to accept the engineering sense of information for contexts where no
bandwidth, no signals per second, no channel capacity, no signal/noise ratio,

32“Mathematical communication theory concerns the signals alone . . . abstracted from all
human uses” and apart from all meaning. The signals control selection of signs from the
receiver’s alphabet, as when “one teletype machine [is] communicating with another. At the
transmitting end, the operator selects and presses keys one at a time; coded electrical signals
are thereby sent to the receiving machine, causing it to select and depress the correct key
automatically. We see the receiver keys going down, as though pressed by invisible fingers.”
(Cherry, 1957 and 1966, pp. 168-169.)
52 FRITZ MACHLUP AND UNA MANSFIELD

and so forth, were involved. When scholars were chiefly interested in cogni¬
tive information, why did they accept a supposedly scientific definition of
“information apart from meaning”? One possible explanation is the fact that
they were impressed by a definition that provided for measurement. To be
sure, measurement was needed for the engineering purposes at hand; but
how could anybody believe that Shannon’s formula would also measure
information in the sense of what one person tells another by word of mouth,
in writing, or in print?
We suspect that the failure to find, and perhaps impossibility of finding,
any ways of measuring information in this ordinary sense has induced many
to accept measurable signal transmission, channel capacity, or selection
rate, misnamed amount of information, as a substitute or proxy for informa¬
tion. The impressive slogan, coined by Lord Kelvin, that “science is mea¬
surement” has persuaded many researchers who were anxious to qualify as
scientists to start measuring things that cannot be measured. As if under a
compulsion, they looked for an operational definition of some aspect of
communication or information that stipulated quantifiable operations. Shan¬
non’s formula did exactly that; here was something related to information
that was objectively measurable. Many users of the definition were smart
enough to realize that the proposed measure—perfectly suited for electrical
engineering and telecommunication—did not really fit their purposes; but
the compulsion to measure was stronger than their courage to admit that
they were not operating sensibly.33
A British scientist had also been searching for definitions that permitted
measurement of information: Donald MacKay. He came up with three dif¬
ferent definitions of measurable information, some of which had been pro¬
posed by statisticians and engineers. [Fisher, 1935; Gabor, 1946.] He distin¬
guished the logon content, the metron content, and the selective-information
content, the latter being Shannon’s measure of information. In his paper in
this volume, he writes: “It would be clearly absurd to regard these various
measures of ‘amount-of-information’ as rivals. They are no more rivals than
are length, area, and volume as measures of ‘size.’ By the same token it
would be manifestly inept to take any of them as definitions of information
itself.” With regard to the Shannon definition (in terms of probability or, as
MacKay put it, unexpectedness), he has this to say: “To try to translate
every reference to information (whether in biology or elsewhere) into a
statement about unexpectedness would be as inept, and as conceptually
Procrustean, as translating all references to a house into statements about

The irrational zeal to measure even where it makes no sense is frankly expressed in the
following statement: “In many ways it is less useful to measure the amount of information [in
Shannon’s sense] than the amount of meaning. In later chapters, however, 1 reluctantly deal
more with measurement of the amount of information than of meaning because as yet meaning
cannot be precisely measured.” (J. G. Miller, 1978, p. 12.)
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 53

its size.” MacKay’s exposition, however, was not without linguistic in¬
felicities. He was very helpful when he warned—as early as 1950—against
confusing information in the original sense of the word (where some mean¬
ingful content is expected) with amount-of-information (the latter referring
to measurable improbability or unexpectedness of signals from a given set).
But then he introduced or accepted the term information-content—which a
commonsensible reader would expect to refer to the meaning-content of the
information—and wanted that term to denote a measurable amount of infor¬
mation-transmission, regardless of any meaning-content.34 [MacKay, 1951a;
1954b; 1969, pp. 56, 156-160, 163-167.]

Confusing Explanations

The available expositions of the essential notions of information in the sense

of signal transmission are incredibly confusing to a reader who is not sat¬
isfied with superficial comprehension but insists on a full understanding of
the major aspects of the subject. The reader is constantly confronted with
terms that are defined but not made clear. We shall illustrate this with
respect to some of the keywords.
Let us begin with selection, the act of choosing one of two, several, or
many messages (or signals) from a given set (ensemble, source, repertoire,
or pool). Who selects and what for? The reader, having been treated to
several examples of coin-tossing and dice-throwing, is led to think that the
message received, for example, tails or five, is entirely a matter of chance.
This interpretation is reinforced by the next example furnished by virtually
all expositors, the case of picking any card, its face down, from a well-
shuffled deck of playing cards. So, the selection of the message is a random
choice? Hold it! The next example, almost universally, is the game Twenty
Questions. Here, the recipient chooses questions not at random but more or
less intelligently and receives presumably correct answers of yes or no. If
each answer constitutes a transmitted message, it is surely not random, but
selected in good faith to the best of the transmitter’s knowledge, be he, she,
or it a human being or a man-made machine. This example ot successive
yes’s and no’s comes closer to what the expositors are trying to show,
namely, that a sequence of steps is needed to arrive at the ultimately sought
answer. However, the conscientious reader may still be puzzled by the
division of labor in selecting in this case: The recipient selects the twenty (or

34 Not only general readers but also logicians mean by contents of information their semantic
contents. Thus, Bar-Hillel and Carnap held that the terminology and the theorems of telecom¬
munication research were unfortunately applied “to fields in which the term information was
used ... in a semantic sense, that is, one involving contents or designata of symbols....” (Bar-
Hillel and Carnap, 1953b, p. 147. Emphasis added.)
54 FRITZ MACHLUP AND UNA MANSFIELD

less or more) questions and the transmitters select the answers, though
before the questioning begins, they have also selected the correct answer to
be guessed. The relevance of all this to the process of signal transmission is
not made clear in any expositions that we have read.
The core of the theory is probability, yet few expositors care to state just
what is probable about a message transmitted. A communication system
such as the telegraph or the telephone does not usually transmit messages
that tell which side of the coin happened to come up or which number of
eyes (or dots or spots) were on the top of the die or which among the playing
cards in the deck was picked. Messages conveyed by telegraph are usually in
the form of strings of words (in a natural or artificial language) and numbers.
It is easy to understand that the signals adopted to represent letters in a
telegram should be chosen so that simpler signals stand for the most fre¬
quently occurring letters. A coding convention of this sort will reasonably be
based on the probability of particular letters being required to compose
words—where probability means the frequency of these letters in the
English (French, German, etc.) language. (The most frequent are e and t,
then i, a, n, and m.)
The frequency of individual letters occurring in an English text (or an¬
other language) matters a good deal when engineers designing a telegraph
system determine what signals ought to be used in encoding, transmitting,
and decoding. The most frequently used letters should be represented by the
shortest sequence of signals, with longer sequences of signals reserved for
less frequently used letters. In most expositions of the theory, discussion of
this encoding problem is followed by a discussion of doublets (digrams) and
triplets (trigrams), that is, sequences of two or three letters that occur fre¬
quently, infrequently, or never in the particular natural language. The bear¬
ing of these different frequencies, however, is less on the assignment of
signals to signs, but rather on the ease of detecting errors in scrambled
telegrams. (For example, sck is not likely to appear in any English word.)
Such detection, however, is unlikely to be built into the architecture of the
system but rather into the cognitive capability of the human recipient.
Turning from telegraph to telephone communication, the physical signals
used to transmit words spoken by a human voice into a microphone and
channeled through wires and cables are surely very different, and the proba¬
bility of particular letters in a written language is no longer significant. Now,
differences in pitch may be important. (We have asked ourselves whether a
telephone system optimal for Chinese speakers would be the same as one for
English.) We can send music over the telephone, and the relative frequency
(probability) of very high, very low, very loud, and very soft tones probably
matters. These issues, however, are not ordinarily included in elementary
expositions of information theory and the so-called amount-of-information
transmitted. The illustrations furnished, even by those who repudiate any
concern with semantic content, usually employ sentences that do have
CULTURAL DIVERSITY IN STUDIES OF INFORMATION 55

meaning, and this further confuses the reader. For example, when it is
pointed out that the probability of a message depends on how many possible
messages there are, and when this is followed by an example of two English
sentences (widely used in practice manuals for students of typing), the
reader will be puzzled about the information contained in either of the two
messages that was selected. Was this a selection by chance, and is each of
the messages equally probable in any other sense? [Elias, 1959 and 1968, p.
254.]
When we are told about the probability of a message, we remain puzzled
about just what is probable about a message. We cannot help (despite all
warnings) thinking first of the probability of its proving true, useful, or
valuable. Second, we think of the probability of its being selected by a
transmitter making a rational choice, either trying to be helpful, perhaps
obeying a command, or trying to deceive the recipient. Third, remembering
information theorists’ advice to forget about meaning and not to forget about
the given pool of possible messages, we think of the probability of the
message being picked at random from all possible messages. Fourth, having
learned about disturbing noise being mixed with genuine signals, we think of
the probability of the message reaching the receiver without undue noise
and, thus, as it is intended by the sender. Finally, backsliding into the
normal attitude of connecting messages with something they are supposed to
convey, we think of the probability of the message being understood by the
recipient. This half-dozen meanings of the probability of a message does not
exhaust the list, and the trouble is that formulations chosen by most ex¬
positors do little to help the recipient of their messages grasp the entire
exercise.35

Information Theory and System Theory

With the difficulties of properly interpreting the many inept and obscure
expositions of information theory, added to the fact that there is not one
information theory but several of different scope and even different objects,
it would be an unduly daring attempt to characterize the methodological
nature of the insufficiently specified discipline. The question, for example,
whether it is an empirical or an entirely analytical theory cannot be an¬
swered before one decides just what the theory says and what it denies.

35 In one of Cherry’s formulations, we are confronted with three probabilities: “the probability
of message x being sent, assessed from past observations of the transmitter ; the probability
of an x being sent, on those occasions when y is received”; and “the likelihood that, if any
particular x had been sent, the specific y would be received. ’ (Cherry, 1957 and 1966, p. 202.)
We interpret this argument as an exercise on deriving the likelihood—that is, the so-called
probability of a single event of a specific type—from prior statistical probabilities, as frequen¬
cies observed in the past, where the first is the frequency of x alone and the second, the
frequency of a conjuncture of x and y.
56 FRITZ MACHLUP AND UNA MANSFIELD

It is possible to interpret the theory as an auxiliary specialty of electrical

engineering or communications technology. Electrical impulses, air waves,
microphone vibrations, telegraphs, telephones, radios, television sets,
wires, cables, and scores of other things that make up communication chan¬
nels, encoders, decoders, and all the rest are undoubtedly observable phe¬
nomena of the physical world, and any research concerned with such phe¬
nomena is empirical (indeed, practical-empirical) in character. On the other
hand, much about the analysis of the fundamental processes involved has to
do with probability—logical and statistical (and some even subjective)—and
this would support the interpretation that the theory belongs to probability
theory, either as an integral part or as an application. To the extent that
statistical probability is based on observed frequencies of events, the theory
has a firm empirical base; to the extent, however, that the relevant observa¬
tions are merely assumed and included in a general, and therefore abstract,
model, the result is analytical. Inasmuch as this analytical-deductive argu¬
ment is formulated in terms of mathematical equations, the theory may be
considered as pure mathematics. Thus, depending on how one looks at
information theory, it may be subsumed under very different categories of
knowledge, ranging from practical-technological via empirical-statistical to
abstract-analytical.
Mihajlo Mesarovic in his paper in this volume proposes to equate systems
theory with information theory. He does so, however, after redefining both
information and information theory. He liberates information from Shan¬
non’s construct, which, he thinks, “refers only to the capacity for transmis¬
sion,” and makes it much richer and thereby suitable for use in the context
of goal-seeking behavior. Accordingly, he extends information theory
enough to make it into a broad theory of information. Then, but only then,
may Mesarovic be justified in saying, as he does, that “information theory
and systems theory are one and the same.” What this asserted equality, or
statement of identity, amounts to is merely a normative pronouncement—
that the two fields ought to be the same; or, perhaps, a protest against the
actual inequality of what now goes under the name information theory and
general system theory. For, in actual academic practice, “Information
theory is a set of concepts, theorems, and measures that were first devel¬
oped by Shannon for communication engineering and have been extended to
other quite different fields . . . .” (J. G. Miller, 1978, p. 12.)
This extension of information theory, as developed for communication
engineering, to other quite different fields has been a methodological disas¬
ter—though the overenthusiastic extenders did not see it, and some of them,
who now know that it was an aberration, still believe that they have learned
a great deal from it. In actual fact, the theory of signal transmission or
activating impulses has little or nothing to teach that could be extended or
applied to human communication, social behavior, or psychology, theoret¬
ical or experimental.
ADDENDUM

Addendum to the Prologue:

Letter from George A. Miller

[Editor's note (U.M.): As well as making suggestions for improving various parts of the
Prologue, George Miller provided us with a comment on its section dealing with cognitive
science. This was contained in a letter to Fritz Machlup dated January 18, 1983, and Professor
Machlup’s immediate reaction was: “We should publish it as it is.” However, he wanted more
time to think about it, but he died unexpectedly before we had had a chance to discuss it again. I
have decided to be guided by his initial reaction, and Miller has kindly agreed to have the letter
published as it is, with only very slight editing.]

Dear Fritz:
It has taken longer than I had hoped to read “Cultural Diversity in Studies of
Information” by you and Una. I hope my impressions will still be of some
interest to you.
I enjoyed most of it, although at points I must confess that your patience with
the territorial claims of the different cultures is greater than mine. I particularly
relished your defense of pluralism, and your interest in good ideas regardless of
whether they come labeled as “science” or “practice.” As one who lives in a
marginal science myself, I long ago discovered that attempts to work solely in
the manner that I had been taught to regard as characteristic of true science
could only stunt my intellectual growth as a psychologist. So I decided to
pursue problems wherever they led; if the result was not science, that was
unfortunate but not important ....
Your discussion of cognitive science seems to me to rely too heavily on the
views of those who take computers and artificial intelligence as providing the
definitive characteristics of this field. Personally, I find it difficult to communi¬
cate clearly my enormous debt to people like Simon, Newell, and Minsky for
helping cognitive psychology take a great leap forward, yet at the same time to
communicate with equal clarity that the human brain/mind system is enor¬
mously more complicated than, and different from, any contemporary compu¬
tational system.
An interesting point that you make might help me straighten out my difficulties
with computer-oriented cognitive science. In my opinion, computer science
applied to the problems of cognitive psychology should yield cognitive psy¬
chology. The fact that it has not, that it has yielded something novel—concepts
of mind and intelligence so abstract that it doesn’t matter whether the system
possessing them is living or dead—may mean that I have mistakenly regarded
computer science as a formal science.
But I suspect that the tensions already tearing the new field apart are worse
than that diagnosis might suggest. Contemporary computers manipulate sym¬
bols, and symbol manipulation is certainly one kind of intelligence. But
whether that kind of intelligence is the only kind remains an open question. I
can accept the claim that the brain must be a universal Turing machine, but that
58 FRITZ MACHLUP AND UNA MANSFIELD

tells me almost nothing as long as I don’t know what kind of universal Turing
machine it really is. (It is like telling me that the correct theory of the brain’s
function could be written in ink.) Unfortunately, I am convinced the brain is
not the same kind of universal Turing machine as are the machines presently
merchandised by IBM.

At this point, therefore, I have personally taken refuge in the plural s that you
endorse so strongly .... The cognitive sciences are those scientific disciplines
sharing an interest in the representation and transformation of knowledge (read
“information” in the present context). But, as you have pointed out some¬
where, sharing an interest and being a single discipline are very different
things. Moreover, my definition of the cognitive sciences is not easily distin¬
guished from your definition of the information sciences, yet it is plain enough
that your information science is a larger, more catholic assemblage than my
cognitive science. At which point I merely shrug. You, I believe, feel a com¬
mitment to rationalizing the difference.

You are certainly right . . . that “representation” is a keyword in cognitive

science. Historically, philosophers since Descartes have assumed that the
mind somehow copies, reflects, or represents the real world, so representation
is hardly a new idea. The philosophers, however, immediately raised the ques¬
tion of how we can possibly know whether or not the mental representation of
the real world is correct, true, valid. At this point Hume, then Kant, then
dozens of others were able to create professional philosophy out of the epis¬
temological (metaphysical) problems that resulted. When cognitive scientists
revert to the problem of representation, therefore, one assumes (or at least
hopes) they have a better strategy in mind than the philosophers did, that there
are other important questions to ask about representations other than their
accuracy, since that question is known to lead straight out of empirical science.

The alternative, as I understand it, is that we now have machines capable of

interestingly complex representations, and we should use them to learn what¬
ever we can about representational processes. Perhaps along the way we will
notice something that will shed light on the structure and function of mental
representations. But I have heard no guarantees, and as the difficulties become
more and more apparent, the tendency for the computer exploration to ex¬
punge psychologism (just as logic did a century earlier) and move off on its
own seems stronger and more attractive. That might be a good development,
but it would spell the doom of cognitive science as a new discipline encom¬
passing both artificial intelligence and psychology.

Some of my misgivings are reflected in your comments about motivational and

emotional processes. . . . But I (and many other cognitive psychologists) would
go further. Consider something like the memory of a pain. If, while you feel the
pain, you describe it verbally to yourself, then when you go to a doctor's office
you can remember your verbal encoding of the experience. That is the kind of
memory that computer-oriented theorists have much to say about. But suppose
you forget to encode a description verbally at the time you feel the pain. By the
time you arrive at the doctor’s office, the original pain experience itself will
have faded and your memory of it will have become quite unreliable. That kind
ADDENDUM 59

of precategorical memory for the “raw feels" of life presumably requires some
kind of analogue representation, but whatever it is, contemporary computer
science does not help us understand it any better. In other words, the impor¬
tant split here is not cognition vs. conation and emotion, but rather a split
within cognition itself: symbolically represented cognition vs. primal, nonsym-
bolically represented cognition. In short, computer-oriented theories of cogni¬
tion are not so much wrong as incomplete ....

Thank you, both of you, for letting me see this. I hope these random associa¬
tions will be of some help.
Sincerely,

George A. Miller
 .

,
SECTION 1
 INFORMATION SCIENCE
Its Roots and Relations as Viewed
from the Perspective of Cognitive Science

Zenon W. Pylyshyn

INTRODUCTION: THE HISTORICAL LINKS

Arthur Koestler once pointed out that it is commonplace for science to

“sleep walk” its way to new stages of development—to progress and
achieve dramatic new world views while not fully understanding, or indeed
frequently misunderstanding, the nature and significance of work that scien¬
tists themselves are doing. Frequently, it is only with the advantage of
hindsight that the significance of various developments can be assessed.
There are clear indications in the history of science [e.g., Butterfield,
1957] that periods of progress are coincident with major new technical and
conceptual developments or sometimes, as in the case of Galileo’s use of
geometry, with taking an existing formalism seriously as a way of under¬
standing the world. Similarly, philosophical understanding also rests on
available conceptual tools. As Susanne Langer put it, “In every age, philo¬
sophical thinking exploits some dominant concepts and makes its greatest
headway in solving problems conceived in terms of them.” (Langer, 1962,
p. 54.)
Several decades ago, the brilliant mathematician and computer pioneer
John von Neumann pointed out that in the past, science had dealt mainly
with the concepts of energy, power, force, and motion, and he predicted that
“in the future science would be more concerned with problems of control,
programming, information processing, communication, organization, and
systems.” (Burks, 1970, p. 3.)
The precise nature of the Weltanschauung that ties together this syn¬
drome of concepts is not yet clear. [See, however, Simon, 1969; Newell and
Simon, 1976a.] It seems to represent a move away from the study of material
substance toward a more abstract study of form. The new conceptual tools
leading up to the development of artificial intelligence and cognitive science
are bound up with such notions as mechanism, information, and symbol.
The earliest harbingers of these developments may well have been in
63
64 ZENON W. PYLYSHYN

symbolic logic, especially the formalist movement in studies of the founda¬

tions of mathematics, initiated in the first half of the twentieth century by
people like Hilbert, and developed by Russell and Whitehead, Frege,
Zermelo, and others. These people began the attempt to reduce the notion of
a mathematical proof to that of a game played with tokens of uninterpreted
meaningless symbols according to certain purely formal rules—rules whose
application depends only on recognizing the occurrence of the type of each
token (i.e., in recognizing anr, whenever it occurred, as being an instance of
the same particular symbol). No sooner had this goal of formalizing mathe¬
matics been formulated than it was discovered to be in principle not achiev¬
able. Godel, Turing, Church, Rosser, Kleene, Post, and others showed that
there exist true but formally unprovable propositions, formally insoluble
problems, and uncomputable functions. [See the collection of fundamental
papers in M. Davis, 1965.]
This work, carried out in the 1930s, was important for many reasons. One
is that it made precise the notion of formal mechanism or a mechanism or
process that functions without the intervention of an intelligence or any
natural being and yet can be understood without knowing about any of its
physical properties. In making this notion precise, these studies laid the
foundations for a way of conceptualizing a wide range of problems in many
different areas of intellectual endeavor—from philosophy of mind and phi¬
losophy of mathematics to engineering, and including almost every facet of
social and biological science.
Although these foundational mathematical developments of the 1930s
were probably the most dramatic signs of a major breakthrough in the under¬
standing of formal symbol systems, they were neither the only such signs
nor were they necessarily the most directly influential. For example, there
had earlier been a tradition in linguistics called structuralism, which proba¬
bly originated with the publication of Ferdinand de Saussure’s lectures in
linguistics in 1916 and soon spread considerably beyond linguistics proper.
This movement influenced the allied fields of anthropology (e.g., Levi-
Strauss), sociology [Barker and Wright, 1955], and psychology (e.g., Piaget)
and even led to what many view as the synthesis of a new interdisciplinary
field of semiotics—a term (attributed to Charles Morris) that refers to the
study of relations of signs and symbols to behavior.
If we were to try to find the major direct influences on the current
Zeitgeist—of the current ways of viewing nonphysical forces that shape our
world and our behavior (as von Neumann foresaw, in the statement previ¬
ously quoted)—we would doubtlessly trace them to a number of technical
achievements that emerged during wartime scientific work in the United
States and Britain, though it is not clear that World War II actually provided
any of the impetus for these developments. [Newell and Simon, 1972.] It is
no accident that this work involved many of the same people who contrib¬
uted to seminal developments in logic and foundational mathematics (partic¬
ularly Alan Turing and John von Neumann), for the technical developments
INFORMATION SCIENCE VIEWED FROM COGNITIVE SCIENCE 65

were very much part of the same intellectual movement. These technical
achievements included the following closely related developments:

1. Control theory, or the theory of closed-loop feedback systems, was

one of the earliest and most far-reaching of the wartime technical develop¬
ments. The need to develop automated systems for aiming and maintaining
the directional stability of radar antennas and artillery on moving vehicles
spawned an enormous surge of research and mathematical analysis, involv¬
ing people like Wiener, Nyquist, Hartley, and others. The broad and pro¬
found significance of this work soon became clear [Wisdom, 1970] and led to
an important paper by Rosenblueth, Wiener, and Bigelow [1943], which
analyzed the philosophically troublesome notion of teleology or purposive
behavior in terms of feedback.
2. Information theory developed in close parallel with control theory
(indeed, Wiener coined the term cybernetics to refer to both as they applied
to the analysis of animal and machine behavior). The quantification of infor¬
mation—in terms of the potential of a certain set of events to affect selec¬
tions—provided an important step in the rigorous analysis of what had
hitherto been known only informally to be a central concept in the under¬
standing of animal and machine behavior. Claude Shannon (along with the
people mentioned in connection with the study of control theory) showed
that there was a well-defined relation between certain general properties of a
physical system (notably its bandwidth and signal-to-noise ratio) and the
maximum rate at which information could be processed or transmitted
through it, regardless of what the information was about and what was to be
done with it. [Shannon and Weaver, 1949.]
3. Mathematical decision theory, the theory of competitive games, and
operations research were important extensions of mathematical analyses to
optimization problems in social, industrial, and management situations.
[See, for example, Simon, 1957.] Along with cybernetics, these develop¬
ments represented a new license to apply formal mathematical ideas to the
social and biological sciences. These developments also legitimized the no¬
tion of decision, just as work in communication and control theory had
legitimized the notion of information, by establishing it on a rigorous analy¬
tical foundation.
4. The development of automatic computing machinery eventually
brought all these ideas together by providing an object that was at once a
tool for carrying out analyses required by these complex studies and at the
same time, an object of theoretical study that involved elements of all these
new analytical methods. The appearance of both analogue and digital com¬
puters in the late 1940s was an event of inestimable importance in bringing
the new view of animal and mechanical commonality into sharp focus. From
the very beginning, computers influenced people’s thinking about the nature
of thought, intelligence, and adaptation and formed the key to the new
mechanistic conception of mind, based on a much richer metaphor (though
 66 ZENON W. PYLYSHYN

some, like me, feel that it is much more than a metaphor) than those that had
been prevalent in earlier attempts to unite the psychological, biological, and
physical modes of explanation. The historical and philosophical issues sur¬
rounding this discovery are beyond the scope of this essay; however, they
are discussed elsewhere in some detail. [Pylyshyn, 1970; Arden, 1980u; and
Pylyshyn, 1982.]
The overlap among these four areas, as measured in terms of the propor¬
tion of people who contributed to them, is remarkable. For example, the
earliest writings on the theory of interdependent competitive decisions
(theory of games), while dating back to the 1920s, became a central concern
in the 1940s, about the time when their author, John von Neumann, became
involved in designing the earliest stored-program electronic digital computer
[Pylyshyn, 1970] and in applying notions of computation to problems of
biology [Burks, 1970]. Claude Shannon also wrote some early papers on
computing and the use of symbolic logic in the design of switching circuits.
Norbert Wiener, who had been a student of Bertrand Russell, contributed to
technical developments in both information theory and the theory of control
systems. Even more important was Wiener spearheading the rapidly grow¬
ing view that, at a suitable level of formal abstraction, these ideas were also
relevant to understanding any complex system, including social and biolog¬
ical systems. This led to the view that there are important abstract principles
that are shared by any highly interacting sets of functional elements, regard¬
less of whether these are sets of cells in a liver or brain, sets of animals in
an ecological system or sets of humans in a social system or sets of elec¬
tronic components wired together to form a complex artifact. This view gave
rise to an area of study sometimes referred to as general systems theory.
Analog computation was also closely tied to both control theory and
information theory. The mathematical tools for analyzing analog systems—
calculus and differential equations—were the ones that were relevant to
continuously varying electronic systems such as those that occur in feed¬
back control loops. By contrast, the study of digital computers, decisions,
and the parts of information theory having to do with encoding were often
discrete. As it turned out, discrete systems were found to be mathematically
more tractable and also much more interesting in a number of respects. For
example, they appear to be far more general (inasmuch as they contain the
class of universal computers, or Turing machines) than continuous ones.
The common mathematical tool that was used in these cases was symbolic
logic.
Almost all early workers on computers and communication systems were
trained in symbolic logic. Indeed, even to this day, mathematical logic has
had a central role in developing ways of conceptualizing problems in the
entire spectrum of information-related disciplines. Despite some early criti¬
cisms of the predominance of logic in the study of artificial intelligence, logic
is becoming more central than ever—both for the analysis of programming
systems and as a formalism for representing knowledge and even programs
themselves.
INFORMATION SCIENCE VIEWED FROM COGNITIVE SCIENCE 67

The one area that has not been mentioned already but which played an
important, though initially highly speculative part in these developments,
concerns the relation of this network of ideas to understanding human
thought and the nature of intelligence. As early as 1950, attempts to apply
certain ideas developed in the study of foundations of mathematics (for
example, the notion of computational mechanism as exemplified by the
Turing machine) to understanding the nature of intelligence were published
by Turing, von Neumann, Shannon, and others. But it was with the avail¬
ability of actual computing devices in the early 1950s that this goal began to
be pursued in earnest.
There were two primary strands in this early work. One was directed at
the class of self-organizing or adaptive systems and emphasized learning,
statistical-pattern recognition, and the modeling of neural networks. This
strand frequently associated itself with the term cybernetics. The second
strand was more molar. It was directed at understanding problem-solving
and thought processes at what we would now call the cognitive or the knowl¬
edge level. These two strands were nearly equally represented at a historic
symposium held in 1958 at the National Physical Laboratory in England on
Mechanization of Thought Processes. In the last fifteen or so years, how¬
ever, work on the class of approaches that are concerned with self¬
organizing systems—especially those influenced by speculation about the
properties of randomly connected neural networks—has all but disap¬
peared. On the other hand, work on the second class of approaches has
grown rapidly and also changed in emphasis in order to reflect a new under¬
standing of the problem of intelligence. The first decade of this work
(roughly 1955-1965) was dominated by the search for general problem¬
solving techniques together with powerful methods called heuristics for con¬
trolling the breadth of trial-and-error search. This work is well represented
by the reports reproduced in a classical book of readings. [Feigenbaum and
Feldman, 1963.]
The second decade [described in Minsky, 1968; Feigenbaum, 1969] was,
by contrast, more concerned with problems of planning and designing good
forms of knowledge representation and the development of high-
performance (or expert) systems in narrower domains. The key terms
changed from learning, classification, and search to planning, knowledge
representation, and semantics. The most recent general picture of the state
of the field generally referred to as artificial intelligence can be found in
a chapter in the collective volume What Can Be Automated? [Arden, 1980a.]

SOME CHARACTERISTICS OF COGNITIVE SCIENCE

In this essay, I am concerned primarily with relations among disciplines and

with scientific and technical developments that have shaped and continue to
influence the development of the attempt to understand the nature of intelli-
68 ZENON W. PYLYSHYN

gence. It is not the only form in which the search for an understanding of
mind is proceeding. Nor, furthermore, is this effort being carried out in only
one academic discipline, for example, the one calling itself cognitive psy¬
chology or artificial intelligence. There are many distinct disciplines con¬
tributing to the effort. What characterizes this particular class of ap¬
proaches—which have recently come to be collectively (and somewhat
loosely) called cognitive science—is an allegiance to the network of ideas
that I roughly outlined and might be summarized as follows:

(a) The approach is formalist in spirit; that is, it attempts to formulate its
theories in terms of symbolic mechanisms of the sort that have
grown out of symbolic logic.
(b) The level of analysis, or the level at which the explanations or theo¬
ries are cast, is functional, and they are described in terms of their
information flow. What this means in particular is that this approach
factors out such questions as how biological material carries out the
function and how biochemical and biophysical laws operate to pro¬
duce the required information-processing function. This factoriza¬
tion is analogous to separating electrical-engineering considerations
from programming considerations in computer science. This does
not mean that such questions of biological realization are treated as
any less important, only that they represent a distinct and, to a large
extent, independent area of study.
(c) In addition to factoring out organic questions, this approach is also
characterized by the techniques it uses in formulating its theories
and exploring the consequences of its assumptions. The most widely
used (though not universal) technique is that of computer implemen¬
tation. Thus, an important methodological goal of cognitive science
is to specify symbolic mechanisms that can actually exhibit aspects
of the behavior being modeled. Adherence to such a sufficiency
criterion makes this approach in many respects like a design disci¬
pline rather than a natural science, at least insofar as the latter typi¬
cally attempts to uncover a small set of fundamental axioms or laws.
Its concern with synthesis makes it, to use Herbert Simon’s phrase,
one of the “sciences of the artificial.’’ [Simon, 1969.] By this, Simon
means that the search for ways in which a certain task can be carried
out within specified constraints—for example, of time and space—is
the mark of a design discipline. Of course, in the case of cognitive
science many of the constraints originate with empirical observa¬
tions. These observations may be not only of what tasks people can
carry out, but also of such things as the time it takes people to solve
certain kinds of problems, which ones they find most difficult, and so
on. Nonetheless, adopting the criterion of constructing a working
system as part of the explanatory process makes the cognitive
INFORMATION SCIENCE VIEWED FROM COGNITIVE SCIENCE 69

scientist’s task less like that of a physicist than a design engi¬

neer.
(d) The approach tends to emphasize a strategy sometimes referred to as
top-down analysis, where a premium is given to the task of under¬
standing how the general cognitive skill in question is possible (con¬
sonant with the constraint of mechanism) in contrast to the task of
accounting for empirical particulars. This difference in style con¬
trasts with the traditional approach in experimental psychology,
which emphasizes the observational fit of models. The contrast has
been carefully examined and discussed. [Newell, 1970; Pylyshyn,
1979; Sloman, 1978a.]
(e) The commitment to the informational level also contrasts the enter¬
prise with the phenomenological approach, where the existential no¬
tions of significance, meaningfulness, and experiential content are
given a central role in the analysis, and with behaviorism, which
attempts to analyze behavior without appealing to internal informa¬
tional states. The philosophical implications of this point have been
treated in recent publications. [Fodor, 1981; Haugeland, 1978;
Dennett, 1978a.]

These general characteristics of cognitive science are also shared in vari¬

ous degrees by other scientific disciplines. The formalist or symbol-
mechanistic character (a) is deeply entrenched in contemporary linguistics
(especially in generative grammar), decision theory, parts of anthropology
(e.g., the work of Levi-Strauss), and biology (e.g., the work of Burks). The
functionalist perspective (b) is now quite general in psychology and philoso¬
phy of mind as well as in technology, where it is referred to as the black-box
approach. Both (a) and (b) are fundamental to computer science as well as to
any science that concerns itself with such notions as the flow of information
or the distribution of control. Such ideas have thus affected everything from
engineering to management science—and even political science. [Deutsch,
1963.]
Criteria (c) and (d) are not quite so prevalent as the first two. For ex¬
ample, the desire to synthesize aspects of the phenomena being modeled, as
part of the attempt to understand them, is not widespread in the social
sciences. Notable exceptions are cognitive psychology and management
science (especially the branch of the latter called industrial dynamics, or the
even more ambitious world dynamics [Forrester, 1971]). The synthetic ap¬
proach is also not yet very common in biology. (See Marr’s critique of
theories in neurophysiology that fail to focus on the constructive computa¬
tional aspect of biological function [Marr, 1975].) Even modern linguistics,
which is, in many ways, a prototypical cognitive science, places little em¬
phasis on the capacity of the theoretical mechanism actually to generate
samples of performance. However, more recently examples of the contrary
trend can be found. [Marcus, 1980.]
 70 ZENON W. PYLYSHYN

The Representational Metapostulate

Although, as we have seen, there are a number of theoretical and meth¬

odological characteristics that pervade a variety of approaches to un¬
derstanding intelligence and human cognition, there is one overriding theme
that more than any other appears to me to characterize the field of cognitive
science. There are a number of ways of expressing this theme—for example,
as the attempt to view intelligent behavior as consisting of processing infor¬
mation or to view intelligence as the outcome of rule-governed activity. But
these characterizations express the same underlying idea: Computation, in¬
formation processing, and rule-governed behavior all depend on the exis¬
tence of physically instantiated codes or symbols that refer to or represent
things and properties outside the behaving system. In all these instances, the
behavior of the systems in question (be they minds, computers, or social
systems) is explained, not in terms of intrinsic properties of the system itself,
but in terms of rules and processes that operate on representations of ex¬
trinsic things. Cognition, in other words, is explained in terms of regularities
in semantically interpreted symbolic representations, just as the behavior of
a computer evaluating a mathematical function is explained in terms of its
having representations of mathematical expressions (such as numerals) and
the mathematical properties of the numbers these expressions represent.
This is also analogous to explaining economic activity by referring, not to
the categories of natural science (say, speaking of the physico-chemical
properties of money and goods), but to the conventional meaning or sym¬
bolic value of these objects (e.g., that they are taken to represent such
abstractions as legal tender or buying power). Although in both economics
and cognitive science, the meaning-bearing objects (or the instantiations of
the symbols) are physical, it is only by referring to their symbolic character
that we can explain observed regularities in the resulting behavior.

PATTERNS OF COMMUNICATION AMONG

CONTRIBUTING DISCIPLINES

It is difficult to judge the degree of current, as opposed to potential, interre¬

lations among the disciplines that comprise cognitive science or the broader
field known as information science. The extent of interest that each field
shows in the problems and findings of the other has varied from time to time
and from topic to topic. Generally, however, research cross-fertilization,
such as that measured by the frequency of citations, has tended to remain
largely within traditional disciplinary boundaries: Technical reports tend to
refer to other technical reports and empirical studies refer to other empirical
studies.
INFORMATION SCIENCE VIEWED FROM COGNITIVE SCIENCE 71

For example, I recently conducted a simple survey to determine the ex¬

tent of cross-disciplinary citation among several prominent journals in
artificial intelligence and cognitive science. For artificial intelligence, I used
what is probably the most representative source of papers in this field, the
International Joint Conference on Artificial Intelligence (IJCAI) held in 1977
in the United States and two years of the journal Artificial Intelligence. A
random sample of 528 references cited in that corpus revealed that at least
300 were clearly classifiable as references to other papers in artificial intelli¬
gence. Of these references, 56 were to other papers in the IJCAI proceed¬
ings; 29 to the AI journal; 16 to the Machine Intelligence series of books; and
157 were to such journals in computer science and engineering as the Jour¬
nal and the Communications of the Association for Computing Machinery
(ACM), the Institute of Electrical and Electronics Engineers (IEEE), and
newsletters of the various Special Interest Groups of ACM (SIGs). There
were only 35 citations of books or journals in psychology and 7 of journals in
linguistics. Several other applied areas were also only occasionally repre¬
sented, among them medicine, music, and management science. There were
about 160 books, dissertations, and other references that could not be ade¬
quately categorized.
The outstanding feature of these tentative figures is the relative rarity of
citations of psychological and linguistic papers (or reports in other empirical
sciences) and the rather high rate of citations of unpublished works in the
form of technical reports and dissertations. The mode of disseminating re¬
sults in unpublished technical reports is quite characteristic of the field of
artificial intelligence and contributes to the difficulty outsiders have in get¬
ting a good idea of the current state of research in that field. This style of
communication can be attributed to a number of factors. It may in part
reflect the fact that up to a few years ago, only a handful of major centers
were turning out the bulk of the most influential work. Indeed, until about
ten years ago, virtually all the work was being done by a small invisible
college of researchers who were known to one another and thus were able to
keep in contact adequately through the private distribution of technical re¬
ports. In addition to these historical reasons for this mode of communica¬
tion, however, there is a more substantive reason, one that is related to the
nature of the field. A great many of the products of research in artificial
intelligence consist of massive computer systems whose contribution cannot
be fully evaluated without a great deal of technical detail about their opera¬
tions and yet whose potential audience is too small to merit publication in
book form.
The relative paucity of citations of empirical research in other fields, such
as psychology, may in part reflect the fact that much current research in
artificial intelligence is concerned with developing computational techniques
for information processing (for dealing with problems of representation and
control). In addition, most people working in artificial intelligence come
72 ZENON W. PYLYSHYN

from computer science or engineering backgrounds (though this is chang¬

ing), and their familiarity and interest in psychological research literature
may be minimal. But there is an even deeper reason that may be relevant to
this tendency to ignore psychological research: Much of what is reported in
psychological literature may be irrelevant to even that part of the artificial-
intelligence community directly concerned with modeling aspects of human
cognition. It is irrelevant because most psychological experimentation is
motivated by the goal of clarifying issues raised in other experiments. It is,
as Newell has very well argued, largely paradigm-driven rather than
motivated by larger theoretical systems. [Newell, 1973/?.]
The interest shown in artificial-intelligence research by publications in
psychology journals does not appear to be any greater than what I found for
the converse. A sample of the last two years of the journals Cognitive Psy¬
chology, Cognition, and Memory and Cognition, totaling some 1200 refer¬
ences, revealed that the vast majority (nearly 1000) were to psychological
articles and books. Indeed, looking at the references to psychological papers
alone, I find that the vast majority of them (571 to 133) were what could be
called journals in conventional experimental psychology rather than those
that could be classified as cognitive or information-processing. Of all the
citations, only 50 were of studies of artificial intelligence or computer simu¬
lation, while another 70 were of linguistics articles and 16 of papers in
philosophy. The latter two categories were represented almost entirely in
the journal Cognition, which especially encourages such cross-disciplinary
work. It would appear from these figures that at present research in cogni¬
tive psychology sees itself as being more closely allied with general experi¬
mental psychology than with any other subfield of cognitive science.
Despite these rough statistical observations concerning the relative insu¬
larity of the two major contributing disciplines of cognitive science, there are
reasons to persist in the view that a major new cross-disciplinary area of
study is developing. For one thing, a somewhat better picture emerges if one
looks at the journal Cognitive Science, which is the official journal of the
Cognitive Science Society and is intended to foster cognitive science as an
interdisciplinary venture. Out of 331 citations of the last two years, 110 were
judged to be clearly psychological (31 books and 79 journal articles in psy¬
chology, with 29 of the articles in cognitive psychology); 55 were artificial
intelligence papers; 14 were articles in computer science journals; 50 were
citations of other articles in Cognitive Science itself; 40 were journal articles
in philosophy and logic; 26 were linguistics papers; 7 were neurophysiology
papers; and the remaining 36 citations were distributed among a variety of
areas, including library science, education, business, anthropology, and
book reviews.
Despite this apparent exception, I believe that the evidence (however
subjective and unreliable it may be) does strongly suggest that from a statis¬
tical point of view, cognitive psychology and computational studies of cogni¬
tion are in fact not yet experiencing a large degree of cross-fertilization. On
INFORMATION SCIENCE VIEWED FROM COGNITIVE SCIENCE 73

the other hand, these indicators pertain to the general population of publica¬
tions: They say nothing about how the most important and most influential
work is being done. Much of the most ground-breaking work is, in my view,
clearly cross-disciplinary and has been extremely influential in shaping peo¬
ple’s thinking about cognitive science. The reason that this does not show up
in statistical patterns of citations is that the number of such seminal works is
(still) small. As an example of this, the aggregate statistics presented do not
reveal that some important cross-disciplinary works' form the bulk of the
small set of cross-disciplinary citations in both psychology and artificial-
intelligence literature. Thus, while the total number of cross-disciplinary
citations may not be large in relation to the total number of publications, a
few of the clearly cross-disciplinary works have been extremely influential.
And that is perhaps to be expected in an area that purports to form a new
discipline as opposed to just being an amalgam of existing fields.
The same pattern emerges when one examines various other possible
indicators of the relation between cognitive science and its contributing
disciplines. For example, membership in various professional societies and
on editorial boards of journals indicates that most people who belong to
these official bodies belong to the traditional disciplines but that, as in the
case of the citation statistics examined, there are a small number of highly
influential cross-disciplinary members. Also paralleling the finding for cita¬
tions, both the editorial board of Cognitive Science and the governing body
of the Cognitive Science Society contain a broad distribution of members
from the contributing fields, especially psychology, computer science, and
linguistics.

THE CONVERGENCE TOWARD COGNITIVE SCIENCE

I have examined historical relations among fields that have recently come
together as cognitive science and have briefly looked at several indexes of
their de facto interrelation in the practice of research in this new field.
However, the practitioners of cognitive science have a general view of what
disciplines currently house the problems that are at the core of cognitive
science. These include such fields as anthropology, computer science, lin¬
guistics, neuroscience, philosophy, and psychology. They also include such
existing cross-disciplinary areas as cybernetics, neurolinguistics, neuro¬
psychology, computational psychology, computational linguistics, philoso¬
phy of psychology, philosophy of language, anthropological linguistics, cog¬
nitive anthropology, and brain evolution. A committee appointed by the

‘Prime examples are Newell and Simon’s Human Problem Solving [1972]; Winograd’s Under¬
standing Natural Language [1972al; Ernst and Newell s GPS: A Case Study in Genetality
[1969]; and such collections as Bobrow and Collins’s interdisciplinary reader [1975],
74 ZENON W. PYLYSHYN

Board of the Alfred P. Sloan Foundation (to advise the foundation on fruitful
areas of frontier research deserving special support) recently put together an
informal internal report on the state-of-the-art in cognitive science.2
The introductory section of this report mentions the preceding areas of
scholarship as being the contributing fields of cognitive science, while warn¬
ing that a general integration of the cognitive elements of these distinct
disciplines is not yet at hand: Cognitive science is still a goal rather than a
reality. However, it is a goal that an increasing number of scholars endorse
and has led in the last few years to the creation of several dozen centers of
excellence devoted to its pursuit.3 The portion of the Sloan Report included
in the appendix describes in general terms the current vision of the scope of
this discipline.
An inspection of the general picture just presented and in the appendix
reveals that cognitive science is not identical with what is commonly re¬
ferred to as information science, inasmuch as it does not mention many
fields that are concerned with storing, processing, and communicating infor¬
mation—such as work on information theory (including theories of optimal
encoding, cryptography, and information security); information retrieval
from large data banks; practical work in network development or library and
document-management systems; and many other such research areas that
are intimately involved with studying information. None of these studies can
be excluded as irrelevant to cognitive science, since any of them could
produce insights on the technical aspects of information use in cognitive
(i.e., thinking) systems. Nonetheless, these studies are not primarily di¬
rected at understanding the nature of human and nonhuman cognition or the
exercise of intelligence in the course of relating systems to their environ¬
ments. Thus, it is fair to view cognitive science as primarily an empirical
natural science concerned with a subset of the class of problems in informa¬
tion science, namely, those that bear on the general question of the exercise
of intelligence by systems that exist at least partially in an autonomous
relation with a natural and social environment to which they are actively
adapting.

2 Authors of this report included Michael Arbib, Carl Baker, Joan Bresnan, Roy D’Andrade,
Ronald Kaplan, Jay Keyser, George Miller, Donald Norman, Zenon Pylyshyn, Scott Soames,
Richard Thompson, Edward Walker, and Edgar Zurif.
3 Among the North American universities that have established centers for cognitive science are
Brown University, University of California at Berkeley, Carnegie-Mellon University, Univer¬
sity of Chicago, University of Colorado, Cornell Medical School, University of Illinois, Univer¬
sity of California at Irvine, University of Massachusetts, Massachusetts Institute of Technol¬
ogy, University of Michigan, University of Pennsylvania, University of Pittsburgh, Princeton
University, University of Rochester, Rutgers University, University of California at Santa
Barbara, University of California at San Diego, Stanford University, University of Texas,
University of Western Ontario, and Yale University. In addition, a large number of industrial
and other research institutions have also developed cognitive science laboratories—among
many others, Bell Laboratories, Bolt, Beranek and Newman, Exxon, Fairchild, IBM, Rand,
Schlumberger-Doll, Stanford Research Institute, and Xerox.
APPENDIX 75

Appendix
COGNITIVE SCIENCE, 1978
Report of the State of the Art Committee to the
Advisors of the Alfred P. Sloan Foundation,
October 1, 1978

Introduction

Cognitive science is the study of the principles by which intelligent entities

interact with their environments. By its very nature, this study transcends
disciplinary boundaries to include research by scholars working in such
disciplines as neuroscience, computer science, psychology, philosophy, lin¬
guistics, and anthropology. The familiar labels of these disciplines have
provided the road map adopted here to explore the state of research in
cognitive science.
It is, however, the richly articulated pattern of interconnection among
these subdomains which makes explicit the basis for the claim that an auton¬
omous science of cognition has arisen in the past decade. The disciplines
contributing to this science, and the major bonds among them, are sum¬
marized by the following diagram. Each of the six fields listed is connected
to the others by a network of interdisciplinary regimens, some of which
represent ancient topics of intellectual concern, others of which raise famil¬
iar and important issues which have not yet become the focus of major
scholarly effort.
Each of the component fields is tied to two or more of the others by a
network of interdisciplinary regimens. Each labeled link represents a well
defined area of inquiry which involves the intellectual and physical tools of
the two disciplines it ties together. Thus cybernetics uses the concepts de¬
veloped by computer scientists to model brain functions elucidated by
neuroscientists. Similarly, psycholinguistics joins two fields in its concern
for the mental apparatus and operations responsible for the acquisition of
language and its production and understanding, and the simulation of cogni¬
tive processes has combined computer science and psychology in order to
formulate explicit theories of thinking and problem solving. Other pairs
share similar sets of concerns.
Each of the 11 solid lines in the figure represents a well defined and
professionally established domain of interdisciplinary inquiry which may be
found within one or more traditional academic departments. Those four
links shown as dotted lines in the figure identify a set of issues, some already
familiar and important, which have not yet become the focus of formally
recognized scholarly effort.
76 ZENON W. PYLYSHYN

Philosophy

Figure 1. Subdomains of Cogni¬

tive Science: 1. Cybernetics. 2.
Neurolinguistics. 3. Neuropsy¬
chology. 4. Simulation of cognitive
processes. 5. Computational lin¬
guistics. 6. Psycholinguistics. 7.
Philosophy of psychology. 8. Phi¬
losophy of language. 9. Anthro¬
pological linguistics. 10. Cognitive
anthropology. 11. Evolution of
brain.

It is possible to consider linked sets of the six major disciplines taken

three or more at a time. The triad of philosophy, psychology, and linguistics,
for example, represents an old area of inquiry concerned with language and
the use of language in cognitive tasks. Each such group represents a valid
and increasingly active area of research whose practitioners have been
trained in two or more of the fields involved. A major concern of this paper is
the argument that the network of interacting disciplines shown here should
be considered as a whole under the name cognitive science. That whole
cannot yet be integrated successfully, but such an integration is the goal
toward which these related groupings are moving.
What the subdisciplines of cognitive science share, indeed, what has
brought the field into existence, is a common research objective: to discover
the representational and computational capacities of the mind and their
structural and functional representation in the brain. Cognitive science is
already being practiced by workers in the fields and subfields listed above.
These workers have accepted the challenge to specify adequate theoretical
descriptions of cognitive systems and to test empirically the predictions of
these theories. The survey which follows presents some of the examples of
their collaboration in research directed toward this objective and provides
ample demonstration of the practical necessity of such a coordinated
scientific attack. The questions now being addressed in the subfields of
cognitive science are fundamentally related; moreover, the theoretical and
methodological apparatus of one subfield is being increasingly applied, and
sometimes improved, to answer questions in another.
The reader will no doubt want to examine in detail the work within one or
more of the subdomains of cognitive science discussed in the survey. Here
we turn to a simple, but relatively well worked-out example to illustrate the
transdisciplinary nature of research in cognitive science.
APPENDIX 77

Consider the names we give to colors. There would seem to be little to

complicate the description of color-naming: there are the perceived colors
and words to name them. A child must learn to match the two, and an adult
uses this matching to refer reliably to colors. In order to appreciate the
serious scientific interest of the matching process, one must consider that
color-naming is a special case of more abstractly characterizable cognitive
processes.
In particular, how do the people who speak a language decide which
objects, events, properties, or relations they will name? Is the set of name-
ables a universal property specified by some innate characteristic of the
mind, or can people partition and name reality in any way convenient to
them? This is not an arbitrary question; it has been the subject of continuing
discussion for at least two centuries. Linguists recognized long ago that the
active lexicons of languages differ in reflection of the interests and concerns
of their speakers.
There are those relativists who feel that culture is free to form any verbal
concept and that it forms those which serve the habitual patterns of thought
of its members. Universalists, on the other hand, feel that while patterns of
usage may differ, the concepts underlying habitual usage are constrained in
every human culture by the nature of the physical world and by the innate
biological mechanisms that have evolved to cope with that world. Most
students of the question feel that some subfields of thought might be best
understood in relation to the habit patterns of a particular culture, while
biological principles that apply universally to human beings constrain
others. Thus, for example, languages differ in surface form sufficiently to
require translation but at the same time share the biologically determined
universal properties which make translation possible.
At first glance, the domain of color names might seem an ideal place to
demonstrate relativism. We know that the range of visible colors from red to
blue and from black to white is continuous. It is possible, then, that cultures
might divide and label the color continuum as they see fit. Different colors
could become salient or memorable in different cultures, for a variety of
reasons, and names would arise to label them. In fact, it is common knowl¬
edge that the speakers of these languages do draw different boundaries
between the color to which their names refer. According to the relativist
thesis, people who label the continuum of colors differently reflect different
cognitive capacities—learned or innate—for thinking about color. They
have different color concepts.
To demonstrate this thesis—that psychological differences are associated
with terminological conventions—early experiments tested the color mem¬
ory of English speakers. An extensively developed European and American
color technology made it possible to control and specify color samples pre¬
cisely, and these calibrated colors were presented to subjects who were
asked to remember them and, later on, to select them from a larger set ot
samples.
78 ZENON W. PYLYSHYN

The samples for which people have reliably communicable color names
(the sort of names which would enable a second person to select the color on
the basis of the name assigned) were found to be just those that could be
recognized and selected from others after a time interval. Colors without
simple names were both difficult to name or describe and difficult to remem¬
ber.
Such results initially were interpreted to mean that language does con¬
strain thought—in this case, that habits of color-naming determined the
accuracy of color memory. This interpretation proved to be inverted. Subse¬
quent studies demonstrated conclusively that particular colors are not dis¬
tinctive because a culture has given them names. Rather, the colors are
given names because they are distinctive.
The evidence comes from several sources. Psychological studies revealed
that the primary colors—those which English speakers call red, green, yel¬
low, blue, black, and white—have special status in color discrimination
tests. Neurophysiological research demonstrated that the visual system or¬
ganizes these primaries into opponent pairs—red/green, yellow/blue, and
black/white. In other words, the organization of color into primary pairs is
an innately given biological property of all humans with normal vision. Fur¬
thermore, anthropological studies of peoples whose languages contain fewer
color terms than English revealed that, in all cases, the “best instances” of
the colors named are close to the psychologically primary colors; that these
color terms are directly translatable to English. Finally, when memory tests
were conducted on people who spoke a language with only two basic color
terms, they remembered primary colors best, even though they had no
names for them.
Clearly, some aspects of color-naming are culturally determined: the
number of terms does vary. However, the use made of available color ter¬
minology is constrained by innate neurophysiological properties and their
relation to cognition rather than vice versa. In short, the conceptual capacity
underlying color-naming conforms more closely to the universalist than to
the relativist position.
Of itself, this conclusion is of no momentous significance: color terms
form only a fragment of the lexicon in any language. Our present interest,
however, is to demonstrate the kind of interdisciplinary research strategy
that is required to establish such a result.
Note first that the intellectual significance of the color-naming studies
derives from a general theoretical question about the nature of the human
mind, a question which is not properly restricted to any one of the disci¬
plines that contributed to the research just described. We assume that this
one research objective of cognitive science is to formulate abstract descrip¬
tions of the mental capacities manifested by the structure, content, and
function of various cognitive systems. We will refer to the subject matter of
this kind of objective as abstraction.
Second, note that the research on color-naming would have been impossi-
APPENDIX 79

ble without the extensive development of the science and technology of

color. This base not only provides the instrumentation needed to carry out
the research, but, more importantly, provides the intellectual resources
needed to specify alternative models of the color-naming process. Often a
cognitive function could be accomplished by a variety of realizable physical
systems. We will refer to the systematic exploration of these alternatives as
the objective of instantiation.
Third, note that not only color technology but psychological, linguistic,
and anthropological methods were required to discover how people in vari¬
ous cultures name and think about colors. The data obtained led to the
conclusion that one model of the color-naming system more plausibly
characterizes that found in humans. That is to say, devices could be de¬
signed and implemented either to allow an arbitrary partition of the color
space or to give priority to a particular set of colors. It is the latter which
more accurately parallels that in humans. We will refer to attempts to
characterize the mental processes underlying cognitive function in living
organisms as the objective of plausibility.
Finally, note that an important element of the explanation of color-naming
was a reasonably convincing neurophysiological account of the capacity.
For more complex cognitive systems, of course, the neurological basis will
be much more difficult to discover and characterize, but the scientific analy¬
sis of a cognitive system cannot be considered complete until its biological
foundations are understood. We will refer to the study of the neurological
mechanisms involved in cognition as the objective of realization.
These four objectives are shared by cognitive scientists, and they consti¬
tute the general problem areas around which work in the subdomains of the
field is oriented. In effect, the result of research directed toward achieving
the objectives just outlined is to place constraints on possible theories of
cognition and its biological representation; constraints based on the abstract
properties of cognitive capacities, the properties of devices for instantiating
those capacities, their plausible implementation in living organisms, and
their neurological realization.
The field of cognitive science has arisen because these objectives unite
the work of scholars in many academic and professional disciplines. The
emergence of the field as an empirical science has, however, preceded the
development of an institutional framework within which research on such
issues could be optimally pursued. Consequently, cognitive scientists all too
often find the transdisciplinary interaction that is so vital to the conduct of
their work difficult to attain or provide. The incorporation of directly rele¬
vant work from another field is sometimes frustrated by the unfortunate
impediments to interdisciplinary collaboration in research and training and
by the specialized terminology which has been developed within each disci¬
pline.
While it is, of course, impossible to guarantee the success of an intellec¬
tual undertaking, cognitive science clearly is a field at the brink of rapid, and
80 ZENON W. PYLYSHYN

perhaps significant, change. As the field matures, there is little doubt that
great progress will be made in many of the aspects of cognition currently
under investigation. Perhaps the single most important contribution which
can be made to further this effort would be to introduce centers of research
in which active scholars and students might come together for extended
collaboration in research and training at all levels. This report is intended to
demonstrate the value of, and immediate need for, such a contribution to the
field.
 COGNITIVE SCIENCE
The View from Brain Theory

Michael A. Arbib

Pylyshyn offers a historical perspective on cognitive science, which sees its

roots in mathematical logic, with artificial intelligence at the core of recent
developments. In the first section of this comment, I offer a somewhat
different history, one that places more emphasis on brain research, and
which suggests that cognitive science is Cybernetics Redux.

CYBERNETICS AND THE ROOTS OF COGNITIVE SCIENCE

Technology has always played a crucial role in attempts to understand the

human mind and body; for example, the study of the steam engine has
contributed concepts to the study of metabolism, and electricity has been
part of the study of the brain at least since sometime before 1791, when
Galvani touched a frog’s leg to an iron railing. In 1750, Offray de la Mettrie
published L’homme machine and suggested that such automata as the
mechanical duck and flute player of Vaucanson indicated the possibility of
one day building a mechanical man who could talk. [Offray de la Mettrie,
1750.] The automata of those days were unable to adapt to changing circum¬
stances, but in the following century, machines were built that could auto¬
matically counter disturbances to restore their desired performance. Per¬
haps the best known example of this is Watt’s governor for the steam engine.
This development led to James Clerk Maxwell’s paper, “On Governors,”
which laid the basis for both the theory of negative feedback and the study of
system stability. [Maxwell, 1868.] At the same time, Claude Bernard drew
attention to homeostasis, observing that physiological processes often form
circular chains of cause and effect that could counteract disturbances in such

This paper refers to research supported in part by NIH grant NS 14971 and by A. P. Sloan
Foundation grant 80-6-13 to the University of Massachusetts. Portions of this paper appeared in
somewhat different form in my article in the Handbook of Physiology—The Nervous System II.
Motor Control [1981],

81
82 MICHAEL A. ARBIB

variables as body temperature, blood pressure, and glucose level in the

blood. [Bernard, 1878-1879.]
The year 1943 was the key year for bringing together the notions of
control mechanism and intelligent automata. Kenneth Craik published his
seminal essay The Nature of Explanation. [Craik, 1943.] Here, the nervous
system was viewed “as a calculating machine capable of modeling or paral¬
leling external events,” suggesting that the process of paralleling is the basic
feature of thought and explanation. In the same year, Rosenblueth, Wiener,
and Bigelow published “Behavior, Purpose and Teleology.” [Rosenblueth,
Wiener, and Bigelow, 1943.] Engineers had noted that if the feedback used
in controlling the rudder of a ship were, for instance, too brusque, the rudder
would overshoot, compensatory feedback would yield a larger overshoot in
the opposite direction, and so on and so on as the system wildly oscillated.
Wiener and Bigelow asked Rosenblueth if there were any corresponding
pathological conditions in humans and were given the example of intention
tremor associated with an injured cerebellum. This evidence for feedback
within the human nervous system led the three scientists to urge that
neurophysiology move beyond the Sherringtonian view of the central ner¬
vous system as a reflex device adjusting itself in response to sensory inputs.
Rather, setting reference values for feedback systems could provide the
basis for analyzing the brain as a purposive system explicable only in terms
of circular processes, that is, from nervous system to muscles to the external
world and back again via receptors.
The year 1943 also saw the publication of “A Logical Calculus of the
Ideas Immanent in Nervous Activity,” in which Warren McCulloch and
Walter Pitts offered their formal model of the neuron as a threshold logic
unit. [McCulloch and Pitts, 1943.] They were building on the neuron doc¬
trine of Ramon y Cajal and the excitatory and inhibitory synapses of Sher¬
rington. They used notations from the mathematical logic of Whitehead,
Russell, and Carnap, but a major stimulus for their work was the Turing
machine, a device that could read, write, and move on an indefinitely ex¬
tendable tape, each square of which bore a symbol from some finite al¬
phabet. Alan Turing had made plausible the claim that any effectively
definable computation, that is, anything that a human could do in the way of
symbolic manipulation by following a finite and completely explicit set of
rules, could be carried out by such a machine equipped with a suitable
program. [Turing, 1936.] What McCulloch and Pitts demonstrated was that
each such program could be implemented using a finite network (with loops)
of their formal neurons. Thus, as electronic computers were built toward the
end of World War II, it was understood that whatever they could do could be
done by a network of neurons.
These, then, were some of the strands that were gathered in Wiener’s
book Cybernetics, or Control and Communication in the Animal and the
Machine, published in 1948, and in the Josiah Macy Jr. Foundation confer¬
ences, which, from 1949 on, were referred to as Cybernetics: Circular
Causal and Feedback Mechanisms in Biological and Social Systems. It is
COGNITIVE SCIENCE: THE VIEW FROM BRAIN THEORY 83

beyond the scope of the present commentary to trace the further evolution
of work under the banner of cybernetics. Rather, let us simply note that, as
the field developed in the fifties, it began to fragment. Much work in cyber¬
netics now deals with control problems in diverse fields of engineering,
economics, and other social sciences, whereas the broad field of computer
science has become a discipline in its own right. Here, we briefly cite five
subdisciplines that have crystallized from the earlier concern with the inte¬
grated study of mind, brain, and machine.

1. Biological Control Theory. The techniques of control theory, espe¬

cially the use of linear approximations, feedback, and stability analy¬
sis, are widely applied to the analysis of such diverse physiological
systems as the stretch reflex, thermoregulation, and the control of the
pupil.
2. Neural Modeling. The Hodgkin-Huxley analysis of the action po¬
tential, Rail’s models of dendritic function, analysis of lateral inhibi¬
tion in the retina, and the analysis of rhythm-generating networks are
examples of successful mathematical studies of single neurons, or of
small or highly regular networks of neurons, which have developed in
fruitful interaction with microelectrode studies.
3. Artificial Intelligence. This is a branch of computer science devoted
to studying techniques for constructing programs enabling computers
to exhibit aspects of intelligent behavior. Among such intelligent be¬
havior is playing checkers, solving logical puzzles, or understanding
restricted portions of a natural language such as English. Although
some practitioners of artificial intelligence look solely for contribu¬
tions to technology, there are many who see their field as intimately
related with cognitive psychology.
4. Cognitive Psychology. The concepts of cybernetics gave rise to a
new form of cognitive psychology that sought to explain human per¬
ception and problem-solving not in neural terms but rather at some
intermediate level of information-processing constructs. Recent years
have seen strong interaction between artificial intelligence and cogni¬
tive psychology.
5. Brain Theory. Because cybernetics extends far beyond the analysis
of brain and machine, the term brain theory has been introduced to
denote an approach to brain study that seeks to bridge the gap be¬
tween studies of behavior and overall function (artificial intelligence
and cognitive psychology) and the study of physiologically and ana¬
tomically well-defined neural nets (biological control theory and
neural modeling).

In the 1970s, a new grouping took place, which brought together research¬
ers in artificial intelligence and cognitive psychology with those linguists and
philosophers of mind who emphasize symbol-processing. The resulting field
84 MICHAEL A. ARBIB

is more a loose federation than an integrated discipline, and it is this cogni¬

tive science that Pylyshyn describes in his article. Few practitioners are
aware of the roots of cognitive science in cybernetics, and most believe that
their ignorance of brain research is a virtue and human intelligence can be
studied as symbol-manipulation without concern for its embodiment. (John
Searle offers a recent critique of this viewpoint. [Searle, 1982«.]) At the
same time, all too many neuroscientists see workers in artificial intelligence
as simply playing with toys. In the hope of better bridging neuroscience and
cognitive science, I offer an integrated perspective on past and potential
contributions of brain theory to the analysis of the role of visual information
in the neural control of movement. Due to lack of space, I shall not discuss
efforts to bring neurologists, psycholinguists, brain theorists, and artificial
intelligence workers together to study language, but simply refer to a recent
collection of papers [Arbib, Caplan, and Marshall, 1982] and a dissertation
[Gigley, 1982],

MAPS AS CONTROL SURFACES

A notable characteristic of the brain is the orderly mapping from one neural
layer to another, be it the retinotopic mapping from the retina to the many
visual systems or the somatotopic mapping of the motor cortex to the mus¬
culature. I briefly look at one hypothesis of how such a map acts as a control
surface such that the spatio-temporal patterns in the map provide input to
some control system in the brain. Pitts and McCulloch offered a distributed-
processing model of the reflex arc that extended from the eyes through the
superior colliculus to the oculomotor nuclei, thereby controlling muscles
that direct the gaze so as to bring the fixation point to the center of gravity of
distribution of the visual input’s brightness. [Pitts and McCulloch, 1947.]
Pitts and McCulloch noted that excitation at a point on the left colliculus
corresponds to excitation from the right half of the visual field and so should
induce movement of the eye to the right; gaze is centered when excitation
from the left is exactly balanced by excitation from the right. Their model is
so arranged, for example, that each motoneuron controlling muscle fibers in
the muscles that contract to move the eyeballs to the right should receive
excitation summing the level of activity in a thin transverse strip of the left
colliculus. This process provides all the excitation to muscles turning the eye
to the right. Reciprocal inhibition by axonal collaterals from the nuclei of the
antagonist eye muscles, which are excited similarly by the other colliculus,
performs subtraction. The quasicenter of gravity’s vertical coordinate is
computed similarly. Eye movement ceases when and only when the fixation
point is the center of gravity.
This scheme shows how to design a retinotopically organized network in
which there is no “executive neuron” that decrees which way the overall
system behaves; rather, the dynamics of the effectors, with assistance from
neuronal interactions, extract the output trajectory from a population of
COGNITIVE SCIENCE: THE VIEW FROM BRAIN THEORY 85

neurons, none of which has more than local information as to how the
system should behave. In other words, the Pitts and McCulloch model of the
superior colliculus shows how the organism can be committed to an overall
action by a population of neurons, none of which has global information
about which action is appropriate.
I have argued that the study of such cooperative computation in so-
matotopically organized networks provides a central paradigm in brain
theory. [Arbib, 1972.] The study of visuomotor coordination in frog and toad
provides an area (by no means the only one) rich in experiments that may
contribute to our understanding of the neural underpinnings of perceptual
structures and distributed motor control. A number of experiments on prey
catching and predator avoidance as well as on the modification of behavior
by the presence of barriers have led to models of prey selection and prey-
enemy pattern-recognition. [Ewert, 1980; Arbib, 1982.] These experiments
provide support for the following general conclusions:

1. Analysis of vision must not only study local features and responses to
point stimuli, but also develop techniques to analyze structured visual
stimuli. Visual processing can be viewed in terms of cooperative computa¬
tion in somatotopically-organized networks, with overall patterns of activity
being generated in neural structures wherein no single neuron has global
information as to what course is appropriate. Here, there is a rich area of
relevant artificial intelligence totally ignored in Pylyshyn’s review, namely,
machine vision. For example, Dana Ballard and Christopher Brown ap¬
proach machine vision in a way that is increasingly based on a brainlike style
of cooperative computation. [Ballard and Brown, 1982.] As discussed in my
“Cybernetics: The View from Brain Theory” in this volume, there is now
developing a general theory of competition and cooperation in neural nets,
with applications to prey selection, mode selection, and stereopsis, and with
similarities to relaxation techniques used in artificial intelligence. An impor¬
tant bridge between machine vision and human vision is offered by Marr.
[Marr, 1982.]
2. It is useful to view a neural map as a control surface providing input
to some control system in the brain. In addition to a knowledge of competi¬
tion and cooperation within the neural nets that constitute each controller,
there must be an understanding of the cooperative computation of control¬
lers. This cooperation may entail coordination turning off all but one of the
controllers, or it may involve a rich interplay between subsystems, each
modulating the other.

THE PERCEPTUAL AND MOTOR SCHEMAS

Pylyshyn has correctly stressed that cognitive science currently tends to

formulate its theories in terms of symbol-manipulation and that theories of
biological organization are ignored. (Pylyshyn is more diplomatic. His exact
86 MICHAEL A. ARBIB

words are: “questions of biological realization are [not] treated as any less
important, only . . . they represent a distinct and, to a large extent, indepen¬
dent area of study.”) In fact, theories apply blinkers to the consideration of
relevant facts, and a cognitive scientist who restricts himself or herself to a
symbol-manipulation theory will never examine the data on brain function
that could show the inadequacy of this approach. Pylyshyn emphasizes the
top-down analysis of much cognitive science. However, he has elsewhere
given a more subtle analysis of the way in which the data the scientist
chooses to examine determine the appropriate grain of the functional ar¬
chitecture of the theory. [Pylyshyn, 1980.] Just consider the very different
demands of linguistic models that address the following three kinds of data:
judgments of sentence grammaticality, response time and eye-movement
data on sentence comprehension, and aphasiological data on the effect of
brain damage in language performance.
In the rest of this comment, I want to show how an analysis of perception
and motor control in the style of top-down cognitive psychology can fruit¬
fully interact with bottom-up analysis enriched by the analog (rather than
symbolic) control theory of cybernetics. The approach grows out of my
concern with brain theory, and is developed in more detail in my contribu¬
tion to the Handbook of Physiology. [Arbib, 1981.] However, the neural
aspects of the analysis will be played down in the present discussion.

Perceptual Schemas and the Action-Perception Cycle

Human behavior is determined by a far greater knowledge of the environ¬

ment than afforded by the current stimulation of the retina, especially be¬
cause so little of that stimulation is foveal. Our actions are addressed not
only to interacting with the environment in some instrumental way, but also
to updating our “internal model of the world.” [Gregory, 1969.] In a new
situation, we can recognize familiar things in new relations and use our
knowledge of those things and our perception of the relations to guide our
behavior in that situation. It thus seems reasonable to posit that the internal
model of the world must be built of units, each of which roughly corresponds
to a domain of interaction, which may be an object in the usual sense, an
attention-riveting detail of an object, or some domain of social interaction.
The notion of schema has been widely used in neurology [Frederiks,
1969]; psychology [Bartlett, 1932; Piaget, 1971]; artificial intelligence, under
such names as frames [Minsky, 1975]; and in the study of motor skills.
Where much of cognitive science talks of knowledge in only some ab¬
stract realm of symbol-manipulation or problem-solving, I here wish to
stress knowledge representations that subserve perception, embedded
within the organism’s ongoing interaction with its environment. As the or¬
ganism moves in a complex environment, making, executing, and updating
plans, it must stay tuned to its spatial relations with its immediate environ-
COGNITIVE SCIENCE: THE VIEW FROM BRAIN THEORY 87

ment, anticipating objects before they come into view. The information
gathered during ego motion must be systematically integrated into the organ¬
ism’s internal model of the world: Information picked up modifies the per-
ceiver’s anticipations of certain kinds of information that, thus modified,
direct further exploration and prepare the perceiver for more information.
Such considerations will, 1 predict, increasingly engage the attention of cog¬
nitive scientists as robotics becomes an increasingly active area of study
within artificial intelligence. The problem of controlling robot arms and in¬
tegrating visual and tactile information will require increasing attention to
control theory, cooperative computation, and sensorimotor integration. This
will combat the overemphasis on symbolic processes provided by
Pylyshyn’s paper.
The intelligent organism does not so much respond to stimuli as select
information that helps it achieve current goals, although a well-designed or
evolved system certainly needs to take appropriate account of unexpected
changes in its environment. Planning is the process whereby the system
combines an array of relevant knowledge to determine a course of action
suited to current goals. In its fullest subtlety, planning involves the refine¬
ment of knowledge structures and goal structures as well as action per se.
Novel inputs (e.g., encountering an unexpected obstacle) can alter the
elaboration of high-level structures into lower-level tests and actions that, in
turn, call on the interaction of motor and sensory systems. Scientists seek to
study programs that are part of the internal state of the system and can
flexibly guide ongoing action in terms of internal goals or drives and external
circumstances. My thesis is that perception of an object (activating appropri¬
ate perceptual schemas) involves gaining access to routines for interaction
with the object (motor schemas) but does not necessarily involve executing
even one of these routines. Although an animal may perceive many aspects
of its environment, only a few of these can at any time become primary
loci of interaction. Therefore, perception activates (i.e., defines a search
space, draws a map), and planning concentrates (lays out the route to be fol¬
lowed).
I use the term perceptual schema to denote the process whereby the
system determines whether a given domain of interaction is present in
the environment. The state of activation of the schema then determines the
credibility of the hypothesis that what the schema represents is, indeed,
present, whereas other schema parameters represent such properties as
size, location, and motion of the perceived object. Consider a schema that
represents, say, a chair; also consider an environment that has two chairs in
plain view. It is clear that two copies of the chair schema (or at least two
separate sets of chair-schema parameters) are required to represent the two
chairs. These two copies are separate instantiations of the same schema,
each with its own set of parameter values. The internal representation of the
environment may thus be viewed as an assemblage of spatially tagged,
parametrized schema-instantiations.
88 MICHAEL A. ARBIB

Motor Schemas

I have used schema to indicate the type of unit from which the internal
representation of the environment can be built. Programs for motor control
may themselves be seen as assembled from suitable units, which are referred
to as motor schemas. These motor schemas are related to synergies (in the
sense of the Russian school founded by Nikolai Bernstein [Bernstein, 1967]).
These schemas may exhibit short-term memory. To control properly the
motion of an object (the controlled system), the controller must clearly know
such relevant parameters of the object as its mass and moment of inertia.
However, in the real world the exact values of these parameters are seldom
available and may actually change over time. (Compare short-term loading
effects on muscles with longer term aging effects and weight changes.) To
adapt to such changes, the feedback loop must be augmented by an
identification algorithm. The job of this algorithm is to identify more accu¬
rately the parameters of the controlled system. To do this, it continually
monitors the behavior of the controlled system and compares it with the
output that would be expected on the basis of the current estimated parame¬
ters. Any discrepancies in the output can be used to obtain more accurate
estimates of the parameters that define the controlled system. These updated
parameters can then be supplied to the controller as the basis for its state
estimation and control computations.
Note well that the identification algorithm can do its job only if the con¬
troller is of the right general class. It is unlikely that a controller adapted for
guiding the arm during ball-catching would be able, simply as a result of
parameter adjustment, to control properly the legs in performing a waltz.
Thus, the adaptive controller (controller plus identification procedure) is not
to be thought of as a model of the brain; rather, each such control system is a
model of a single motor schema that can be activated when appropriate.
This framework for analyzing visually guided behavior of a complex or¬
ganism is based on four general premises:

1. The action-perception cycle: As the organism moves—making, ex¬

ecuting, and updating plans—it must maintain an up-to-date repre¬
sentation of its spatial relations with its environment.
2. The model of the environment is an active, information-seeking
process composed of an assemblage of perceptual schemas, each in¬
stantiation of which represents a distinct domain of interaction with
such relevant properties as size and motion represented by the cur¬
rent values of parameters of the schema.
3. Activation of perceptual schemas provides access to related motor
schemas but does not necessarily entail execution of these schemas.
Planning is required to determine the actual course of action. The
plan is updated as action affords perceptual updating of the internal
model.
COGNITIVE SCIENCE: THE VIEW FROM BRAIN THEORY 89

4. The plan of action is to be thought of as a coordinated control pro¬

gram composed of motor schemas, each viewed as an adaptive con¬
troller that uses an identification procedure to update its representa¬
tion of the object being controlled. Thus, the identification procedure
can be viewed as a perceptual schema embedded within a motor
schema.

COORDINATED CONTROL PROGRAMS

Karl Lashley laid many of the foundations for neuropsychology, that most
cognitive of the neurosciences. In an important critique of stimulus/response
theory, he raised questions about serial order in behavior that are answered
at the conceptual level as soon as one thinks of the brain’s computations not
in terms of stimulus/response couplings or chains of associations but, rather,
in terms of coordinated control programs. [Lashley, 1951.] Although our
knowledge of computer programs removes the conceptual problem of serial
order, the question of how such control strategies can be neurally imple¬
mented is only beginning to be answered. Much of the neurophysiological
analysis of movement has focused on spinal mechanisms (especially feed¬
back mechanisms in posture and locomotion) and on higher level single-cell
correlates of stimulus or response. Future research clearly must aim to
better analyze the distribution of planning operations within cortical struc¬
tures and understand the signal flow this planning must impose on the cere¬
bellum and other regions that modulate this planning.
Even though the neural mechanisms for the planned coordinated control
of motor schemas seem to be beyond the range of current experimental
investigation, I suggest that artificial-intelligence approaches to planning
may provide a framework for the development of such investigations in the
future.
A pattern of action may be quite complex, with the actions intertwined
and overlapping. Simultaneous actions must be coordinated, and successive
actions must be smoothly phased, one into the next. In this section, I discuss
the concept of a coordinated control program as the type of structure that
orchestrates the interwoven activation of motor schemas controlling differ¬
ent actions.
Biological control theory usually studies neural circuitry specialized for
the control of a specific function, be it the stretch reflex or the vestibulo-
ocular reflex. Yet, most behavior involves complex sequences of coor¬
dinated activity of a number of control systems. Thus, I explore the notion
of a coordinated control program as a combination of control theory and the
computer scientist’s notion of a program suited to analyzing the control of
movement. Control theorists use a block diagram to represent a system.
Each box represents a subsystem that is continually active, whereas the
lines linking the boxes illustrate the transfer of data, showing how the output
90 MICHAEL A. ARBIB

of one system helps determine input to another. By contrast, the boxes in

flow diagrams used by computer scientists represent not subsystems but
patterns of activation of subsystems. The computer has various subsystems,
such as memory registers, arithmetic units, and test units. At any time in a
computation, certain data are stored in these subsystems, and one box in the
flow diagram is activated in the sense that it is used by the computer to
determine what tests and operations are to be carried out by the subsystems
and how data are to be transferred among them. The lines of the flow dia¬
gram then specify how activation is to be transferred from one instruction to
another.
In a coordinated control program, control schemas are so scheduled that
simultaneous actions are coordinated and successive actions are smoothly
phased one into the next. Although certain basic programs are hard-wired
into the organism, most programs are generated as the result of an explicit
planning process. A hypothetical program for reaching to a target showed
that perceptual schemas need not be defined as a separately specified part of
a coordinated control program but may, instead, enter automatically as the
identification algorithms that are required to define the appropriate values of
the parameters in the motor schemas entering into the program. [Arbib,
1981.]
Concepts from computer science and artificial intelligence have been ex¬
posed to little in the way of neuroscientific experiment but may serve to
stimulate new models of perceptual structures and their role in planning and
controlling movement.

CONCLUSION

I have examined the role of visual information in the control of movement

from the perspective of brain theory, seeking to bridge the gap between the
cognitive-science topics addressed in Pylyshyn’s paper (artificial intelligence
and cognitive psychology) and the study of physiologically and anatomically
well-defined neural nets (biological control theory and neural modeling).
Complexity seems to demand intermediate levels of analysis to mediate
between a top-down analysis of the flexibility of an animal’s behavior in a
richly structured environment and the data of neuroscience. The environ¬
ment can be represented in terms of an assemblage of instantiations of
perceptual schemas, each instantiation representing relevant parameters of
some particular object or domain of interaction relevant to the organism.
Artificial intelligence provides techniques for analyzing processes whereby a
plan of action is created on the basis of the organism’s goals and current
perceptions. Cybernetic analyses and Russian studies of skilled movement
give the notion of motor schemas as the units from which each plan of action
is constituted; the concept of a coordinated control program suggests how
motor schemas might be interwoven. Control-theoretic studies of feedfor-
COGNITIVE SCIENCE: THE VIEW FROM BRAIN THEORY 91

ward and identification algorithms are useful in describing the structure of

motor schemas at a level that can be connected with neurophysiological
investigation. The growing refinement of experiments in the motor-skills
literature, especially when coupled with electromyogram recording, begins
to let us essay functional accounts of coordinated control programs for a
number of behaviors that involve the integration of several motor schemas.
However, the neural implementation of such programs, as distinct from the
motor schemas themselves, remains elusive.
We thus see that when we turn from abstract problem-solving to the study
of perception and movement, we can make use of a spectrum of concepts
that ranges from those firmly rooted in the findings of the neuroscientist’s
laboratory and the control theory descended from Wiener’s cybernetics to
those that have proved their efficacy in the design of robots. I have also
observed that there are whole areas of artificial intelligence, such as machine
vision, that are absent from Pylyshyn’s picture of cognitive science. More
centrally to the present paper, I have suggested that cognitive scientists
(those working in vision being the most striking exception) are “blinkered”
in their choice of data and thus dismiss neural data as irrelevant, holding that
a top-down analysis is necessarily sufficient. Pylyshyn certainly reinforces
this perception when he surveys 331 citations in the journal Cognitive Sci¬
ence and finds that only 7 were to sources in neuroscience. Another com¬
mentator might note that anthropology, too, is still too limited in its impact
on cognitive science. The aim of this commentary has been to encourage the
cognitive scientist to learn from both brain theorist and experimentalist in
developing a theory of the human mind that can benefit from the rich body of
data and models on animal behavior, comparative neuroanatomy, neuro¬
ethology, and neurophysiology, and the clinical data of the neurologist.
Two last comments: In the diagram provided in the appendix to Pyly¬
shyn’s paper, one of the missing links is that between philosophy and
neuroscience. Surely this is a mistake. Discussion of the brain-mind problem
antedates the current fashion for cognitive science by centuries and is still
going strong, as witnessed by the active interest generated by the (in my
opinion, mistaken) dualist position of Karl Popper and John Eccles in their
volume on The Self and Its Brain. [Popper and Eccles, 1977.] I would also
draw the reader’s attention to Robert Young’s Mind, Brain, and Adaptation
in the Nineteenth Century, which should engender humility in us all by
showing how thoughtfully neurologists of the last century considered prob¬
lems at the center of cognitive science today. [Young, 1970.]
 COGNITIVE NEUROSCIENCE
More Plain Talk

Michael S. Gazzaniga

Zenon Pylyshyn has laid out with striking clarity what cognitive scientists
do, which is, mainly, test the design of information systems that model
aspects of a behavior. That is not what cognitive neuroscientists do, and my
bet is that the approach he describes will be inefficient in designating cogni¬
tive systems that do accurately describe how human minds work. That is the
task cognitive neuroscientists have set for themselves, and their objective is
to take up the problem experimental psychologists left for neuroscientists
but that neuroscientists did not pick up. This is because the current view in
neuroscience is that the cellular study of brain function will yield more
substantial insights into, well, into something, and the question is what is the
something? Accepting completely Pylyshyn’s case for the levels-of-analysis
approach, cognitive neuroscientists believe that the cellular approach of
their colleagues in basic neuroscience will tell us about cells. That is an
admirable task, but one that leaves begging the task of understanding human
cognition.
In order to compare the approaches, it is helpful to take the example of
the problem of memory. The basic neuroscientists, taking the lead from
work on brain-damaged humans, have decided to focus a large part of their
energies on a brain structure called the hippocampus. This structure, which
is implicated in the human memory process, can be removed from experi¬
mental animals, placed in tissue culture, and the metabolic and electrophysio-
logical properties of its cells studied. A series of elegant reports has demon¬
strated that changes in cellular activity occur to repetitive stimulation that
are long lasting, indeed, seemingly permanent. The biochemical machinery
responsible for this response seems to be in hand and investigations are well
on their way to determining the site of morphological changes on the
synapse that must occur to explain the long-lasting effect. Nice, perhaps
even brilliant, but how such changes connect with how a person remembers
a telephone number when it is breezed by at a cocktail party is not even
attempted.
At the level of description at which cognitive scientists work, the task is
to describe how artifacts and/or humans process information. Over the
93
94 MICHAEL S. GAZZANICA

years, cognitive scientists have established limits on the amount and kind of
information that can be apprehended, on strategies that facilitate the acquisi¬
tion of information, and on the possible different kinds of memory the cogni¬
tive system handles. These elegant studies generate data on response char¬
acteristics of these human and nonhuman information systems, and theories
are constructed about how the system must be organized. What is left open,
of course, is whether or not the human brain is organized the way the
cognitive scientists think it must be, just as Pylyshyn plainly says.
Alas, the cognitive neuroscientists’ approach! The subject for experimen¬
tal observation is the human with brain disease and/or focal lesion that can
be accurately localized and described. The measures are on this human’s
ability to perform memory tasks of a wide variety. The results allow one to
construct theories about how the memory system is organized by consider¬
ing how systems make errors that are reliably generated by specific patho¬
logical states. The approach puts constraints on the theories of the cognitive
scientists and the neuroscientists. At the present time, the cognitive neuro¬
scientist is rewriting the backdrop for researchers in both areas. The neuro¬
scientist is being redirected to other brain areas as well as facing the reality
that more than one cellular process may be responsible for memory. Cogni¬
tive scientists are having to adjust their theories to the fact that there is not
one cognitive system that is in charge of all information-processing. Recent
developments would suggest that the cognitive system has a modular-type
organization and that these modules may have specialized features.
Of course, in reality, all of the approaches seem terribly exciting, and any
contemporary scientist must become reasonably conversant in the three
fields—cognitive science, basic neuroscience, and cognitive neuroscience—
but each of us, happily, has our own preference.
 COGNITIVE SCIENCE
A New Patch in the Quilt?

Saul Gorn

Pylyshyn first presents a historical introduction discussing a general ideolog¬

ical drift from material and power concepts to the study of their interactive
relations and our symbolization of them. He points out that the resulting
paradigms were cybernetic and, finally, also included the study of communi¬
cation and control of symbol systems (analog and digital computers). These
“influenced . . . thinking about the nature of thought, intelligence, and adap¬
tation,” to quote Pylyshyn. He distinguishes a cybernetics of continuous
processes from one of the discrete, much as Moses, in his comments on my
paper in this volume, distinguishes the paradigms of system theory from
those of information science. Where the first set of paradigms studied self¬
organizing and adaptive systems, the second addressed itself to the simula¬
tion of thought processes involved in problem-solving, that is, to artificial
intelligence.
Pylyshyn’s second section discusses what he feels are the more or less
common attitudes of the group of disciplines that “attempt to understand the
nature of intelligence” (philosophy, linguistics, anthropology, neuroscience,
computer science, and psychology—as he points out in his last section and
as appears in the appendix from the report to the advisors of the Sloan
Foundation). He feels that these shared attitudes are (1) formalist, (2) func¬
tional, (3) in the habit of using simulation techniques to “specify symbolic
mechanisms that can actually exhibit aspects of the behavior being mod¬
eled,” (4) addicted to top-down analysis, and hence (5) in sharp contrast
with phenomenological and behavioristic approaches. He finds aspect 3 the
most characteristic, calling it, I believe, “the representational metapos¬
tulate.”
Pylyshyn’s third section discusses the intercommunication of the parent
discipline and the new area by examining citation statistics. He finds that
communication is heaviest within the new area and fairly slim across the
boundaries.
Finally, Pylyshyn quotes the Sloan report as saying that the integration
into one discipline is not yet at hand. In any event, he believes such an
95
96 SAUL CORN

integrated discipline would not be identical with information science, be¬

cause it is not involved with storing, processing, and communicating infor¬
mation or with information retrieval. Hence, he sees “cognitive science as
primarily an empirical natural science concerned with a subset of the class of
problems in information science.”
Anyone looking at both Pylyshyn’s paper and mine would immediately
recognize many attitudes in common; a large portion of our historical refer¬
ences is (as was to be expected) the same: mathematics, logic, and mecha¬
nisms. But where he includes cybernetic history, I stress the semiotic and
pragmatic; he includes the neural and psychological generation and recogni¬
tion of ideas and their symbolism where I stress their linguistic, social, and,
perhaps, anthropological generation and recognition. To revert to the pre-
Marxian meaning of the word ideology that I describe in my paper, both
historical approaches, Pylyshyn’s and mine, are ideological, he stressing the
physical side and I the social. But not only the historical questions we raise
are ideological! The very subjects we are discussing are two aspects of what
might be called the art, science, and engineering aspects of ideology. Cogni¬
tive science is concerned with the generation and recognition of ideas; infor¬
matics is concerned with the generation and recognition of symbolizations of
ideas. We are both discussing ideology ideologically.
Since each discipline has its own pragmatically (in the technical sense)
distinct ideology, will this general theory of ideologies be capable of having a
single ideology of its own, or will it have to remain an alliance of disciplines,
what I call a superdiscipline, the way I believe history, education, library
science, and cybernetics must? Pylyshyn’s question of whether there will be
a convergence to one cognitive science is a most valid one. I think the
answer will depend on what I have called interdisciplinary politics. Will an
appropriate brainwashing curriculum for a single major in the subject de¬
velop to form what in the United States educational system is called a
department? It would have to have courses covering the physical, chemical,
biological, psychological, and social generation of the recognitive aspects of
cybernetics and explain or describe or model character recognizers, word
recognizers, sentence recognizers, concept recognizers, picture recog¬
nizers, accepting automata, the variety of technological as well as biological
receptors, and the general processes of understanding and problem-solving!
Destutt de Tracy’s book [1817] on ideology also tried to cover grammar and
logic. I am skeptical, and I suspect Pylyshyn is too.
There is one characteristic ideological attitude that both Pylyshyn and 1
present: What Pylyshyn calls the representational metapostulate, I describe
as part of the mass ego of informatics, as follows:

The very fact, however, that they have symbolisms with well-defined
manipulative procedures may make it useful for them to simulate their pro¬
cesses symbolically. It is one of the informatician’s basic insights (prejudices,
superstitions, ideologic attitudes, or peculiar ways of looking at the world?)
COGNITIVE SCIENCE: A NEW PATCH IN THE QUILT? 97

that any process that can be precisely specified is capable of being symbolically
simulated .... (From my paper in this volume.)

I believe this is not much different from what Pylyshyn says in the second
section of his paper. However, I find his item (1) in that section, the attribu¬
tion of formalist properties (as though it were mathematical) puzzling; for
example, artificial intelligence is much more empirical than formal.
In this connection, I have a question to ask Pylyshyn the psychologist
rather than Pylyshyn the computer scientist: Are not the semantics of the
symbols representing the basic cognitions of most disciplines definitely not
symbol-manipulative? Do they not refer to what they consider to be real
things and not symbols? Is it not only the informatician whose basic con¬
cepts are about symbol manipulation? Even mathematicians resent being
confused with arithmeticians or accountants. How, then, should I interpret
this sentence from Pylyshyn’s paper: “In all these instances, the behavior of
the systems in question (be they minds, computers, or social systems) is
explained, not in terms of intrinsic properties of the system itself, but in
terms of rules and processes that operate on representations of extrinsic
things.” Is this not a computer scientist talking, rather than a psychologist
or a cognitive scientist? Perhaps the title of Pylyshyn’s paper should have
been reversed to read “Cognitive Science: Its Roots and Relations as
Viewed from the Perspective of Information Science.”
Finally, a remark on Pylyshyn’s analysis of the citation statistics is in
order. I believe he uses them as evidence in response to his question about
convergence into one discipline. Although I go along with him on the conver¬
gence question, I do not think citation indices form conclusive evidence.
When a new disciplinary area is being formed, one would naturally expect a
transition period during which the fusion process accompanies the fission
process by which the new body is separated from its parents. It is precisely
when there is a Kuhnian scientific revolution that the new generation breaks
contact with the old (which continues to follow its same old paradigms). Is
this not the standard way in which a new patch is formed in the patchwork
quilt of human lore? In this case, the corners of eight abutting patches (the
previously mentioned cognitive disciplines) are cut off and fused to form a
new patch. If, then, a convergence is occurring (and, I repeat, I am skept¬
ical), present statistics would still be not only what is expected, but would
most likely be misused by the old guard as an argument against the young
Turks.
.
 REFLECTIONS ON THE
STRUCTURE OF
AN INTERDISCIPLINE

Allen Newell

In his “Information Science: Its Roots and Relations as Viewed from the
Perspective of Cognitive Science,” Zenon Pylyshyn outlines the develop¬
ment of cognitive science, an interdiscipline that lies within something called
information science, this latter being too amorphous and wide-ranging to be
termed even an interdiscipline. Here follow a few thoughts on the nature of
interdisciplines.

COMPLICATIONS

The disciplinary structure of science is a crazy quilt. Disciplines emerge and

extend, shrink and disappear, merge and fracture, overlap and surround.
However, even by such balkan standards, the intellectual history of cogni¬
tive science seems to me especially complicated, although no doubt it is just
my professional myopia. Pylyshyn does a good job of making this complex¬
ity apparent. I agree with all the historical strands he lists; however, it seems
to me he gets only half of the complexity out on the table.
For instance, cognitive science is often taken to have six disciplinary
pillars, as in the figure used by the Sloan Committee and reproduced in
Pylyshyn’s paper: psychology, philosophy, linguistics, anthropology,
neuroscience, and computer science. All of these except computer science
(which is a stand-in for artificial intelligence) are essentially absent from
Pylyshyn’s account.

This research was sponsored in part by the Defense Advanced Research Projects Agency
(DOD), ARPA Order No. 3597, monitored by the Air Force Avionics Laboratory under Con¬
tract F33615-78-C-1551. The views and conclusions contained in the paper are those of the
author and should not be interpreted as representing the official policies, either expressed or
implied, of the Defense Advanced Research Projects Agency or the United States government.

99
100 ALLEN NEWELL

To take psychology, with which I am somewhat familiar, a good case can

be made that the situation described in Pylyshyn’s story is right—that psy¬
chology was transformed by an external intellectual movement whose cen¬
tral core comprised control theory, information theory, computation, and
operational mathematics. However, psychology was not a passive patient; it
transformed that transformation and added to it significantly. Some features
of the transformed psychology may follow directly and necessarily from the
central ideas of information-processing; for example, the acceptance of men¬
tal mechanisms in the head and the separation of the study of human and
animal behavior. But the same cannot be said of focusing on memory as the
important structure to research, which has certainly characterized the main
line of the information-processing revolution in psychology. This has its
roots, of course, in psychology’s long-standing concern with learning. Mem¬
ory is the internal mental structure that most directly lies behind learning.
Indeed, lack of any mechanistic notion of memory blocks the study of learn¬
ing in process terms, so that learning itself cannot be taken over immediately
and made the central object of study in an information-processing psy¬
chology.
Another important response of psychology was to fail to develop the
study of problem-solving and higher cognitive processes into a really major
stream of research. Problem-solving and the use of strategies was a striking
feature of early work that influenced psychology, coming partly from game
theory but mostly from artificial intelligence. However, higher cognitive
processes have remained a relatively minor, although integral and conceptu¬
ally important, part of modern cognitive psychology. For many cognitive
psychologists, the cognitive revolution really starts with Ulric Neisser’s
Cognitive Psychology in 1967, ten years after the usual date of the late 1950s,
as pegged by the work of Broadbent, Bruner, Chomsky, Miller, Newell and
Simon, and Swets and Tanner. [Neisser, 1967.] Neisser’s book explicitly
took cognitive psychology to be centered on perceptual and memorial mech¬
anisms. In part, this was due to the emergence in the late sixties of a highly
successful mental chronometry, with the work of Neisser, Posner, and
Sternberg, which, in concert with the focus on memory, locked psychology
into the study of the underlying micro-organization of human cognition.
The continued commitment to a basic experimental methodology was
another important internally determined component of psychology’s re¬
sponse. Although the amount of theory has gradually increased, the central
path to knowledge remains the well-controlled and carefully designed labo¬
ratory experiment, generated to test some question or hypothesis arising
from prior experimental work. Indeed, it remains the norm that theory and
experiment should be coupled in the same person and in the same paper. It is
rare for a purely theoretical paper to appear in the major journals of cogni¬
tive psychology. This commitment to method may even be taken as an
overarching determiner of psychology’s response. Scientists are driven by
what they know how to do next. However fascinating and beckoning a
STRUCTURE OF AN INTERDISCIPLINE 101

potential path, if a scientist does not know how to do the next piece of
research down that path, he or she will rarely go that way.
The point is that modern cognitive psychology is as much a product of the
earlier history of psychology and developments internal to cognitive psy¬
chology, as it is of the external ideas of information-processing. In turn, this
cognitive psychology is a major determiner of the character of the interdisci¬
pline of cognitive science and cannot be left out of the picture.
The story of linguistics, even more than psychology, seems to me one of
internal determination, although I am on less sure ground here. To an out¬
sider, the Chomskian revolution arises from the same general intellectual
milieu that Pylyshyn sketches, impacting linguistics rather than psychology.
The algebraic character of the new linguistics and the fundamental idea of
generative rules characterizing behaving systems—these seem a product of
the same intellectual viewpoint of operational mathematics that gave rise to
operations research, game theory, and control theory. But to an insider,
perhaps to Chomsky himself, the development of the new linguistics looks
almost entirely internal. When he traces roots, as in his Cartesian Linguis¬
tics, it is in terms of internal linguistic history and such general philosophical
positions as rationalism and empiricism. [Chomsky, 1966.] It is certainly not
the Descartes of analytical geometry that the book’s title refers to.
Again, from an internal linguistics perspective, I would suspect that the
continued development would be seen mostly as flowing outward. Certainly,
one can document a transformation of psycholinguistics in the sixties that
flowed directly from linguistics, but left linguistics only modestly changed, if
at all. And there has remained a barrier, both intellectual and political,
between linguistics and computer science, which has kept computational
linguistics in a rather ambiguous status. My concern here is not primarily to
assess or evaluate such relations, but only to note how they complicate the
already complex picture that Pylyshyn sketches.
Another kind of complexity exists, which can be illustrated by the case of
artificial intelligence. Here we have a field that has a fair amount of coher¬
ence, but only a short time ago (the sixties) was itself considered an interdis¬
cipline. How come it is now seen as a disciplinary pillar on which one can
build a new interdiscipline of cognitive science? Indeed artificial intelligence
is not a discipline, in the usual sense of the word, but rather a part of
computer science—which itself is newly come to disciplinary status. In their
wisdom, the makers of the Sloan figure did put computer science as a node
rather than artificial intelligence. But the reality is that almost the entire
connection with computer science lies with artificial intelligence. As an indi¬
cation of how jerry-rigged the structure is, artificial intelligence did not get
itself its own national professional society—the American Association
for Artificial Intelligence—until after the Cognitive Science Society was
formed.
The complexities continue when we consider neuroscience. It also is an
interdiscipline—the neuro parts of anatomy, physiology, pharmacology, and
 ALLEN NEWELL
102

biochemistry—just working itself up to disciplinary status, so the remarks

about artificial intelligence apply here as well. But much more important,
neuroscience does not share a common scientific Zeitgeist with the rest of
cognitive science. Rather, it participates in the central paradigm of modern
biology—as exciting and dynamic a scientific development as the second
half of this century is likely to witness. True, some neuroscientists work
across the boundary, and many on all sides have yearnings for a larger
integration. But such connections seem unlikely to suffice for an interdisci¬
pline. Indeed, taking Pylyshyn’s interconnection statistics seriously, I would
expect the cross-referencing to be almost totally disjoint. To pick a random
statistic of this kind, there happen to be eight Andersons referenced in the
1981 Annual Review of Neuroscience, but among them one does not find any
of the four Andersons familiar to cognitive scientists (James A., John R.,
Norman H., or Richard C.). [Cowan, 1981.]
The complexities continue. Cognitive science shows up as an official
interdiscipline about twenty years after the beginnings of the revolution and
quite a few years after the major pillars had been transformed or created,
each in their various ways (and with much heralding and discussing). So
cognitive science is not so much a harbinger or a precipitation point as a
consolidation or a mid-course correction. In fact, there do not seem to be
any substantive events that mark the creation of cognitive science—no new
intellectual discoveries or even changes of perspective. If a scientific histo¬
rian were given the history since the mid-fifties, with the specific organizing
activities of cognitive science deleted, and told to guess when it occurred,
almost any time at all would do. Perhaps the one aspect that might serve as a
marker is the more active involvement of philosophy, whose engagement
with the implications of information-processing systems before the mid¬
seventies was fairly minimal. This does not, of course, prevent there from
being a few seminal individual contributions, such as those of Hilary
Putnam. [Putnam, I960.]
In this respect, there is a fascinating parallel between the history of cogni¬
tive science and the history of behavioral science, which came into existence
in the mid-fifties. Then it was the Ford Foundation and the pillars were
psychology, social psychology, economics, political science, sociology, and
anthropology; now, it is the Sloan Foundation and the six fields already
mentioned. Then, the argument was whether history was a behavioral sci¬
ence; now, it is whether neuroscience is a cognitive science. Then, even
more than now, the pillars had existed for a long time, and no scientific event
or breakthough signaled the time for integration but, rather, there was a
spreading awareness of the need for integration. Behavioral science is no
longer an active organizing focus. However, the loosening up of disciplinary
boundaries that occurred then remains, as do some more specific legacies,
such as the Center for Advanced Study in the Behavioral Sciences. Behav¬
ioral science has become a social institution somewhat more ephemeral than
that of an interdiscipline but still more real than information science, which
STRUCTURE OF AN INTERDISCIPLINE 103

is almost entirely a descriptive term of the analyst. Cognitive science is still

on the ascendancy and its fate is not yet revealed to us. But this parallel
story indicates that the zoo of social institutions may be more varied than we
usually think.
The complexities keep coming. What of semiotics? According to a recent
article,

Semiotics is the study of signs, i.e., of those entities which effect communica¬
tion between interpreters of signs. Quite a variety of things can function as
signs. A word, a sentence, a gesture, a facial expression, a photograph, a
diagram, etc., are all signs because we, their interpreters, are more concerned
with what they stand for or represent than with what they are merely in them¬
selves. (Laferriere, 1979, p. 434.)

Surely semiotics is a part of cognitive science—or is cognitive science or

includes cognitive science. But we would not know of the existence of
semiotics from the cognitive science literature; even the enumeration of all
the cross-relations in the Sloan figure leaves it out. Pylyshyn, to his credit,
does mention it in passing. But the situation is symmetric. Except for the
occasional quotation from great men common to us all (Warren McCulloch
and Roman Jakobson), the referenced article bears no witness to either the
sources described in Pylyshyn’s paper or any of the massive work done
since then that cognitive scientists know and love.
All disciplines make claims; we should hardly be surprised at that. Like¬
wise, perhaps, it is in the proper nature of claims to overlap. But the com¬
plexities get thicker when multiple fields, here cognitive science and semio¬
tics, claim almost exactly the same intellectual territory and then pass in the
night. And the complexity of this labeling phenomenon bedevils the partici¬
pants in cognitive science as well: Each claims to be the science of much
of the same phenomena. Each, of course, has special slants (including
methodological ones) to bring to the party; but slants are add-ons to a sci¬
ence, subject to change as the field moves on. Again, the point is not to pass
judgment on this situation. It is to note the difficulties that arise in making
sense of the state and structure of science in this area. And so it goes. I
suspect that when the whole set of papers on the disciplines of information is
spread out, along with their multiple commentaries, my few additions will
hardly seem the greater part.

IS THE COMPLEXITY REAL?

Perhaps the difficulty is that we are casting the story in terms of the disciplin¬
ary structure of science. There are certainly alternative ways to tell the story
of an intellectual domain such as information science or cognitive science.
Perhaps the situation would not seem so complicated, if we dealt directly
 ALLEN NEWELL
104

in some other coin—in ideas and discoveries or major scientists or the

paradigms of Thomas Kuhn or the research programmes of Imre Lakatos.
[T. S. Kuhn, 1962a; Lakatos, 1970.] As we all know, the notion of a
discipline continually gets a bad press, especially from academics unhappy
with the felt constraints of departments officiating in the name of this or that
discipline. „ . ,
One possibility is to describe the structure in terms of scientists. Pylyshyn
himself notes how a few scientists show up repeatedly—Warren McCulloch,
George Miller, Herbert Simon, and John von Neumann. But this way lies
madness, it seems to me; such people are invariably polymaths. Their intel¬
lectual histories are about as intricate and convoluted as can be imagined a
task for the skilled biographer rather than a device for making the structure
of fields clear. .
Recently, called on to provide some historical background of artificial
intelligence, I decided to enumerate the global intellectual issues that have
gained prominence over the course of its history. Such issues procedural
versus declarative or serial versus parallel—though usually stated crudely,
seem to capture enduring concerns of a field. In another essay for this
collection, focusing on artificial intelligence, I lay out these issues, which
provide one such alternative description. Alas, one of the most striking
features of artificial intelligence is the large number of issues that have been
prominent at one time or another and which pile up on each other several
deep. Many of the issues there reflect rather directly one or another of the
issues raised by Pylyshyn or previously added by me. Additional issues of
this list, I suspect, will be identifiable in other essays in this collection.
This list hardly constitutes real evidence, though it is probably a bit better
than my Anderson index. Nevertheless, I conclude that the complexity is
real. Whether it is more complex than in other fields remains an open ques¬
tion. But it is a crazy enough quilt to require understanding, and we might as
well continue to examine it through the lens of disciplinary structure.

WHAT IS A DISCIPLINE?

How are we to understand disciplines, so we may understand interdisci¬

plines? Like so many other things concerning humans, there seems to be
both a cognitive dimension and a social/motivational dimension. (For in¬
stance, think of task feedback to humans, which can be cast either cogni¬
tively, as knowledge of results, or motivationally, as reinforcement.)
Cognitively, a discipline seems to be an interreading population of scien¬
tists. Two scientists are in the same discipline if they read the same literature
and feel themselves responsible for keeping up with the same set of ideas.
The notion is not absolute, of course—specialities are well recognized.
However, the notion of a specialty implies a common base. In fact, specia ty
is to discipline as figure is to ground. The discipline becomes important in
STRUCTURE OF AN INTERDISCIPLINE 105

defining the intellectual ground against which scientists do their work; that
is, pursue their specialties.
Thus, a significant aspect of artificial intelligence not being a discipline in
itself and being part of the discipline of computer science is what the scien¬
tist in artificial intelligence knows about besides his or her own research,
say, on expert systems. If artificial intelligence were a discipline, all the
scientist would know about are other aspects of intelligent systems, such as
vision, theorem-proving, and natural language. With artificial intelligence
part of computer science, the scientist also knows about work on multipro¬
cessing architectures, verification, algorithm analysis, and so on. Corre¬
spondingly, of course, he or she knows only a little about what is going on in
cognitive psychology and nothing at all about what is going on in social
psychology.
If, by some chance of fate or fashion, artificial intelligence had become
part of psychology or philosophy or, say, a developed cybernetics, then
quite different things would be part of the ground. Advances in behavior
modification, phenomenological philosophy, or muscle control, respec¬
tively, would be a living part of the artificial intelligence scientist’s world.
Correspondingly, VLSI (very-large-scale integration) and the structure of
protection systems would not.
What intellectual ground a scientist stands on is immensely important. It
affects what intellectual resources are drawn on in choosing frameworks for
analysis, interpretation of difficulties, sources of inspiration, and the kernel
ideas for solutions to conceptual problems. It affects to whom the scientist
can talk about his or her problems—namely, those who share common
ground—and, therefore, what types of suggestions he or she obtains. In
short, though it does not affect the position or velocity of the work, which is
determined by the research task itself, the discipline affects all higher de¬
rivatives.
One fundamental cognitive reason why disciplines exist seems appar¬
ent—it comes from resource limitations to human cognition. There are only
so many hours in the day; they can be filled with only so much talk and only
so many articles. These limits assure that no scientist will be familiar with
more than a small fraction of science; thus, science cannot be the seamless
web we all desire. However, such limits do not determine that science will
partition itself into (disciplinary) clumps, in which many scientists read ap¬
proximately the same scientific literature and few read heavily across disci¬
plines. It seems equally plausible that each scientist would trace out an
idiosyncratic problem-oriented trajectory through science. The whole of sci¬
ence would then have the appearance of a randomly woven but homoge¬
neous-appearing mat—a crazy quilt at the individual level rather than at the
discipline level.
There is an entirely different side to disciplines and their structure,
namely, they are social institutions and serve social and motivational needs.
They provide the rocks to which scientists attach their identities. “Hello,
 ALLEN NEWELL
106

who are you?” “I’m a physicist” or a philosopher or a biochemist. That

people will answer “I’m a computer scientist” shows that computer science
has become a discipline. Disciplines carry a point of view—to answer ‘T’m
an anthropologist” declares the view that culture is an important explana¬
tory ingredient in human affairs. One does not pronounce on another disci¬
pline’s subject matter without appropriate deferential behavior. This is not
just a matter of familiarity, for the hierarchy of the sciences is in part a
social-dominance hierarchy. Physicists can pronounce on psychology with
less deference than vice versa.
Again, on the surface, there seems little mystery behind the institutional
nature of disciplines. They are nurtured by the organization of universities
along disciplinary lines into departments, supported by professional
societies. Critical to this is the strong tradition of graduate education, lo¬
calized in the departments, with its almost exclusive command of the
young’s attention during the period when career identities are shaped. This
is compounded by the faculty of a department being composed almost exclu¬
sively of scientists with PhDs from the discipline. Thus, from a social view¬
point, there are ample grounds for the disciplinary structure that exists in
science perpetuating itself. However, it seems less clear that organizing
science according to disciplines has any favored status over other ways of
organizing, in terms of social mechanisms. Its emergence may strongly
reflect historical factors.

WHAT IS THE ROLE OF INTERDISCIPLINES?

Given the disciplinary structure, as just described, what then is the role (or
set of roles) of interdisciplines? There are two obvious ones. First, science is
in fact all of a piece; thus, any decomposition into disciplines must necessar¬
ily be imperfect. Interdisciplines represent a spillover as groups of scientists
with common interests read each other’s work and begin to cooperate with
each other. In this role, interdisciplines can be permanent and stable institu¬
tional structures in science.
Second, the division into disciplines changes with history. Early in the
development of science, new sciences precipitate out of a less differentiated
scientific and intellectual activity—for example, natural philosophy. The
emergence of physics and chemistry as separate disciplines would be an
example, or the emergence of statistics. As science matures, all of the terri¬
tory is covered in one way or another by some science, then new sciences
are carved out of existing ones. Interdisciplines are simply early forms of
disciplines on the way to existence; biochemistry provides an example.
Other roles are possible: An interdiscipline could be an intermediate stage
where part of one discipline moves to another. It could be a temporary
scaffolding while some important ideas are transmitted to a set of fields, only
STRUCTURE OF AN INTERDISCIPLINE 107

to disintegrate when this task has been accomplished. It could simply be a

device for periodic widening of scientific awareness, more analogous to a
professional meeting (though on a much longer time scale) than a disciplin¬
ary structure. Or, it could be similar to a special-interest group, permanently
established to signal the importance of some aspect otherwise ignored.
Psycholinguistics is perhaps an example of this latter. All of these are cogni¬
tively centered roles, but social roles are also possible, such as being a
structure within which to provide funding.
So what is true of cognitive science? It seems to me unlikely that it is the
nascent form of a new discipline, carved out of psychology, linguistics,
computer science, and so forth. The determining factor, it seems to me, is
the role of the parts in the parent disciplines, and the determining factor
would seem to be the ability of the parts to be excised from their parent
disciplines. This would be eased if a part were a specialty or peripheral in
some way, easiest perhaps if a strain line already existed along which cleav¬
age could occur. Psychology seems to me the key: The part of psychology
included in cognitive science is cognitive psychology, which is the central
core of the parent discipline—what experimental psychology was of yore. It
seems inconceivable that there could exist a psychology without cognitive
psychology; what would it be? And without cognitive psychology, it is in¬
conceivable that there could be cognitive science. For the other parts, the
situation is more favorable to founding a new discipline (though with other
consequences of course). The linguistics part of cognitive science could just
become the new psycholinguistics, and linguistics could proceed without it,
more or less as it has without psycholinguistics. Artificial intelligence has a
fault line down the middle, separating the concern with engineering and
applied systems from the concern with the basic nature of intelligence and
human intelligence. This latter half is the part that participates in cognitive
science; a split is certainly possible along this cleavage line. Thus, artificial
intelligence would not leave computer science, but a part of it would. The
other three cognitive science partners—philosophy, anthropology, and
neuroscience—seem less critical in determining whether a new discipline
will congeal.
It seems much more likely that cognitive science plays one of the other
roles. Perhaps, it is a more or less permanent spillover, reflecting imperfec¬
tions in the particular arrangements of current disciplines concerned with
the nature of mind. The imperfections are certainly clear to all. With so
many parties to the endeavor, hence, so much need for work at this or that
point across a boundary, even if arrangements falter from time to time, they
will be continually reactivated.
Pylyshyn notes that cognitive science may not even be much of an inter¬
discipline. Thus, perhaps even this last position is too strong. Recall, again,
behavioral science, which seemed strong in its day, but has faded to much
less than an interdiscipline. Pylyshyn bolsters his concern with some simple
108 ALLEN NEWELL

counts of the extent of intercitation. Despite my slight caricature with the

Anderson statistic (because I was too lazy to do better), I am in favor of any
attempt to provide some data, however scrappy, for discussions of this sort.
In fact, however sketchy Pylyshyn’s own data, I find I have little urge to
disagree with the pattern revealed. On the other hand, I believe, in fact, that
the component fields in cognitive science are strongly interdisciplinary.
Thus, it seems necessary to find a plausible interpretation of the pattern
Pylyshyn found.
First, I think a set of fields are interdisciplinary to the extent that ideas
interpenetrate. The ideas may be any type—theories, data, methods, prob¬
lems, or perspectives. However, they must receive their development in
field X and eventually be used in some form in field Y. Commerce must occur
in all directions and continue over some period, thus involving a population
of ideas. The citation data presented by Pylyshyn refer to the channels over
which the penetration occurs. I find it quite plausible that much of the
penetration goes via one or a few seminal publications or works. Scientists
in general cite publications that have not only influenced them but that they
have read. In general, these will be only the seminal works—they are
sufficient to convey the idea. There is no need to examine the original work
in the parent field or to follow in detail its continued advancement there. In
general, only the gross aspects of an idea—to wit, the idea itself—is useful
in the recipient field. Actually, a significant transform is usually necessary to
make the idea applicable to the recipient field, which is accomplished by the
transferring publication. Thus, there is no reason to go beyond it.
I will not attempt to construct a map to demonstrate the interdisciplinary
nature of cognitive science; that is beyond the limits of an already long note.
I assert cognitive science scores quite well, with repeated ideas flowing back
and forth. The flow is not equal at all times and between all subfields, nor
should it be. There will certainly be some patterning to it. But it contrasts
strikingly with a unidirectional, one-time penetration, such as that from
information theory into psychology in the early 1950s. To take just a single
example, that of semantic nets, several ideas have flowed back and forth
over a period of 15 years (so far), primarily between psychology and artificial
intelligence. In each of the instances, only a couple of publications were
involved, and they were each reports of bridging research results. [For
example, Quillian, 1968; Anderson and Bower, 1973; Schank and Abelson,
1977.]
This view of the nature of being interdisciplinary seems to make it un¬
likely that cognitive science is just a more widespread but weaker interread¬
ing population of scientists, which is what the spillover role would imply. It
is more like an intermittent-communication device, in which the continuous
intercommunication, as in common journals and common meetings, is a way
of keeping the background level high enough to trigger new ideas occasion¬
ally and, although it seems less necessary, for the reception of those seminal
publications when they occur.
STRUCTURE OF AN INTERDISCIPLINE 109

WHAT WILL HAPPEN TO IT ALL?

I end where I began. The disciplinary picture for cognitive science is a crazy
quilt and even more so for information science. I believe it will stay that way
almost indefinitely. The intellectual domain of understanding human nature
is too extended and admits too many approaches to enforce unification. All
that is required to avoid any serious pressure for unification is that en¬
deavors be separate in their problems. Overlapping claims need not violate
this, because claims are made by area, but separation is determined by
problem. As long as the approaches in a field that answer one problem do not
suffice to answer nearby ones, then many approaches can coexist with little
pressure to unify them. The progress of an approach taken in isolation is not
relevant—it can be real, measurable, and even dramatic, without implying
that other approaches to the same arena should be abandoned.
Will it always be this way? That is too far in the future to guess. I believe
the essential condition for the gradual simplification of the disciplinary pic¬
ture for cognitive science lies in the development of techniques that solve
well-specified scientific problems and answer scientific questions in routine
ways. The phonetic alphabet is a good example; we all depend on linguistics
for that. The representations for grammars are another; I believe they will be
around and used routinely by all scientists, even when our conceptions of
language have evolved considerably. Similarly, from psychology comes sig¬
nal-detection theory, with the concept of the operating characteristic, and
from artificial intelligence comes the LISP programming language. These all
become islands of certainty that all disciplines borrow and use when needed.
Though sometimes such techniques become irrelevant and drift away, by
and large they form a permanent accumulation. They can be expected to
increase gradually (and sometimes rapidly) to form a framework that will
ultimately force simplification of the disciplinary structure. Until that time,
there is no reason for those professing an interdiscipline not to let the crazy
quilt continue. Furthermore, there are good reasons to be confident that
ideas will continue to move from discipline to discipline.
.

■
 IN FORMA VORES

George A. Miller

Both the human tendency to think analogically and the dangers of reasoning
by analogy are familiar themes to students of the history of ideas. One
cannot help but wonder what about the human mind compels it to flirt so
outrageously with the potentially disastrous consequences of false analo¬
gies.
The general pattern of analogical thinking, of course, is to explain some¬
thing that is poorly understood by noting its similarity to something that is
well understood. Thus, it is not uncommon to explain complicated natural
phenomena by analogy to human artifacts, since people generally feel that
they understand reasonably well those things they themselves have actually
manufactured. In the life sciences, one thinks of such examples as ex¬
plaining the motions of the limbs by analogy with levers, explaining the
circulation of the blood in terms of the analogy of the heart to a pump,
explaining eyes and ears by analogy to cameras and microphones, or ex¬
plaining metabolism by analogy to heat engines.
Attempts to understand human cognition have generated their own
analogies. The mind has been likened to a cave on whose walls events cast
their shadows, to a slate on which experience can write, to a hydraulic
system for pumping energy into alternative activities, to a telephone switch¬
board that connects ideas to one another or responses to stimuli, even to a
hologram that stores representations distributively. The current favorite, of
course, is the modern, high-speed, serial, stored-program, digital computer,
an analogy that has catalyzed the rebirth of cognitive psychology as an
active scientific enterprise.
In What is Life?, a little book that opened up biology for physicists, Erwin
Schrodinger pointed out that organisms survive by ingesting, not food, not
calories, but negative entropy. [Schrodinger, 1945.] It is no accident, of
course, that the mathematics of entropy are also the mathematics of infor¬
mation. The analogy is obvious: Just as the body survives by ingesting
negative entropy, so the mind survives by ingesting information. In a very
general sense, all higher organisms are informavores.
But how should this suggestive parallel be developed? No sooner was the
question asked than computer scientists offered an answer. The computer
ill
112 GEORGE A. MILLER

seemed to provide a precise, well-understood system that could serve as a

laboratory for the analysis of all kinds of informavores. The human informa-
vore seemed merely a special case of the more general theory of information
processing that remained to be discovered. And so the informavore analogy
was transformed almost immediately into the computer analogy.
Perhaps it is unfair to speak of these hypotheses as analogies. To someone
caught up in an analogical line of reasoning, the hypothesis under considera¬
tion must seem much more than a mere analogy. Unless an analogy is taken
very seriously, the motivation to explore its implications will be lacking.
Instead of calling the hypothesis an analogy, therefore, it may be promoted
to the status of a theory. But calling an analogy a theory does not make it
one; only hard, empirical work can accomplish that. (In the study of cogni¬
tion, an analogical flavor is often preserved in the preference for speaking of
models rather than theories.)
Pylyshyn draws an accurate picture of cognitive science today, a picture
in which the languages that computer science has provided for formulating
cognitive theories and the tests of those theories that computer simulation
allows seem to outrun by far any mere analogy. Still, the heart of the matter
is the exploration of an analogy.
Which is not a bad thing. To say that modern conceptions of machines
will continue to evolve, and that richer cognitive analogies will evolve along
with them, is not to criticize the progress that has already resulted from
exploiting analogies presently available. It is merely to suggest that current
conceptions may still be incomplete.
Consider an analogy between analogies. Insofar as a limb is used as lever,
the theory of levers describes its behavior—but a theory of levers does not
answer every question that might be asked about the structure and function
of the limbs of animals. Insofar as a mind is used to process information, the
theory of information processing describes its behavior—but a theory of
information processing does not answer every question that might be asked
about the structure and function of the minds of human beings.
Or does it? Suppose all psychologists restricted themselves to theories
that could be simulated and tested on contemporary computers. Would any¬
thing of importance be lost?
Such questions may seem premature while so much still remains to be
done to explore the limits of the computer analogy, but it is important to
have as clear a perspective as possible on any analogy we happen to be
considering. Pylyshyn, in his paper, comments that a “commitment to the
informational level” places cognitive science in contrast to “the phenom¬
enological approach, where the existential notions of significance,
meaningfulness, and experiential content are given a central role.”
To some psychologists, these would be serious losses indeed. If one
defines psychology as the science that deals with significance, meaning¬
fulness, and experiential content (and some psychologists do), then progress
in the information sciences will be bought at the end of this century at the
informavor.es 113

same price progress in formal logic was bought at the beginning of the
century—at the price of a total divorce from psychologism.
But the informational level may be richer than we now know. That is to
say, computers as we presently know them may prove to be incomplete
analogues of informavores. The claim would not be that there is something
computable that they cannot compute but simply that evolution may have
produced a species of informavore—a kind of special purpose computer—
that we as yet have no idea how to construct or imitate.

* * *

The opposite of analysis, in one sense, is synthesis. In that sense, the

computer analogy offers a priceless antidote to behavioral analysis—the
possibility of synthesizing and testing integrated, goal-oriented systems. It is
this possibility that is contributing so generously to what Pylyshyn calls “the
convergence toward cognitive science.”
In another sense, however, the opposite of analysis is acceptance. In that
sense, analysis characterizes an active, problem-solving attitude of mind. In
the analytic mode, one asks questions that impose analytical categories and
symbols on experience; by contrast, in the accepting mode, one accepts
each moment as a new miracle and appreciates it with gratitude and wonder.
In this sense of analytic, present-day computers may be too analytic to
model completely a human mind capable of both modes of learning. It is
noteworthy, for example, that information-processing models of human
memory have a great deal to say about the storage and retrieval of verbal
messages but almost nothing to say about memory of a pain, of the sound of
a person’s voice, or of the taste of a good wine.
Perhaps the kind of processing that modern computers do so rapidly and
so well is not the only thing informavores can do with information. The
future of the information sciences may be even brighter than anyone can
now foresee.
 REPRESENTATION,
COMPUTATION,
AND COGNITION

Zenon W. Pylyshyn

The five commentaries on my chapter are an excellent sample of the diverse

viewpoints that comprise cognitive science. They go a long way toward
redressing my own biases in emphasizing certain features of this new disci¬
pline or interdiscipline. For this reason, I have little to say to the commen¬
tators individually beyond thanking them for their positive contribution.
Instead, I take this opportunity to remark in a general way on a number of
themes that recurred in much of the discussion. One is the notion of a
discipline and the other is the question of the scope and character of cogni¬
tive science as explanatory science.
Allen Newell and Saul Gorn both raised the question of whether cognitive
science is a discipline, an interdiscipline, or (in Gorn’s terminology) an
idealogy. Disciplines, in the usual sense of this term, are a lot like academic
departments. They can be based on almost any kind of cohesive purpose: a
concern with what appear superficially to be similar puzzles (Why do people
behave the way they do? What makes it possible for certain organisms to
reason and adapt intelligently?); a concern with the same objects of study
(people, animals); the use of similar formal techniques (algebra, difference
equations, field theories, etc.), similar empirical methods (statistical experi¬
mental designs), or even similar physical facilities. In my own university,
psychology is considered to be a social science for undergraduate-teaching
purposes and a biological science for graduate and research purposes,
chiefly because for the latter purposes, it requires specialized laboratory
space and facilities. Thus, the survival of cognitive science as a departmen-
tally recognized discipline or even as an interdiscipline depends at least as
much on political, economic, and social factors as it does on scientific ones.
Clearly, these are not the sorts of issues that concern us in trying to under¬
stand the character of cognitive science as a science.
What is really central in that case is the question of whether there can be
an explanatory, theoretical science of cognition. This is the question whether
115
 ZENON PYLYSHYN
116

our present notion of the subject matter of such a science is close to being
what philosophers call a natural kind. It has been widely recognized ever
since Plato first drew our attention to it—that in order to develop a success¬
ful explanatory system, we must “carve nature at her joints." The way we
group and distinguish properties and phenomena is crucial to the success of
the scientific project. Thus, for a successful theory of motion, we must not
distinguish, as did the ancients, between natural and violent motion, though
we do have to distinguish between weight and mass, a distinction with which
medieval thinkers struggled but which was not clarified until the seventeenth
century. Such distinctions serve to carve out a natural and relatively homo¬
geneous scientific domain within which a set of explanatory principles apply.
It had been assumed for centuries that the appropriate natural domain of
study for what we informally call cognition (or higher mental functions) is
something like complex biological systems. In the 1940s and 1950s, this
domain was expanded beyond biology to include what were called cyber¬
netic systems. Gradually, and through the influence of the sorts of histoiical
developments to which several of the commentators alluded (most particu¬
larly the work in artificial intelligence), a picture of natural domain began to
evolve. It looked more and more as though the “natural kind to which the
theories would apply were those systems that were governed by repiesenta-
tions or by knowledge and goals. Newell, in a similar analysis of the current
state of the field, referred to this as the “physical symbol system hy¬
pothesis.” [Newell, 19806.]
This was not simply a case of expanding the domain of study by finding
increasingly abstract ways of viewing systems (as has sometimes been
claimed of general systems theory). The representation view also has a
restricting effect. Some of the central concepts of psychology, as George
Miller notes, are excluded in the process. For example, while the distinction
between conscious and unconscious representations is glossed over, the
subjective-feeling component (the “qualia") of mental states falls outside
the domain of such cognitivist explanations. Similarly, the physical forms in
which representational systems are instantiated becomes a separate field of
study. The work of cognitive neuroscientists like Michael Gazzaniga and
Michael Arbib is taken to be a contribution to our understanding of how one
particular class of machines—the biological one—is able to carry out repre¬
sentation-governed processes. Their work is a study of the underlying func¬
tional architecture, not the rule-governed representational processes them¬
selves, since these constitute an autonomous natural domain with
regularities that can be captured independently of particular physical instan¬
tiations. The domain of cognition is thus viewed as a natural kind encom¬
passing all informavores (as George Miller puts it so colorfully) rather than a
human or even a biological category of functioning. What is lost in the
processes is the concept of a domain of study defined in terms of our naive
interests rather than in terms of its scientific tractability. But such losses are
routine in scientific evolutions: Astronomy, too, lost much glamor and won-
REPRESENTATION, COMPUTATION, AND COGNITION 117

der when it was discovered that astrological questions were not in its pur¬
view.
I end with some brief comments prompted by Saul Gorn’s remarks con¬
cerning the notion of a formal symbol and semantics. Being governed by
representations is not the same as being formal and computational. Nonethe¬
less, computation is the only model we have of how a physical system can be
governed by representation. As such, the computation view of mind is
neither a logical necessity nor a metaphor: It is a far-reaching empirical
hypothesis. Thus, Saul Gorn is exactly right when he says that I am viewing
cognition from the perspective of information science rather than the con¬
verse. That is because I believe this is precisely the view that characterizes
cognitive science; it is a sweeping reanalysis of the notion of cognition itself.
What I called the representational metapostulate is not the same as the “basic
insight of informatics” that Gorn refers to; namely, “that any process that can
be precisely specified is capable of being symbolically simulated.” Rather, it
is the much stronger hypothesis that cognition consists in (not “is capable of
being simulated by”) manipulating physically instantiated symbols. It is the
quite literal proposal that cognition is computation.
Because of this postulate, symbols do not have the same status in cogni¬
tive science that they have in either physics or semiotics, though for differ¬
ent reasons in the two cases. In physics, symbols represent physical mag¬
nitudes and the relations among these: They are a notation for the subject
matter only. The symbols themselves are not assumed to have any realiza¬
tion except in the expression of the theory. Thus, the choice of a notation,
though it has important consequences for the utility of the theory, does not
matter for the empirical truth of the theory. In cognitive science, as in
computing, symbols not only refer to some extrinsic domain (say, numbers
or beliefs), but they are also real objects in their own right that the machine
(or, by hypothesis, the mind) manipulates in a manner that preserves the
semantical properties of those symbols. In the case of numerical computa¬
tion, for example, symbol manipulation is carried out in such a way that
when interpreted as numbers according to some uniform scheme, the ex¬
pressions remain congruent with number-theoretic properties. That is why
we can quite literally say that the machine carries out arithmetic processes.
According to the view I have been describing, exactly the same story is to be
told about cognition in general. Because of this, in cognitive science the
notation matters: It constitutes a substantive claim about the cognitive pro¬
cess. Gorn is right in remarking that this is the computer scientist talking,
but that is because computer science and cognitive science converge pre¬
cisely at this point.
Cognitive science is also quite different from semiotics in the kind of
symbols that it studies, which may account for the quite wide divergence
between the two approaches (as Newell notes). Semiotics is concerned with
what might be called secondary symbols. These are the symbols that we
humans use in order to convey meanings. They are not symbols that func-
 ZENON PYLYSHYN
118

tion directly as intrinsic causes of behavior, the way symbols do in com¬

puters. The symbols of the semiotician have no meanings and exhibit no
behavior unless there is an intelligent, knowing agent to interpret them.
These secondary symbols get all their meaning from their social and conven¬
tional (and sometimes personal) use. Hence, to study these secondary sym¬
bols is to study the body of cultural conventions, intentions, aspirations, and
so on, of individuals and groups. By contrast, mental symbols of the sort
that concern cognitive science, like computational symbols, have intrinsic
meaning (semantics) by virtue of being instantiated in a physical mechanism
in such a way that they interact causally with each other and the world
outside (through transducers).
These few reactions do not adequately address some of the genuinely
contentious issues that several of the commentators raised. My original
paper was written in the spirit of investigative and analytical reporting, not
in the spirit of scientific imperialism. If it sounded as though I was claiming
that the computational perspective of cognitive science is an important new
direction in the study of mind, that is because I really do believe that if the
representational metapostulate can be empirically sustained, it will prove to
be the biggest breakthrough in psychological conceptualization since
Descartes. The representational metapostulate will give us not only a handle
on the important problem of how intentional rule-governed systems can be
compatible with physical causality, but it also comes equipped with a set of
working tools for actually building testable theories of the mind.
SECTION 2
 INFORMATICS (COMPUTER
AND INFORMATION SCIENCE)
Its Ideology, Methodology, and Sociology

Saul Gorn

My task is to discuss my area—computer and information science—and its

relations to some neighboring areas, among which I mention library science,
information retrieval, information science, cybernetics, cognitive psychol¬
ogy, artificial intelligence, semiotics, and linguistics. I add computer en¬
gineering, management, and decision science, even education, and, of
course, the field in which I was brought up, pure mathematics. The purpose
of this discussion is to distinguish my new area from these others pragmati¬
cally and methodologically but especially ideologically.

THE PLAN FOR THIS PAPER

Let me, first of all, choose a shorter expression than computer and informa¬
tion science. Like the French (informatique) and the German (Informatik),
let me choose the word “informatics.” It contains the idea of information
and has an ending like that of mathematics, implying a formally based
theory. It is too bad that informatics loses the computational component in
its name and also leaves no impression of an experimental basis, both of
which I consider equally important, as we shall see.
Previous papers of mine on the subject were addressed to applied mathe¬
maticians, pure mathematicians, library scientists, logicians, linguists,
pedagogues, behavioral scientists, and political scientists. [Gorn, 1965;
1967; 1968.] I even discussed why each audience needed a different presen¬
tation. The audience on the present occasion is the mixture of the neigh¬
boring areas I mentioned first, and the purpose of this paper and those from
the other areas is to help us sort out our separate identities, if they are indeed
separate.
I propose to do my part by first discussing my picture of the disciplines of
human knowledge and action, their ontogenic development, and resulting
121
122 SAUL CORN

distinct ideological identities and phylogenic relations. Then, I will identify

the ideology and areas of study within informatics and examine its relation to
the other areas; and finally, I will discuss their stability in the social sense.
Let me outline the sections in greater detail. For the section on the disci¬
plines of human knowledge and action, I have an agenda of twelve points: (1)
The social specialization into disciplines and, in particular, the definition of a
discipline and the relation of disciplines to language and the history of disci¬
plines to the genesis of language. (2) The meanings of the word ideology. (3)
The dependence of the sciences, arts, and crafts on the extension of human
perceptive and motor ranges and their consequent interdependence with
technology. (4) How this dependence makes empiricism itself dependent on
advanced technology, with many well-known examples from the history of
science. (5) How this dependence creates the ontological commitments that
I have called waves of agreement to perceive; they correspond to Kuhn’s
paradigms and occur during his scientific revolutions, especially when they
lead to a social agreement to support the economics of the production of
advanced receptor and effector instruments. (6) The changes in the meaning
of the word science. (7) How taking a wave theory rather than a particle
theory of knowledge keeps us from forgetting the pragmatics of symbolism.
(8) The language and the metalanguage developments in disciplines and how
the metalanguage level in symbolic thought can cause a separation of knowl¬
edge and action. (9) How mathematics and the physical sciences are sup¬
posed to be independent of the pragmatics of their language (where the word
pragmatics is used in a technical sense). (10) How the metalinguistic as¬
sumptions of determinism versus free will distinguish the knowledge-
oriented from the action-oriented disciplines, and how the professions try to
tie them together to expand our powers despite our limitations. (11) The
diachronies of the languages of disciplines and the stages at which they
communicate with mathematics and informatics. (12) The mass ego of a
discipline.
In the section called The Informatic Outlook, I shall discuss eight issues:
(1) How symbol-manipulation activity was accelerated by electronic com¬
puters because they demand decision-making instructions and common stor¬
age of instructions and data. (2) How the programming activity becomes
another symbol-manipulating activity capable of mechanical handling and,
therefore, a self-referencing activity. (3) How the first result is the area of
artificial intelligence, and the second result is the emergence of the principle
of logical equivalence of hardware and software, which yields the character¬
istic idealized concept of processor and the ambiguous meaning of assembly.
(4) The syntactic, semantic, and pragmatic aspects of informatics. (5) The
syntactic aspects—automata theory and formal languages—and the resul¬
tant theory of computation with a specification of the limits of computation.
(6) The semantics of symbols versus the semantics of symbol manipulation:
How symbol manipulation simulates the processes symbolized. (7) How to
distinguish disciplines by their ideologies: the differing semantics of physi-
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 123

cists, mathematicians, and informaticians. (8) How the areas within infor¬
matics will vary from the formal and mathematical, through the empirical
and experimental (especially where unsolvable and unsolved problems are
concerned), through aesthetic and ethical reactions, into the variety of ap¬
plied informatics in a number of professional activities.
Finally, the last section, on the intercommunication and stability of disci¬
plines, will deal with seven topics: (1) The two meanings of ideology that we
are using (as contrasted with the Marxian meaning and its relatives); the
sociology of ideas versus the biology and psychology of ideas and their
connection because of the need of social support for the advanced technol¬
ogy needed to maintain them. (2) How to compare and contrast the ideolo¬
gies of disciplines. (3) Cybernetics as a superdiscipline and robotics as a
professional activity. (4) The managerial-professional activities related to
informatics. (5) Pedagogy as a superdiscipline; the impossibility of a single
pedagogic ideology and the similar impossibility of a single ideology for
information retrieval. (6) The library profession and its long-range stability:
the dependence of the stability of a discipline on both the existence of stable
ideologies (to maintain communication) and the existence of unsolvable
problems (to prevent technological obsolescence). (7) How the stability of
informatics depends on maintaining the interlocking of its scientific, artis¬
tic, and professional activities.

DISCIPLINES AND THEIR IDEOLOGIES

We all know that ever since the agricultural and urban revolutions humans
have tended to specialize their activities. The result is that, by now, human
lore, especially in Western civilization, is a patchwork quilt of areas of
activity called arts, sciences, and professions. In spite of such philosophies
as operationalism, some of these areas still seem to be mainly knowledge-
oriented, such as mathematics, the natural and social sciences, the humani¬
ties, and other academic disciplines, while others seem to be mainly action-
oriented, like farming, sports, hunting, cooking, mining, manufacturing,
constructing, and the like. Still others seem to have as their purpose the
transformation of knowledge into action, such as the professions of en¬
gineering, medicine, law, education, and so forth. Let me use the word
discipline to cover them all, since the dictionary gives the following mean¬
ings: branch of knowledge involving research; rule or system of rules affect¬
ing conduct; and even, training that corrects, molds, strengthens, or perfects.
The patchwork quilt of disciplines is like a series of slightly distorted
plane projections that a mapmaker might make of, say, a doughnut-shaped
planet. Each, therefore, represents a portion of human activity from a
specialized point of view. It seems to me that these specialized representa¬
tions, especially for the more sophisticated disciplines, are for separate ide¬
ologies with definite histories of development and even changes of territory
 124 SAUL CORN

covered; that these histories of development are similar to the diachronic

histories of the development of language and are even similar to early stages
in the development of linguistic ability in a child; and that the results are
rather distinct Weltanschauungen that change in time. [Gorn, 1965; 1967.]
Although a meaning of ideology in one of my dictionaries is the integrated
assertions, theories, and aims that constitute a sociopolitical program, espe¬
cially as the word was used by Hegel and Marx, the earlier broader meanings
are still recorded: a systematic body of concepts; a manner, or the content of
thinking, characteristic of an individual, group, or culture. I believe that
Destutt de Tracy also intended it to mean the development and presentation
of ideas and would have considered the present developmental psycholo¬
gists such as Piaget to be direct descendants. [Picavet, 1891 and 1971; De¬
stutt de Tracy, 1817.]
Human knowledge began by being strictly limited by human perceptive
devices the senses—and their limited ranges. We have a limited audio¬
frequency range and visual-frequency range and similar limited ranges for
our other senses as well as for our motor controls. It took technological
inventiveness to extend our perceptions and controls and, indeed, the physi¬
cal ranges in which we could continue to exist. To change, or even to
understand, the temperature range of our surroundings already called for
controlling fire; the invention of tools, utensils, and containers; the invention
of clothing and shelter; and observing the seasons, so that even before the
agricultural revolution, technology had begun. The division of labor that
followed, leading into the urban revolution, resulted in a variety of arts,
crafts, and professions. The more knowledge-oriented disciplines of the type
now called physical sciences, however, had to wait for the appropriate tech¬
nological advances that extended our perceptions beyond the range of our
senses. Devices that extended our counting ability, or our ability to measure
length, area, and volume, had to appear before arithmetic could develop; the
abstractions of number, length, area, and volume could not even exist in the
human mind to permit the development of mathematics prior to such techno¬
logical advance. As we were to see over and over again, our decision in each
science of what new things or processes were worth perceiving depended on
an advance in technology. [See, for example, John Dewey, 1929.] And,
thenceforth, physical scientists and technologists together formed a Jacob’s
ladder, each bootstrapping the other’s knowledge into a more advanced
state.
Similarly, even an advanced and sophisticated stage in such arts as sculp¬
ture, architecture, painting, and music could not occur without the technol¬
ogy sufficient to standardize—and therefore permit the harmonization of—
line, mass, perspective, and color vision and their composition for our
eyes, or rhythm, melody, and timbres of sound and their composition for our
ears. We had to standardize, not merely our perception of these arts through
new measuring devices, but our production of them through new instru¬
ments and techniques. In fact, the necessary technology for these arts did
not develop until the sixteenth through the nineteenth centuries.
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 125

We see, then, that it was only when technology had advanced sufficiently
that empiricism could even become conceivable, let alone possible. The
human spectra and ranges of perception and control had to be extensible by
machines, or there would be no evidence on which to base empiricism. The
idea of empiricism may have had early philosophical roots—from Aristotle
to Roger Bacon—but it was not until the seventeenth century that Francis
Bacon’s form of it was expressed, and it took centuries before empiricism
could really flourish.
For example, statics and dynamics developed in the time of Archimedes
(300 B.C.), together with various hydraulic and ballistic technologies. That
was when the concepts of force and weight and pressure outside the human
range really became perceptible and measurable. Similar developments tied
optics to astronomy and biology before measurement could advance into the
macroscopic and microscopic ranges. Certainly, the biological concept of
the cell could have no empirical base before then. It took many decades
before temperature could be measured and centuries before its relation to
mechanics could be appreciated. Thomas Kuhn points out that it is impossi¬
ble to indicate the moment of discovery of oxygen. Over some decades,
Priestley, Lavoisier, Scheele, and others did some individual groping with
experiments and technological devices to control the materials and mutual
communicating on what to capture and how, before it could be said that
anyone knew what oxygen was. [T. S. Kuhn, 1962a; 1977.] It took years for
these scientists to decide what it was that ought to be perceived and to
extend their senses and controls by technological devices in order to per¬
ceive it. This is the social phenomenon that I have elsewhere called a wave
of agreement to perceive. And there had to be this social agreement to
perceive what was unperceivable before, or else the social expense required
to manufacture the artificial effectors and receptors would not be accepted.
This is especially marked in twentieth-century physics and its dependence
on accelerators.
The basic chemical concept of an element, as it was introduced by Dalton
and is thought of today, could not even exist (in spite of the ancient forms of
atomism), let alone have an empirical basis, until after a number of occur¬
rences similar to the story of oxygen; the principle that chemicals combine in
fixed proportions and the principle of conservation of mass was inconceiv¬
able, let alone unmeasurable, until then. Similar things could be said about
the concepts of energy in each of the variety of meanings it has had over the
millenia. The principle of conservation of energy was inconceivable and
inexpressible before the late nineteenth century.
When Thomas Kuhn talks about scientific revolutions, where some of the
paradigms of the science are replaced, I interpret this to mean that the
community forming that science has decided that something considered
worth perceiving before is no longer worth perceiving and is to be replaced
by a new basic concept or a new basic principle, using a new insight. [T. S.
Kuhn, 1962a.] On a millenial time scale, phlogiston disappeared in a flash, to
be replaced by oxygen; on a time scale that could look at day-to-day, week-
 126 SAUL CORN

to-week, and month-to-month changes, phlogiston died hard, and the birth
of oxygen was difficult.
Kuhn’s analysis is restricted to those knowledge-oriented disciplines
called the sciences and, more specifically, the empirical sciences and espe¬
cially the physical sciences. Machlup has pointed out that for thousands of
years, the word science referred to the absolutely certain, such as mathe¬
matics and logic and metaphysics but excluded natural philosophy; and that
it was only in the nineteenth century when an empirical base became a
requirement for science that the tables were turned to drive metaphysics
out. [Machlup, 1980.] Machlup points out, however, that in languages other
than English, the word science still includes other disciplines (even law); but
I note that all his examples of the term in other languages are what I have
called knowledge-oriented as distinguished from action-oriented. It seems to
me that even the action-oriented disciplines develop their characteristic con¬
cepts, ideologies, Weltanschauungen, and what I have elsewhere called
mass egos, in a similar way. I will return to this question shortly. Mean¬
while, let me discuss this attitude toward knowledge implied by those con¬
cerned with its accumulation.
Those fond of physical metaphors in psychology and epistemology often
compare knowledge to physical entities in order to invent methods for
weighing and measuring it. They often visualize knowledge as an inert quan¬
tity of stuff independent of the knower. (I believe it was Karl Popper who
criticized and sharpened this point of view. [Popper, 1968.]) Knowledge is
en/Zg/ztenment, and those fond of physical imagery, in talking about it,
should remember that light was not fully understood in the physical sciences
until Huyghens’s wave theory was developed as well as Newton’s particle
theory.
Charles S. Peirce pointed out that the study of signs (symbol systems) was
not merely about a dyadic relation between the symbol and what it means
but was really discussing the triadic relation among symbols, their meanings,
and their users or interpreters. (I will come back to this point when I discuss
the ideology of informatics.) In a similar way, knowledge, which is what we
try to symbolize when we communicate, is not merely a relation between
what is known and the knower; knowledge is the result of an attempt to
communicate and, therefore, involves a community of knowers as well as
what is known. This communication process works like a boundary-value
problem among the knowers of the community, and the things and processes
the community knows in common result from something like a wave phe¬
nomenon. When the concepts of cardinal number, length, area, volume,
mass, weight, temperature, energy, element, or cell, or the processes of
which they partake, were invented (not discovered!), this was a production
of new knowledge that occurred over a period of time. And this social
process was what I called a wave of agreement to perceive. This is the
process that some historians of science call discovery, and whose retraction
and leplacement Thomas Kuhn calls the change of paradigm in a scientific
revolution. [T. S. Kuhn, 1962a.]
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 127

Furthermore, the continued transmission of these social agreements to a

larger community is the part of the wave of agreement to perceive that we
commonly call education, although some call it brainwashing.
It seems to me that this process occurs in all disciplines, not just in the
sciences in a restricted sense. But before I go into this question, I want to
consider why the division of labor among us has separated knowledge-ori¬
ented disciplines from action-oriented ones. Why did understanding divorce
itself from purposefulness, knowledge from power? To use the cybernetic
terms, why did communication separate itself from control? In Schopen¬
hauer’s words, why is the world as “idea” distinguished from the world as
“will”? How did aesthetics come to separate itself from ethics, the true and
the beautiful from the good?
We have all noticed squirrels alternating active skittering around with
passive, quiet observation. The human animal seems to be the only one that
has specialized these two states. Perhaps neurophysiologists can tell us if it
is characteristic of the chordata phyllum to alternate receptor-signals going
to the brain with effector-signals coming from it along the spinal cord. But it
would take sociological or anthropological information to tell us why West¬
ern civilization used this dichotomy in its division of labor among those
professing action-oriented and knowledge-oriented disciplines. Why did
Plato consider the question of the usefulness of knowledge demeaning? Why
have science and technology become so divorced when the bases of both are
so interlaced, as we have already observed? What were the reasons for the
sharp separation of pure and applied mathematics, especially in the twen¬
tieth century? Why are the words academic and practical used as mutual
insults? Although I agree with operational philosophers like Anatol Rapo-
port that this separation has had a bad effect, the social phenomenon in¬
volved still needs explaining. [Rapoport, 1953 and 1967.]
In the case of those disciplines sophisticated enough to have developed
formal symbolic systems, such as mathematics and physics, I offer a partial
answer by observing what is required of sentences in their languages, known
as laws, ideal laws, logical laws, or analytic sentences in mathematics and
material laws, laws of nature, or synthetic sentences in the physical sci¬
ences, as observed by the logical positivists. [Carnap, 1939 and 1969.] In any
case, they are declarative sentences whose verbs appear in the indicative
mood. They must be formed using nouns and verbs for the basic things and
relations or processes that are either precisely what those communities de¬
cided to be their basic percepts, or else linguistically constructed phrases
and clauses that use them in accepted syntactic structures.
The additional concepts must be defined in terms of the basic ones. Carl
Hempel tells us that these definitions are rarely simply nominal or real
definitions but often employ what Carnap has called reduction to an expe¬
riential basis. This means that the definition is not simply the presentation of
a necessary and sufficient condition for recognizing the new concept but is
preceded by a special condition or is accompanied by special measurement
requirements. [Hempel, 1939 and 1969.] It seems to me that these conditions
128 SAUL GORN

and measurements are concerned precisely with specifications, outside the

immediate range of human perceptions, that the discipline has decided to
extend itself to. The only syntactically acceptable sentences in these lan¬
guages are sentences that can be verified as true or tested for falsity by
formal logical or standard empirical means. The results are then, first of all,
strictly descriptive because they are declarative—hence, knowledge-ori¬
ented; second, their acceptance criteria must be independent of the variety
of representation systems (or even languages) they are presented in. For
example, even as simple a statement in mathematics as the arithmetic one
that 7 + 5 = 12 must have its truth value independent of the notational
language for numbers or the addition procedure used. (The puzzle if 7 + 5 =
13, what is 6 + 8 is not really a mathematical one.) Similarly, the sentences
in physics expressing the motions of a system of, say, twenty bodies or
whatever number is used to model the solar system, must be equally valid
when translated into another coordinate system. In fact, Einstein used the
Tensor notation precisely because a grammatical sentence in it could not be
physical nonsense as a result of being essentially dependent on which coor¬
dinate system it appears in. A physical law should be the same indepen¬
dently of the language used to describe it! This is a guarantee of objectivity. I
submit that it was the quest for objective truth, independent of the knowers,
that caused the dichotomization of knowledge and action. [J. Dewey, 1929.]
Thus, the laws of mathematics and physics must be independent of their
users or interpreters. To use Charles Morris’s interpretation of Peirce’s
semiotics that I have already mentioned, the symbology of these disciplines
must be independent of pragmatics, and possibly even formal. [Morris, 1939
and 1969.]
The very sophistication in the language of the discipline that allows one to
define composite concepts and composite processes from the basic ones by
some standard assembly method, whether definitions be nominal or real—
(or explicatory in Hempel’s sense)—refocuses one’s attention from the
things and processes considered by the discipline to the symbolic expres¬
sions and symbolic manipulations needed to express them. One is now be¬
coming involved with the metalanguage of the language of the discipline;
that is, one is now not merely talking in the language of the discipline, one is
talking about its grammar as well. The action orientation is now more sym¬
bolic than substantive, and the discipline as a whole appears more knowl¬
edge-oriented. [See Hermann Hesse on the Glass Bead Game, 1949.]
The model I am presenting of the ideological development of disciplines
has these disciplines building up special symbolic languages. At their birth,
and during revolutions of the Kuhnian type, a number of basic expressions
are introduced or replaced; expressions referring to processes the commu¬
nity has agreed to perceive are the basic verbs of the language. Expressions
referring to the things the community has agreed to perceive are the basic
nouns of the language. Definitional habits developed by the community per¬
mit it to specify composite processes and composite objects to yield phrases
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 129

and clauses or new nouns and verbs of the language. Communication habits
developed by the community make its members build these syntactic types
into sentences and assemblages of sentences.
When the discipline is mainly knowledge-oriented, as in the academic
type, the tendency seems to be to have the majority of the sentences in the
declarative mood; furthermore, the manner of assembling these sentences is
often in deductive or inductive groups, each discipline having its own
method of reasoning or argument. For example, mathematics and physics
use formal and extensional logic, and the assemblages are definitions and
proofs and demonstrations.
When, however, the discipline is mainly action-oriented, the majority of
the sentences tend to be imperative, hence, neither true nor false (though
possibly good or bad). Moreover, the important assemblages are assign¬
ments, allocations, and programs. Assembly grammar is not logic; it uses the
methods of programming. Communication is mainly in command languages,
as in industrial orders, game rules and protocols, recipes, military com¬
mands, or governmental programs. The language is that of management, and
the sentences are like legal laws, not natural laws or valid truths. The basic
assumption in the action-oriented disciplines seems to be free will, while the
basic assumption in the empirical sciences seems to be determinism. I sub¬
mit that these are presenting laws of nature but belong to the metalanguage
of the discipline. They are decisions on how their statements are to be
viewed and organized in the language.
Each of the two extreme types of discipline has a few sentences belonging
to the other type. Management peppers its prescriptive commands with
descriptions of the conditions for alternative actions or initial states or the
goal to be achieved. Theoreticians pepper their extensive descriptions of
the states of their worlds with verification procedures (specified often in the
metalanguage they use in discussing the methods).
Scientists of the operational type demand a better balance. But it is the
disciplines called professions, whose mission is to transfer knowledge into
action, that employ the indicative and the imperative in more equal doses.
(But note that even medicine has its surgeons and internists; law, its prac¬
titioners and reviewers.)
I have elsewhere detailed the development of disciplines into phases that
mainly describe the complexity of the expressions in their languages from
the primitive noun system (taxonomic) to the sophisticated metalinguistic
expressions (formal deductive or simulative). [Gorn, 1965.] My claim has
been that an advanced stage in the development of a discipline’s language
and symbols makes it develop a contact with mathematics, and an even
more sophisticated—in fact, self-conscious—stage brings it into contact
with informatics. However, before I consider the ideology of informatics
that causes this relation to other disciplines, I should say something about
what I have called the mass ego of a discipline.
The claim that every discipline whose language is sufficiently sophis-
130 SAUL CORN

ticated (in fact self-conscious) establishes contact with informatics (and

many such with mathematics) is not, I feel, merely personal; it is a claim that
I believe any informatician would make for reasons I will explain shortly.
This is an example of what I would call the mass ego of informatics. Every
fairly advanced discipline regards itself as a critically important factor with¬
out which the warp and woof of the fabric of human knowledge and action
would unravel. Its participants feel that the concepts they have agreed to
perceive cover a certain set of dimensions of human perception that no other
discipline can do justice to and that all others depend on. Furthermore, it is
with its closest neighbors in the topology of disciplines that it is likely to
have the severest ideological arguments; after all, it is almost impossible for
a microscope to be kept in focus for overlapping views simultaneously. The
informatician is no different from the others in these respects, as we shall
see.

THE INFORMATIC OUTLOOK

Informatics is concerned with symbolic expressions and their manipula¬

tions. Such symbol manipulation, especially in numerical and analytic com¬
putations, has been changing slowly over the millenia. Since the Renais¬
sance, and especially after the industrial revolution, it spread over a number
of disciplines and accelerated. It was, however, the amplification of this
phenomenon of the spreading and changing of symbol manipulations, caused
by the appearance of electronic computers, that made the process extremely
apparent, because its pace had become visible on a day-to-day basis.
Such symbolic manipulation had already transcended numerical computa¬
tion in the calculus, with formal differentiation and integration, and in sym¬
bolic logic, especially because of the school of formalists at the turn of the
century, but even with Aristotle when he discussed syllogisms.
When, however, the electronic speeds of the new computers required
decision-making instructions and a programming process separated from the
time of computation and when the necessary program occupied the same
storage as the numerical data, it became obvious that most of the program¬
ming process itself was just another symbol manipulation that the same
machine could accomplish. It then became clear that numerical computation
was a very special example of a much broader concept of computation, one
that included the linguistic process of programming. Furthermore, the com¬
mon storage of instructions and data made the language of programming
capable of self-referencing operations.1 Programs could possess some of the

'I have called this property of some machines and their languages unstratified control. The
metalanguage of programming is contained in the object language. Hofstadter calls this property
“a strange loop.” [Hofstadter, 1979.]
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 131

self-referencing subtleties we had thought to be characteristic of speech and

thought, and, hence, what we previously thought to be unmechanizable.
The result was not only the design of more sophisticated programming
languages more suited to human use but the design of programs of the type
now known as possessing artificial intelligence. In fact, some informaticians
are trying to make natural language reveal more mechanism, and some keep
trying to make mechanical languages more natural.
Also any instruction, and, hence, any set of instructions, could either be
wired into the circuitry or placed in the storage. The difference between
hardware and software disappeared as a logical distinction—as the very
peculiar words hardware and software indicate.
So far, everything I have said about computation is action-oriented and
tied to the computer. But prior to the appearance of digital computers, a
theory of computation had already appeared. [Kleene, 1979.] Symbolic logi¬
cians had already investigated the logical limits of computation; Turing had
analyzed computation, designed the universal Turing machine, and proved
the unsolvability of the halting problem; Godel had shown the limits of
formalism with his undecidability theorems; Church, Kleene, and Curry had
analyzed computation in recursive-function theory and combinatory logic;
Thue and Post, and, most recently, Markov, had presented general compu¬
tation as word problems; and most important, all these approaches were
shown to be equivalent. Thus, when computers appeared, Noam Chomsky’s
discussion of natural linguistics was in terms of computation, and a new
computational view of psycholinguistics was introduced. [Chomsky, 1963;
Chomsky and Miller, 1963; Miller and Chomsky, 1963; and Fodor, 1980.]
This new development produced linguistic descriptions of programming pro¬
cesses, a mathematical automata theory, and formal languages. These, in
turn, affected the design of programming languages and machines. In short,
the presence of the machines amplified and made tangibly and visibly appar¬
ent what had been a slow process of development in symbol manipulation
before their appearance.
We now perceive informatics as concerned with the synthesis and analy¬
sis of systems of symbolic expressions and the synthesis and analysis of
processors that interpret, translate, and manipulate such expressions. More
prosaically stated, informatics is concerned with the study, design, and use
of data structures and their transformation by mechanical means, as a recent
text has put it. [Tremblay and Sorenson, 1976.]
In its present youthful stage, this new discipline covers the art, science,
and profession of symbol manipulation. The action-oriented and the knowl¬
edge-oriented aspects of computation are less likely to move apart, because
the description of symbolic expressions remains tied to the prescriptive
language of its interpreting processors. In other words, the knowledge-
oriented concept of data structures and the action-oriented concept of data-
processors remained harnessed instead of going through the usual polar¬
ization. I prefer to speak of mechanical languages and their interpreting
132 SAUL CORN

processors rather than data-structure systems and data-processors. And the

new concept of processors that the informaticians have agreed to perceive in
a new ontological commitment is, as I see it, a blending together into one
entity of mechanisms (hardware), programs (software), and rigidly con¬
trolled human procedures (in symbol manipulation). The assembly of a pro¬
cessor can be performed equally well in assembly language by using as
software the program that is known as the assembler, or by wiring machines
together (whether big or small—chips in fact), and only our decision as to
which method happens to be more efficient determines the choice. This is
what I have called the principle of logical equivalence of hardware and
software; I believe that it is part of the peculiar ideology of informatics.
The concept of assembly in informatic systems refers equally to such
activities as splicing programs, wiring mechanisms together, calling pro¬
grams that are stored in mechanisms, or having people push buttons, punch
cards, or type instructions and follow procedures specified explicitly by
machine outputs. These combinations of machines, programs, and algorith¬
mized people are the “users and interpreters” of symbolic expressions that
Charles S. Peirce considered to be the third element in the triadic relation
evoked by symbolism. In the study of symbols that Charles Morris called
semiotics, he referred to the relations of symbolic expressions to their users
and interpreters as pragmatics. [Morris, 1939 and 1969.]
Informatics must concern itself precisely with these pragmatic questions,
where, as we have seen, mathematics and the physical sciences must be
independent of them, certainly in their products if not in their methods. In
this respect, informatics is more like linguistics, psychology, the behavioral
sciences, philosophy, and the professions. This pragmatic tie between sym¬
bolic expressions and the design of symbol manipulators, the processors
forming the basic concept that informaticians have agreed to perceive, is
likely to keep both the science and technology harnessed together, as I have
already remarked.
The study of relations among, and operations with, symbolic expressions
independent of their semantic content or pragmatic context is called syntac¬
tics. It is in the study of syntactic questions that informatics is most formal
and closely related to mathematics and its methods. The study of relations
between symbolic expressions and their meanings, independent of their
method of use or interpretation, is called semantics.
I have stressed from the beginning of this paper that the different disci¬
plines are distinguished from one another, especially if they are sophis¬
ticated enough to have well-developed symbolic systems, by having ideolog¬
ically distinct types of semantic concepts. The very fact, however, that they
have symbolisms with well-defined manipulative procedures may make it
useful for them to simulate their processes symbolically. It is one of the
informatician’s basic insights (prejudices, superstitions, ideologic attitudes,
or peculiar ways of looking at the world?) that any process that can be
precisely specified is capable of being symbolically simulated, because a
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 133

precise specification is already a symbolic simulation. Notice that the sym¬

bol system need not be numbers; it could consist of analytical expressions
from mathematics or physics; formulae in one, two, or three dimensions
from chemistry; graphic representations of polypeptide chains; double-helix
representations of genes; graphic representations of communication sys¬
tems; symphonic scores; choreographic scores in laban-notation; or even
animated motion pictures; wiring diagrams; specifications for chip designs in
microprocessors; architectural drawings in animated graphics; photographs
for pattern recognition in artificial intelligence; or actual people being
viewed and heard. And the symbolic medium could be holes punched in
cards, electronic signals, magnetic signals, visual signals, sounds, even
speech, or whatever. Adding receptors and effectors simulating or going
beyond human ranges to the input/output devices could extend applications
beyond symbol manipulation to be fully cybernetic in the design of robots.
This symbolic mass ego of informatics is not a schizophrenic dream. Most
sophisticated disciplines, even action-oriented ones concerned with games,
agree that the informatician has this useful power and maintain contact with
him or her in spite of the fact that their own semantics, when they use their
symbols, are different from those of the informatician.
The informatician’s machines may deal with real semantics if they have
the appropriate sensors and effectors, but the informatician can always deal
with the real semantics of symbol manipulation by using the unstratified
control in his or her machines. This difference between semantics in differ¬
ent disciplines can create confusing arguments, especially when, as we have
seen, the concepts they symbolize are closely related.
It is not difficult to distinguish the biologist’s meaning of the English word
function from the mathematician’s, but see how different our interpretation
of the analytical expression 1 — Vigt2 is when we are thinking like physicists,
mathematicians, or informaticians. In the first case, t might mean time and g
the acceleration of gravity, and the formula might represent the trajectory of
an object being dropped from an elevated moving vehicle—a physical pro¬
cess. To the mathematician, on being told that g = 32.2, symbol / is almost
meaningless, and any other free variable would serve as well. On being told
that its domain is real numbers, he or she sees not a physical trajectory but a
static set of ordered pairs of numbers that is all there at once; he or she might
visualize it as a graph, but a static one, and not a moving point. Further¬
more, the formula (1 — \/~Vig r)(l + \f Vig t) would be to him or her, another
representation of exactly the same function. Finally, the informatician
would see the two formulae as crude descriptions of two different computa¬
tional processes that would only yield the same value for the same value of t
under very special representations of f, to the informatician, real time is
represented not by t, but the access of the value of t from storage and the
time of interpretive computation—in this case, calculation.
Having run through the syntactic, semantic, and pragmatic aspects of the
symbol-manipulating activity with which informatics is concerned, we can
134 SAUL CORN

now see how different people in the field were able to stress different kinds
of activity.
The formal theoretical aspects close to logic and mathematics cover such
topics as the theory of computation, the analysis of algorithms, switching
theory, automata theory [Hartmanis, 1979], the study of formal languages
[Greibach, 1979], the study of data structures, and the syntactic representa¬
tion of semantics. Among these topics, one learns what can and what cannot
be computed and for those problems that can be computed, what varieties
demand what kinds of computing rhythms and storage controls and at what
cost. One also learns the capabilities of various programming techniques,
the symbol-manipulative aspects of formal logic and theorem-proving, and
the formal techniques of program verification. [Ibid.]
The informatic activity that is empirical and experimental rather than
formal turns up in modeling human symbolic behavior in heuristic searching
and problem-solving. [See Newell and Simon, 1976a.] Whenever theory
informs us that some class of problem is unsolvable, we immediately search
for subclasses of problems that are solvable and do so by experimenting with
new solution methods until our theoreticians can prove to us that the
methods we find are obvious, limited, or fallacious. Even when theory in¬
forms us that a general-solution method exists but is very costly, we thrash
around for heuristic procedures that may fail to be universally successful
(these are called algorithms) but are likely to succeed much sooner. Much of
this modeling of human problem-solving behavior that remains empirical is
the subject of the area called artificial intelligence. [See also Amarel, 1976;
Newell and Simon, 1976a; Salton, 1976; Simon, 1976; and Pearson and
Slamecka, 1976.] This part of informatics is the part that is applied to the
area of engineering called robotics.
This empirical side of informatics, together with the theoretical side called
analysis of algorithms, is the area some call a major portion of informatics; it
is known as the design of algorithms. If we note that these procedures called
algorithms have equivalent representations in software and hardware, we
recognize our basic concept of processor again; this area is, therefore, what
I called the synthesis and analysis of interpreting processors. However, we
should not forget that in addition to the algorithms, there are devices that
feed them and collect their results, which are similar to human receptors and
effectors. These input and output devices can extend the semantics of the
symbol systems being processed to the various extensions of our senses and
motor controls needed in robotics. Thus, the various studies in graphics,
vision, and speech recognition, and their use in image processing and pattern
recognition, are also extensive portions of empirical informatics.
So far, we have examined the areas of informatics that are most scientific
in the recent English sense of the word. In both the theoretical and empirical
portions, people are concerned with truth—the truth about symbol manipu¬
lation. But many are engaged in informatic activity because of aesthetic or
ethical reactions. Some are attracted to the beauty, and some, especially
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 135

those with professional interests in applications, are interested in the good

things that can be achieved and the further increase of our powers, while
others are appalled by the stupid or evil applications that are possible.
[Weizenbaum, 1976.] The information-handling machines have been called
intelligence amplifiers. It is true: They are, but they are also stupidity ampli¬
fiers. They are symbol-manipulation amplifiers that can be used for good or
ill.
Knuth’s Turing lecture presented the aesthetic view. He talked about
beautiful and clearly understandable programs. [Knuth, 1974a.] And we can
readily imagine that most of the designers of processors, hardware, soft¬
ware, or systems may react aesthetically, especially now that microproces¬
sor chips can be designed for production by automatic VLSI (very-large-
scale integration) techniques and produced without exorbitant capital
investment (unlike cyclotrons). The ranks of hardware designers may be¬
come almost as populous as those of the software engineers and, in fact, may
largely overlap them. Because of the principle of logical equivalence of
hardware and software, the machine designer and programmer did not have
to have different aptitudes. Now it will be much more usual for them to be
one and the same person.
These aesthetic and ethical views of professional aspects of informatics
have been there from the beginning, connected first with formal aspects of
logic design, then with structured programming rules for software engineers
[e.g., Dijkstra, 1976], and finally with the variety and value of the applica¬
tions, especially as they begin to transcend the symbol-manipulative.
It is, however, this area of applications to all other disciplines that should
absorb the vast majority of professional informaticians. These professionals
would apply their knowledge of informatics to the production of programs,
programming aids, and programming systems—or their hardware equiva¬
lents—for the other disciplines. For example, these professionals might de¬
sign processors called simulators—psycholinguistic simulators, economic-
system simulators, traffic simulators, management-information systems,
military-information systems, science-information systems, report genera¬
tors, inventory-control systems, and so forth.
In any case, the application to the other discipline would require that its
own personnel and informaticians dredge up the basic concepts from the
mass ego of that discipline and model its language syntactically in its sim¬
ulators. The model might even be semantically based if the appropriate
receptors and effectors are available, as they might be in robotics.
This sums up my picture of the ideology of informatics and its practition¬
ers, whether they behave as formal theorists (such as pure mathematicians),
empirical scientists, or professionals with aesthetic or ethical goals in a
variety of applications. I believe it is the ontological commitment deeply
entrenched in this ideology that keeps knowledge-oriented and action-ori¬
ented behaviors harnessed together and makes the complete separation of
pure informatics from applied informatics less likely. However, the variety
136 SAUL CORN

of applications is so tremendous, even more than in the case of mathematics,

because of the pragmatic aspects of the subject, that it is quite likely that a
number of distinct professions will spin off, whereas mathematics only pro¬
duced an amorphous community of applied mathematicians.
This leads me to consider community relations between informatics and
other disciplines and especially its immediate neighbors.

THE INTERCOMMUNICATION AND STABILITY OF DISCIPLINES

I remarked early in this paper that the word ideology has acquired a number
of meanings; I have been using two of those meanings. The first refers to a
systematic body of concepts, especially as applied to a manner or the con¬
tent of thinking characteristic of an individual, group, or culture; in our case,
the relevant groups or cultures are called disciplines, and the body of con¬
cepts includes what Thomas Kuhn calls paradigms. The second meaning is
that of Destutt de Tracy2 and the ideologues [Picavet, 1891 and 1971] around
the time of Napoleon. This meaning was the study of the development of
ideas from sensations, especially in humans; the ideologues intended ideol¬
ogy to be a branch of zoology, specifically of human biology, particularly
neurology and psychology. Destutt de Tracy’s [1817] book on ideology ac¬
knowledges descent from Condillac and has a second part on grammar and a
third part on logic. Clearly, he would have included today’s psycholinguists
and such developmental psychologists as Piaget among the ideologues. He
believed, for example, that ideas were initiated by direct perception and
extended by inner perceptions called memory, judgments (the comparison of
outer perceptions with memories), and will (the perception of desires and
action caused by this perception).
Our description of informatics and its relation to other disciplines, in fact,
of the generation of ideologies of disciplines and their Kuhnian transforma¬
tions, extends the basis in perception that the ideologues insisted on to a
basis in the extended perceivers and controllers beyond the human spec¬
trum, one that is provided by technology. In the case of informatics, it is the
electronic computer that provides this extension, just as it is the accelerator
in modern physics, the electron microscope in material science and biology,
and similar devices and techniques in genetics and anthropology and as it
was the microscope and telescope in the time of Galileo.
A true unification of human knowledge (such as the sciences) and action
(such as the arts and professions) would call for the ability to patch the
variety of ideologies together when they have been generated by extensions
in many directions beyond the human range. Such patchwork is going on in a

21 am indebted to Bertrand de Jouvenel for pointing out to me that my method of identification

of informatics was ideological in this sense.
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 137

large variety of interdisciplines, such as biophysics, biochemistry, sociobiol¬

ogy, ecology, bionics, the aforementioned psycholinguistics; and in such
superdisciplines as cybernetics, history, anthropology, philosophy, and
pedagogy.
This process of mapping all of human lore has been going on since the
urban revolution, and the system of disciplines has moved around and
changed kaleidoscopically. We have been filling in and constantly modifying
this map. I believe this to be a more realistic model than the nineteenth-
century model of science progressing by indefinite accretion; such an
infinitely progressing model is obviously unstable, even apart from the
criticisms of Thomas Kuhn and Karl Popper. Historians, philosophers, and
sociologists of science would mainly concern themselves with our first
meaning of ideology but would also be concerned with Destutt de Tracy’s
meaning in order to understand the way the first develops.
Stable disciplines have distinct ideologies, and we can, therefore, distin¬
guish them from one another by contrasting those ideologies. For instance,
in my presentation of the ideology of informatics, I showed that it was not a
branch of mathematics by, for example, pointing out that it must concern
itself with pragmatic questions where mathematical concepts must be inde¬
pendent of them. Another way of distinguishing disciplines is by examining
their proportions of descriptive, prescriptive, and professional activities
(those translating knowledge into action).
Let us use these methods to discuss some of the variety of neighboring
disciplines and subject areas, specifically cybernetics, robotics, communica¬
tion theory, library science, information retrieval, information science, man¬
agement-information systems, decision science, and management science.
As Norbert Wiener defined it, cybernetics is the study of “communica¬
tion and control in the animal and the machine”; I have already called it a
superdiscipline. Wiener intended cybernetics to include neurology, com¬
munication theory, control theory, what is now called general systems
theory, and bionics. Informatics, I think, is included because it concerns
itself with communication by explicitly designed symbolic systems and the
control of such communication by mechanisms, procedures, algorithmized
people, or suitably assembled systems of these (what I have called pro¬
cessors). If input/output devices attached to computers are extended to
include more general receptors and effectors, the resulting extension of the
field would include robotics and also be a part of cybernetics. However,
although robotics makes use of the area of informatics known as artificial
intelligence, it is clearly involved with concepts and principles of the sci¬
ences related to extended perceptions provided by the receptors and effec¬
tors used, and robotics is pragmatically concerned with actions needed in
processes for which robots are designed. In short, robotics is mainly con¬
cerned with translating from the knowledge of a number of disciplines (infor¬
matics included) into action; it is a professional activity. Whether robotics
138 SAUL CORN

will develop a unified ideology, and, therefore, become a single discipline,

will be determined by its practitioners. If its members set up a tight com¬
munication system and decide that their new recruits need a standardized
curriculum in their education, robotics will, indeed, become a separate disci¬
pline. Cybernetics is likely to contain a number of disciplines, and informat¬
ics is likely to have a number of such professional disciplines developing
from its applications. The areas variously called decision science, management-
information systems, medical-information systems, and management sci¬
ence are other examples of professions of the cybernetic kind neighboring on
informatics. If the word science in some of these titles remains attached to
them, even if only as a convenience in obtaining economic supports, then
that word science will have gone through still another stage in its evolution. I
would tend to call these disciplines professions rather than sciences, be¬
cause their emphasis is on system design rather than being strictly knowl¬
edge-oriented. That is my interpretation of the present meanings of the
words science and profession. Thus, the designers of medical-information
systems would be applying knowledge of medicine and informatics to a
particular management area.
The professional activity of the pedagogues, in this country known as the
field of education, is another example of what I have called a superdisci¬
pline. Each discipline, in order to educate its newcomers, develops a com¬
plicated curriculum in order to imprint its ideology. I have heard some talk in
the past of having universal pedagogic techniques independent of the subject
matter taught. I am skeptical, for pragmatic reasons, that such tools can be
sufficient. Distinct ideologies requiring, as they do, different habits of
thought must, it seems to me, call for different pedagogic techniques.
The pedagogue must be able to recognize when a pupil understands a
subject and to generate such understanding when it does not exist. He or she
must, therefore, have a technique to elicit activity from the student, either to
reveal the state of the student’s understanding or to generate such. Different
ideologies of different disciplines demand different techniques, and the
pedagogue who introduces one must be steeped in it himself or herself. One
cannot expect the majority of pedagogic techniques, which must ipso facto
be pragmatic to suit human users and interpreters, to be pragmatically inde¬
pendent of the ideologies of the disciplines taught.
It seems to me that the area of information retrieval makes a similar
pragmatic demand. In order to retrieve relevant information, one must
understand the structure in which it is stored—usually in accordance with
the ideology of the storer—and compare it to the ideology of the purpose for
which it is to be retrieved. Hence, I believe that this pragmatic dependence
on the ideology of the subject applies not merely to the superdiscipline called
pedagogy, but also to that most ancient profession of library management
and the area related to informatics known as information retrieval.
Because the different disciplines of knowledge and action will keep
INFORMATICS: IDEOLOGY, METHODOLOGY, SOCIOLOGY 139

changing their scope, knowledge about each discipline will never be needed
in the same order or in exactly the same language in which it was produced.
Thus, the organization of disciplinary information as it is produced will
never be the same as the appropriate organization needed to recall it effi¬
ciently for future use. It follows that the organization of the library will
always need changing and, hence, the problems of library organization will
never be solved once and for all. Thus, the library profession will never be
static enough to be totally mechanizable. Each generation of librarians may
suffer a certain amount of technological obsolescence, but the profession as
a whole never can. Since its main problem is unsolvable, it is guaranteed to
be a stable profession.
In fact one can say, conversely, that only those professions will be stable
and, at long range, safe from technological obsolescence whose problems
are unsolvable. For example, physicians, whose main duty is to keep us
from dying, will always be with us. Similarly for teachers, whose job it is to
keep mankind from becoming stupid; or mass-communication professionals,
who must keep us informed.
It is attempts to solve new problems, especially the unsolvable ones, that
are responsible for establishing new disciplines and their ideologies, espe¬
cially the stable ones. The application of computers to these problems is,
therefore, likely to present symbolic simulations of their pragmatic de¬
mands, hence, of their ideologies. I would therefore expect a number of
professions to keep spinning off from informatics.
As a young discipline, however, theoretical aspects are still closely con¬
nected with the action-oriented design and handling of computational and
linguistic mechanisms. Fundamental results from the mathematical theory of
computation present a host of provably unsolvable problems of symbol ma¬
nipulation and a host of impracticable solutions to overly general problems
together with proofs that they cannot be improved. They mark the limits of
computation that every professional informatician must be aware of, as well
as the unsolvable areas that can be conquered only by subdivision. This
essential tie of the knowledge-oriented base to action-oriented aspects and
the consequent empirical activity may succeed in keeping informatics from
separating into distinct knowledge-oriented disciplines, the way mathemat¬
ics did after the Pythagoreans and Plato; grammar, after the Stoics; and
logic, after the logistic and formalist schools. This fact, coupled with the
host of unsolvable problems, may maintain informatics as a stable discipline
without its unravelling into separate sciences, arts, and professions. If such
is the case, informatics will not become a metascience, as did the just-
mentioned disciplines, or, for that matter, metaphysics.
In short, we should not try to separate computer science from information
science but should try to maintain one discipline of informatics. Any attempt
to cause such a separation by, for example, trying to create another metasci¬
ence [see Otten and Debons, 1970] would separate the action from the
140 SAUL CORN

knowledge, as has happened with Pythagorean mathematics, Sophistic

rhetoric, Aristotelian metaphysics and the Aristotelian organon. Stoic gram¬
mar, and the logic and grammar of the logical positivists. Such a separation
would cause the cessation of the very effervescence that the mixture of
knowledge and action maintains. In the case of the library profession, even
nonpragmatists and extreme idealists would agree that this is the wrong way
to go. [See Wright, 1979.]3

3For further insights, see Furth [1974], Heilprin [1974], D. Hillman [1977], Kochen [1974],
Otten [1974], Sager [1977], Slamecka and Pearson [1977], Whittemore and Yovits [1974], and
Yovits, Rose, and Abilock [1977],
 PERSPECTIVES ON INFORMATICS
AS A SEMIOTIC DISCIPLINE

Charls Pearson and Vladimir Slamecka

The purpose of this paper is to analyze the logical and methodological rela¬
tions between information science and computer science and some of the
other subject areas that contribute to the scientific study of information.
Gorn proceeds toward this destination in two stages. First, he develops two
main theses: (1) Informatics (encompassing information and computer sci¬
ence, engineering, and technology) is a semiotic discipline; and (2) informa¬
tion science and information engineering should not and perhaps cannot be
separated very much. These theses provide a perspective that is then used to
analyze the relation between informatics and several neighboring disci¬
plines.
The gist of our comments may be summarized as follows. We are in full
agreement with the first thesis—that informatics is a semiotic discipline.
Given this perspective, we also fully concur that most of the neighboring
disciplines (including robotics, cybernetics, what is currently called informa¬
tion science, etc.) are professions or technologies rather than sciences and
that most uses of the word science in this area are honorific rather than
scientific. On the other hand, we are not convinced that information science
and informatic engineering should never be separated and that the honorific
use of the term science by information professions and technologies will
necessarily persist.
Our sympathetic regard for viewing informatics as a semiotic discipline is
easily explained: This view has been the common denominator of the aca¬
demic programs of Georgia Tech’s School of Information and Computer
Science since its establishment in 1964, and a major if not central thrust of its
research programs. [Slamecka and Gehl, 1978.] Thus, Gorn’s and our views
have largely coincided for nearly two decades [see Gorn, 1963; Slamecka,
1968]; and our research interests contributed to several of the topics men¬
tioned in his paper, notably the theorem of hardware-software equivalence
[Poore, Baralt-Torrijos, and Chiaraviglio, 1971], logic of plans and programs
[Baralt-Torrijos, 1973], and particularly empirical semiotics [Pearson,
1977]. Our comments on Gorn’s first thesis are thus largely an amplification
141
142 CHARLS PEARSON AND VLADIMIR SLAMECKA

of his views, as well as an argument for a broader interpretation of the

domain of semiotics relevant to information science.

ON THE SEMIOTIC NATURE OF INFORMATICS

Discussions of semiotics are often buried in linguistic analyses; for this

reason, we first discuss the semiotic background and then comment from its
perspective on several positions in Gorn’s paper. In the following develop¬
ment of the topic, we make use of a language developed specifically for use
in the empirical analysis of meaning and information, called the Language of
Menetics. [Pearson, 1977.] Much of the origins of this language stems from
Charles Peirce, Charles K. Ogden and Ivor A. Richards, Rudolf Carnap, and
Charles Morris.
In order to perform an information experiment, it is necessary to control
all known variables and sources of variation for the purpose of analyzing
unknown sources of variation and the relations among them. This implies
the idealization and simplification of an information situation into its minimal
atomic units and the selection—or design—of one such unit for observation
or study. To use an example from physics, it is possible to measure the heat
equivalent of energy by shaking the Golden Gate Bridge with giant machines
and measuring the temperature rise across the span. This would be a most
inefficient way of carrying out such an experiment and unthinkable, of
course, in a modern physics laboratory. If we are interested in better under¬
standing (Gorn’s knowledge-orientation) the relation between mechanical
energy and thermal energy, we would idealize the situation down to a single
controllable and analyzable source of mechanical energy, boundary condi¬
tions that control the flow of heat, and a means for measuring temperature
rise. Similarly, in the study of information, we must simplify our messages
and communication situations—the things the information community has
agreed to perceive—down to their minimal atomic units and assemble them
in simple, controllable, and analyzable configurations, so as to be able to
study the unknown sources of variation among them and the relations among
these sources.
In the information and communication situation, the minimal atomic ele¬
ments are called signs. Thus, signs are the elementary carriers of meaning
and information, and the basic science of information concerns the structure
of signs—how they carry information and meaning and how they are pro¬
cessed. This study has been called semiotics1 (in various forms of the word)

'The word semiotics should not, however, be attributed to the American philosopher, Charles
Morris. Morris never called the discipline semiotics; he consistently called it semiotic, as did
also the founder of modern semiotics, the American logician and philosopher, Charles Peirce.
Modern usage of the term semiotic stems from its reintroduction into philosophy by the British
philosopher John Locke. The expression semiotics came only later, and Morris himself noted
that someone else had called the discipline semiotics. The anthropologist Margaret Mead is
often given credit for coining this term.
INFORMATICS AS A SEMIOTIC DISCIPLINE 143

since the times of the pre-Socratic philosophers. Gorn is eminently correct

in his assessment of the role that semiotics plays in the study of information.
Charles Peirce divided the structure of all signs into three dimensions
dealing with (1) the medium, body, or existence of the sign; (2) the object or
designation of the sign; and (3) the interpreter, interpretation, or interpretant
of the sign. Later, Charles Morris found all signs to have these same three
dimensions and gave them their currently accepted names: syntactic, se¬
mantic, and pragmatic dimensions. Although this can be viewed as Morris’s
interpretation of Peirce, Morris developed it before he read Peirce. (Peirce’s
work was not publicly available until several years later.)2
Signs also have both an internal structure and an external structure in all
three dimensions. The external structure is observable (e.g., the shape of a
sign, its medium or embodiment, or the interpreter of a sign). The external
components of a sign are its information generators, and the various mea¬
sures of information (in the empirical sense—the term information measure
also has a distinct mathematical sense) are the empirically interesting ob¬
servable aspects of these information generators or external components.
The internal structure of a sign is theoretical rather than observable; an
example would be the intension of a sign. The internal components of a sign
are its meaning components. Extension and intension are semantic compo¬
nents of meaning. Morris, who originated the term semantics (apart from a
completely different use for the same word by Breal), defined semantics as
the study of relations between signs and their objects (this object is to be
understood here in its ideal, theoretical sense).
The division of semiotics into syntactics, semantics, and pragmatics al¬
lows us to classify the various kinds of signs. Peirce’s most famous classifi¬
cation was a three-way semantic classification into indexes, icons, and sym¬
bols. Indexes have extension and object as their only semantic structure
apart from any syntactic or pragmatic structure they may have. Icons have
extension, intension, object, and ground in their semantic structure. And
symbols have extension, intension, cognision, object, ground, and cognitive
mentellect in their semantic structure. (Peirce originally used his philosoph¬
ical categories of firstness, secondness, and thirdness to draw this distinc¬
tion.) Another well-known three-way classification of signs is a syntactic one
into tokens, types, and tones. Peirce sometimes used the terms sinsign,
legisign, and qualisign to draw this same distinction using his philosophical
categories. A final three-way classification of signs involves the pragmatic
dimension: Pragmatically, signs can be divided into rhemes, phemes, and
dolemes.
There are, thus, nine elementary kinds of signs: tokens, types, tones,
indexes, icons, symbols, rhemes, phemes, and dolemes. Since a sign must
fall into one of each of the syntactic, semantic, and pragmatic categories, the

2 It may be of interest in this regard to note that Morris was a student of George Herbert Mead,
who worked quite closely with John Dewey at the University of Chicago and that Dewey, if not
actually a student of Peirce’s, studied him quite thoroughly.
144 CHARLS PEARSON AND VLADIMIR SLAMECKA

actual structure of a sign must be a combination of three of these elementary

kinds.
This classification scheme helps us to understand the relation between
semiotics and informatics. Accordingly, in its concern with semiotic struc¬
ture, informatics is concerned with not only symbolic expressions and their
manipulations (as Gorn suggests) but with all elementary kinds of signs;
symbols are just one of the nine such kinds of semiotic structure. Along the
same line of reasoning, Gorn’s definition of semantics as the study of rela¬
tions between symbolic expressions and their meanings would seem to ne¬
glect syntactic meaning, pragmatic meaning, external semantics, and the
semantics of nonsymbolic signs (the semantics of a string quartet, for in¬
stance). Peirce and Morris both recognized their versions of semiotic as the
study of all signs, not just symbols; for instance, in one paper on aesthetics,
Morris has analyzed the meaning of icons.
Having clarified the semiotic background, we are now in a position to
comment on several other statements in Gorn’s analysis. One comment
concerns the distinction between the indicative and imperative moods. The
distinction is important for the subject at hand, because one of the distin¬
guishing features of science and technology concerns the essential role
played in science by declarative sentences and in technology by imperative
sentences. Gorn is aware of this role and of the role the attitude of the
interpreter plays in distinguishing between scientific text and technological
text (thus confirming his agreement with the basic triadic nature of informa¬
tion and the sign). Mood is the syntactic encoding of a pragmatic attitude. It
is a grammatical device that indicates the relation between the interpreter of
a sentence and his or her attitude toward the propositional content of the
sentence (e.g., indicative, imperative, interrogative, etc.).
Another related comment refers to the essential existence of nouns and
verbs. The statement that sentences in disciplines sophisticated enough to
have developed formal symbolic systems must be formed using nouns and
verbs does not indicate an essential nature of information, since nouns
and verbs are not universal semiotic categories. The existence of nouns and
verbs in all of the sentences under discussion only indicates the Indo-Euro¬
pean origins of such formal systems and the reliance of current methods of
formalization on the grammatical structure of the Indo-European languages,
especially ancient Greek and late Latin. Incidentally, current research in
semiotics or information science seems to neglect this interesting relation
between the grammatical structure of natural language and the resulting
structure of the methods of formalization adopted by various intellectual
communities within the society of that language’s speakers. On the other
hand, the fruitful analogy suggested by Gorn between languages in the
conceptual study of a discipline and coordinate systems in the mathematical
study of a discipline has been given some attention. [Pearson, 1977.]
The suggestion that the symbology of mathematics and physics is possibly
formal deserves comment for two reasons. First, the symbology of mathe-
INFORMATICS AS A SEMIOTIC DISCIPLINE 145

matics is formal because mathematics is the formalization of the syntactic

encoding of our reasoning processes. The symbology of physics, on the
other hand, can never be completely formal, because of the semantic nature
of physics and, indeed, of all the sciences—including information science.
Two other semiotic comments that share a specific relation to logic, or the
methods of reasoning, are in order. In addition to the two methods (the
deductive and inductive) of manipulating information in the style called
reasoning, listed by Gom, mention may be made of what Peirce called the
most important method of reasoning—the abductive method, now more
often called retroductive. This is the method of inventing hypotheses that, if
true, would explain some known results. Abduction is absolutely indispens¬
able for investigation in all disciplines called nomological sciences.
Gorn’s insightful comparison of the action-oriented technologies with the
knowledge-oriented sciences acknowledges the role that imperative sen¬
tences play in technology, and the role that the indicative mood plays in the
sciences. This suggests to us that research into the nature of the sentential
mood and the development of a logic of the sentential mood will be impor¬
tant for obtaining a better understanding of information science and the
other semiotic sciences. Consider, for example, Gorn’s suggestion that since
imperative sentences are neither true nor false, good and bad may be used as
alternatives. However, acceptable/unacceptable and practical/impractical
are other equally important pragmatic alternatives; so are beautiful/ugly and
moral/immoral, though not as pure alternatives to true/false. This example
indicates the increased complexity pragmatic structure bears relative to se¬
mantic structure.

KNOWLEDGE- VERSUS ACTION-ORIENTED INFORMATICS

The main issue of the paper, as far as the purposes of the entire study are
concerned, is the relation between informatics and some of the other disci¬
plines that contribute to the scientific study of information. Gorn draws a
distinction between the knowledge-oriented component of informatics (in-
formation-and-computer science) and its action-oriented component (infor¬
mation/computer engineering).
The inseparability of information science and computer science is well
worth emphasizing. Even though a few scholars, including Gorn, have
viewed this field in such a way for years, it is only recently that the similarity
or identity of goals and approaches has begun to surface (attested to, for
example, by the 1982 Symposium on Empirical Foundations of Information
and Software Science). Referring to our earlier comments, a note is appro¬
priate, however, regarding the scope of information and computer science.
Gorn refers to information-handling machines as symbol-manipulating am¬
plifiers, a definition that would seem to restrict such machines to digital
computers only—a restriction probably not intended. While symbols may
146 CHARLS PEARSON AND VLADIMIR SLAMECKA

play some role in many analog computers, it is the iconic structure that plays
the essential role; analog computers are icon-manipulating amplifiers. Simi¬
larly, simulators (in the engineering-training, not the software sense of the
term) are indexical information processors or index-manipulating amplifiers.
A key point of Gorn’s paper is that the knowledge-oriented and the ac¬
tion-oriented components of informatics cannot and should not be sepa¬
rated. The arguments supporting this belief need to be carefully assessed.
Informatics is perceived, in a fashion more characteristic of information
engineering, as being concerned with synthesis and analysis of symbolic
expressions and processors. The action-oriented (technological) and the
knowledge-oriented (scientific) aspects of computation are said to be less
likely to move apart, because the description of symbolic expressions re¬
mains tied to the prescriptive language of its interpreting processors, hence,
invariably forming a pragmatic tie.
Two comments seem appropriate here. First, the merger of the knowl¬
edge- and action-oriented components of informatics is a historic fact. The
mixture of knowledge and action does, indeed, maintain a pregnant environ¬
ment that can be harmed by forcing their separation; the late Mortimer
Taube, perhaps the first philosopher of informatics, was fond of saying that
there is no such thing as pure and applied science, only good and bad
science.
On the other hand, the merger of knowledge- and action-oriented compo¬
nents of informatics is not absolutely necessary. The pragmatic tie between
them refers to the pragmatics of doing technology, not to the pragmatics
studied by a triadic science. At times, such a separation may be necessary
for the major idealizations and abstractions characteristic of the great ad¬
vances in science to take place. To argue as Gorn does that such a separation
will of necessity “cause the cessation of the very effervescence the mixture
of knowledge and action maintains,” is to argue that the separation of
mechanical engineering and physical thermodynamics in the nineteenth cen¬
tury was also a similar mistake. While the original inventions of the steam
engine and the internal-combustion motor were made by practitioners, no
one would argue today against the drastic revolution in the level of technol¬
ogy enabled by the pure knowledge-engendered advances made in physical
understanding by such action-indifferent scientists as Gibbs, Maxwell,
Helmholtz, Kelvin, and others. This is effervescence at its best!
In informatics, such advances may be sought by standing back from the
interpreting process to look at the prescriptive language of interpreting pro¬
cessors, and then describing abstractly what it is they do. This, in fact, is one
reason for the necessity of introducing semiotic analysis into informatics—
to abstract from the study of pragmatic structure and to describe it indica-
tively.
Gorn is entirely correct in his assessment of other informatics-related
disciplines as being largely professions. The interesting movement we are
beginning to note is the emigration of areas from informatics to other profes-
INFORMATICS AS A SEMIOTIC DISCIPLINE 147

sions: information systems and database areas into management science;

computer-aided design into mechanical engineering and architecture; artifi¬
cial intelligence into such problem-solving fields as chemistry and medicine;
and very-large-scale integration (VLSI) onto a new turf shared by physics,
electronics, and materials science and engineering. This trend—if it is a
trend—is comforting in at least two ways: It documents the utility of the
concepts engendered by informatics, and it preserves the relative stability
and natural evolution of scientific disciplines.
As the “informationalization” of other sciences, professions, and tech¬
nologies continues, it is interesting to ponder what will eventually remain in
the core of information and computer science and how extensive that core
and its community of practitioners will be. Our perceptions lead us to specu¬
late that the knowledge-oriented component of this core will be concerned
principally with the theory of signs, sign structures, sign processes, and
algorithms. The disturbing thought is that if this scenario is correct, the
educational programs in today’s departments of computer science and infor¬
mation science are not very relevant.
.
 THE ROLE OF INFORMATION
IN COMPUTER SCIENCE

Alan J. Perlis

Saul Gorn’s paper treats the relation of computer-and-information science to

some of its neighboring areas, for example, library science, information
retrieval, information science, cybernetics, cognitive psychology, artificial
intelligence, semiotics, linguistics, computer engineering, management, de¬
cision science, education, and pure mathematics. He seeks to distinguish the
former from the latter ideologically. His choice of the word informatics as a
convenient replacement for computer-and-information science is an inter¬
esting one and fits well with the views he develops but also leads to a de¬
emphasis of the key role played by the computer; that is, the consequence of
the ubiquitous, high-speed, and almost effortless execution of our symbolic
systems. While informatics is a newly synthesized and pleasantly neutral
word, it accents information, which is definitely not the key concept in¬
volved in issues treated in Gorn’s paper. The key, in fact, is the computer—
as silicon and metal, as metaphor, as silent servant—and it accounts for
most of the phenomena treated by Gorn. It particularly accounts for the
incredible dynamism that sharpens distinctions between computer-and-
information science and its neighbors.
Consider Gorn’s treatment of disciplines and their respective ideologies.
He points out that human knowledge develops as the available perceptive
devices extend and refine our range of distinguishable sensation. Because of
their almost arbitrary complexity, we are forced to translate these accumu¬
lated sensations into symbolic models whose properties and consequences
precipitate further extension and improved control of sensation. The sheer
bulk of accumulated sensation-extension, and its manipulation and control,
can only be managed by specialization. As these artificial things become
visible, they become knowable, usable, and suggestive. It is the dynamics of
sensation perception that matters, and from that appreciation springs the
importance of the processes by which we alter the focus of our individual
and cultural perceptions. Social agreement to perceive and social commit¬
ment to invest arise because of present traffic and future prospects, and all is
greased by the recognition of metaperceptions, the symbolic processes.
Gorn points out that human beings, and, hence, their disciplines, alternate
between states of perception and analysis and those of synthesis and action.
149
150 ALAN J. PERLIS

The forms of representation and communication within their languages tend

to differ between the two state-sets. Once a discipline extends its concerns
beyond the senses, a notion of truth is needed that operates on symbolic use
of language and has important properties invariant of representation. In
time, every notion of truth becomes dynamic and leads to the development
of metanotions, envelopes within which we manipulate our symbol systems.
We might ask: “What are the corresponding notions for dealing with the
representational and invariant properties of synthesis and action?”
It is here that the contact of all other disciplines with computer science
becomes pronounced. Gorn calls this claim a consequence of the mass ego
of computer science, much as reductionism is attributed to the mass ego of
physics. The representation of our syntheses as computer programs, their
experimental study as the execution (action) of these programs, and the
observations of the effects of these “runs” on our world (real and symbolic)
now constitute a major paradigm in our quest to extend sensation. The
widespread adoption of this paradigm is being accompanied by enormous
expansion in the computer itself. The expansions have become less func¬
tional and more concerned with increases in redundancy and ability to com¬
municate between machines and processes, their software counterparts. It is
a continual game of musical chairs in which at each stage either man, com¬
puter, or symbolic process is unsatisfied, and an assault to stabilize is
launched, whose success merely shifts the dissatisfaction to an ensuing stage.
Gorn makes the interesting observation that those disciplines that deal
with unsolvable problems flourish. The social commitment and the willing¬
ness to perceive are preserved because unsolvability is indefinitely post¬
poned by creating new perspectives that give the illusion that their solution
either gets us closer to that of the original problem or suggests a different
problem equally fundamental and, hence, unsolvable. Computer science
revels in unsolvable problems. One of its important subareas, artificial intel¬
ligence, in seeking symbolic models linking thinking to program execution,
is concerned only with unsolvable problems.
In summation, the Gorn paper presents an accurate picture of the inter¬
faces between computer-and-information science and other disciplines, but
the dynamics of the flows across the interfaces are less well treated because,
in this reviewer’s opinion, the computer is not put forth as generator and
pump but only as transformer. Informatics is too weak a word to identify the
rich dynamics of computer-and-information science. For that matter, com¬
puter science alone is quite adequate as an identifying term and does not gain
clarity or accuracy by including the word information. Computer science
does not deal with information differently than any other field. This can be
seen from the fact that it treats its own information in much the same way as
biologists or educators do theirs.
Though it shares little in methodology, notation, tools, training, and prob¬
lem focus, biology is closer to computer science than any of the areas listed
by Gorn, because of their mutual preoccupation with growth, evolution,
reproduction, and the complexity of ramifying “cell” structures.
 THE PROVINCE
OF COMPUTER SCIENCE

Vladimir Zwass

I am dividing my comments on Saul Gorn’s paper into three parts. In the first
part, I shall discuss the definition of computer science; in the second, I shall
reflect on the epistemological domain of this discipline; and in the third, I
shall assess the place of computer science among the family of disciplines
concerned with the study of information.

ON THE DEFINITION OF THE DISCIPLINE

The domain of computer science may be viewed either in relation to the

active object of its investigation, the computer, or in relation to the purpose
of computer operations, that is, deriving information. The proponents of the
second approach, notably Gorn and Slamecka, merge computer science with
information science. [Gorn, 1967 and also in this volume; Slamecka and
Gehl, 1978.] I shall argue that the two definitions may be seen to converge,
but that the point of convergence requires closer study.
Since information has no generally accepted definition at this time, we
may accept the operational meaning implicit in the preceding paragraph or,
following Boije Langefors, consider information to be an increment of
knowledge. [Langefors, 1977.]
Gorn offers several definitions of informatics, the term he uses in the
continental European fashion for the unified computer-and-information sci¬
ence. According to his initial definition, as a discipline concerned with ma¬
nipulating symbolic expressions, informatics would include formal logic and
mathematics—surely too broad a scope. In Gorn’s paper, the definition is
sharpened subsequently by relating symbolic expressions to their processors
and treating both as objects of research. This definition is congruous with the
one I will advocate for computer science.
Computer science is concerned with the limits and methods of processing
symbolic information by a synthetic processor and with synthesizing effec¬
tive processors. It then appears that the definition of the discipline has to
151
152 VLADIMIR ZWASS

focus on the processor, since it is the general capabilities of the processor

that delimit the discipline’s domain. The essential commitment of the disci¬
pline is to study computing within these limits.
Computer science is thus defined here as a study of the manipulation of
symbolic information by synthetic processors, called computers, and,
hence, the investigation of effective algorithms, processors, and representa¬
tion of information for such manipulation.
The word effective in this definition stresses the fact that the discipline
investigates goal-oriented human activity to a larger extent than natural
phenomena, and thus we are discussing one of the sciences of the artificial,
to use the phrase coined by Herbert Simon. [Simon, 1969 and 1981.] Com¬
puter science was originally defined in terms of the computer. [Newell,
Perlis, and Simon, 1967.] A definition of a scholarly discipline in terms of a
single artifact appears justified by the almost universal information-process¬
ing capabilities of the computer. In this definition, the computer is not a
certain physical implement, but an ensemble of well-defined capabilities.
Indeed, in the development of the field, the fundamentals of the theory of
computation preceded the appearance of the stored-program computer.
However, if we take away the computer as a technological phenomenon,
knowledge gained in the theoretical domains of this science may be accom¬
modated within the family of older disciplines, such as mathematics, logic,
or linguistics. In fact, some of the fundamental work was originally done in
mathematics. [Turing, 1936; Church, 1941.]
To remind ourselves that we are discussing a very young science, it is well
to consider the (perhaps harsh) words of Christopher Strachey about the
“pitiably small body of generally accepted fundamental laws and principles”
it has formulated. (Strachey, 1977, p. 124.)

THE EPISTEMOLOGICAL DOMAIN OF COMPUTER SCIENCE

As just defined, computer science aims not only at gaining knowledge of

existing natural phenomena, but, in the first order, at investigating the possi¬
ble in the artificial. Theoretical investigations are combined with empirical
findings, with the latter dominating. [Newell and Simon, 1976«.] The science
is strongly influenced by computer technology and the practice of comput¬
ing. Homo faber, the technologist of computer science, and homo cogitans,
the pure scholar in search of knowledge, are often one and the same person.
This person assumes the role of technological expert/leader, to use
Znaniecki’s typology of “men of knowledge.” [Merton, 1973.] Only the
scholarly rigor of a specific investigation makes the Ryleian know-what
stand out from the know-how.
If the definition I have offered is accepted, the following major areas of
computer science may be distinguished:
PROVINCE OF COMPUTER SCIENCE 153

1. Theory of Computation. A combination of formal disciplines em¬

ploying deductive reasoning, this field includes the theories of auto¬
mata and formal languages and investigates algorithmic complexity,
program verification, and the limits of automatic computability.
2. Syntactics, Semantics, and Pragmatics of Program Specification.
Broadening the framework of the study of programming languages,
this area investigates the means for specifying tasks to the computer,
the translation (interpretation) of this specification, and the execution
processes arising from the specification. An evolution, or rather a
build-up for a discontinuity (as posited by Thomas Kuhn), has been
taking place: a departure from the sequential prescriptive languages
of the von Neumann machine toward stronger descriptive languages,
possibly giving rise to parallel processes. [Winograd, 1979.] The aim
is to bridge the semantic gap between the person specifying the task
and the computer, with the level of the bridge adjusted to that of the
semantic domain of the person providing the specification. Alterna¬
tives for expressive specification range from querying by example
[Zloof, 1977] to functional programming [Backus, 1978].
3. Architecture of Processors. This investigation of the effective com¬
bination of hardware, software, and communications needed to cre¬
ate a desired computing environment is strongly influenced by the
underlying technologies. A shift from centralized to distributed pro¬
cessing has already occurred; the distribution vector has been a sub¬
ject of significant theoretical and empirical research.
4. Synthesis and Analysis of Algorithms. A combination of science
with craft, this area has deep foundations in theoretical computer
science.
5. Representation of Information. The study of data structures,
reflecting largely the physical aspects of data representation (efficient
placement of data in computer memories), has become a component
of a larger discipline, seeking to provide representational models of
real-world entities and relations among them, whose structure possi¬
bly varies over time.

A particular promise in the last area is held out by artificial intelligence, an

area of research in computer science whose immediate goal is to devise
computer-based systems for some of the tasks generally considered to re¬
quire human intelligence. The larger concern of artificial intelligence is to
establish a theory of intelligence based on the information-processing para¬
digm. Thanks to these approaches, “computer science makes it possible to
manipulate ideas as though they were things.” (Minsky, 1979, p. 392.)
These research areas are interrelated with technology to organize user
systems, such as management-information systems, computer-aided design
154 VLADIMIR ZWASS

and manufacturing (including robotics), or information storage and retrieval

systems.
The predominantly empirical nature of computer science is reflected by
its typical research cycle, usually following these steps, which closely corre¬
spond to Karl Popper’s theory of knowledge acquisition [Popper, 1972]:

A set of concepts is formed that is related to expanding the functionality

of computing in a certain direction (this often occurs in response to tech¬
nological innovations).
An invisible college of investigators refines and expands the conceptual
basis, often with the use of mathematical and simulation models, and
formulates the basic hypotheses to be tested in an operational system
incorporating the proposed ideas.
An exemplar, to use Thomas Kuhn’s terminology, is built. (Often several
alternative exemplars are studied.)
Experimentation and modification of the exemplar lead to generaliza¬
tions. The original concepts, often significantly reformulated, enrich the
body of computer science and may gain acceptance in the practical envi¬
ronment.

An excellent example of such research is the development of the working-

set theory of virtual-memory management, well documented by its origina¬
tor. [Denning, 1980.] A larger, paradigmatic change, which has been matur¬
ing over the last decade, is the predicted rejection of the von Neumann
computer architecture, centered on a single processor with linear memories
that store programs and data interchangeably, in favor of data-flow architec¬
ture. [Dennis and Misunas, 1975.] Taking advantage of opportunities offered
by the very-large-scale integration (VLSI) technology, data-flow architec¬
ture offers the promise of a high degree of operational parallelism and is
suitable for powerful functional programming.
To conclude, computer science is one of the fields where “external social
need’’ (T. S. Kuhn, 1962a and 1970, p. 19) influences many developmental
directions. The Platonian ideal of a disinterested search for knowledge is one
of the (weaker) motors of the disciplinary development; Juris Hartmanis, for
example, stresses the strategic role of the marketplace. [Hartmanis, 1981.]

COMPUTER SCIENCE AMONG RELATED DISCIPLINES

To appreciate the specific nature of computer science, we need to look at its

position in the family of disciplines.
“There have been signs in the past decade that the rather artificial separa¬
tion of disciplines may be coming to an end,” observed Noam Chomsky a
decade ago. [Chomsky, 1968 and 1972.] This statement may be well sup-
PROVINCE OF COMPUTER SCIENCE 155

ported in the case of computer science. To gain a perspective on the separa¬

tion of disciplines, we have to realize that it has occurred largely in the last
century and a half.
Computer science has an individual profile as well as close ties with other
disciplines. Gorn suggests that we not try to separate computer science from
information science. Though an intimate relation exists between the two, the
specific investigative domain of computer science previously discussed re¬
quires, I believe, that we treat this science as a separate field of inquiry.
To understand the relation between these two sciences, we need to ana¬
lyze the domain of information science. A number of scholars define infor¬
mation science as a metascience striving to develop fundamental theories of
information phenomena, ultimately couched in semiotics. Tefko Saracevic
claims that this science has no interest in the computer itself or no pragmatic
concerns in general. [Saracevic, 1970.] If we accept this interpretation,
avoided by Gorn, computer science could become a beneficiary of funda¬
mental theories of information science, offering in return, theoretical in¬
sights into artificial processing of information and the design of information
systems.
Machlup’s definition of information science as a study of processes and
systems of knowledge-transfer implies a far broader relation with computer
science. [Machlup, 1979a.] Many aspects of computer use would fall into the
domain of information science, so defined. But computers are increasingly
used to extend knowledge; computerized systems assist people not only by
accelerating cognitive processes, but also by augmenting them or substitut¬
ing for a specific process. It appears then that computer science and informa¬
tion science have partially overlapping concerns: first, with artificial pro¬
cessing of information and second, with the nature of information and
processes of its transfer.
Computer science is related to almost all other disciplines in multiple
ways. These relations may be classified as follows:

1. Sciences that furnish computer science with paradigms of conceptual

thinking. These disciplines include systems theory, which investigates con¬
cepts of wholeness and organization; cybernetics, a science of systems,
communication among systems, and regulation of systems; and information
theory, particularly later attempts to link it to semantic information. A clear
demarcation separating these disciplines would be difficult; they investigate
behavior of systems using different paradigms. Cybernetics has gained scant
acceptance in this country, but it has given impetus to the development of
artificial intelligence.
2. Sciences that share areas of inquiry with computer science. Because
of the interdisciplinary nature of computer science, its practitioners deal
with fields carved out by other established disciplines. This research ranges
from socioeconomic studies of computing [Kling, 1980] through work shared
with information science to cognitive psychology. The latter discipline,
156 VLADIMIR ZWASS

applied together with the methods of artificial intelligence and linguistics,

provided the foundation for cognitive science, which investigates problems
of knowledge, understanding, inference, and learning.
3. Sciences that share methods of inquiry with computer science. The
most prominent example is, of course, mathematics; the close relations
between the mathematical and computer cultures have been probed by sev¬
eral scholars. [Wegner, 1970; Knuth, 19746; Arden, 1980a.] Another such
discipline is the mathematical theory of communication. [Shannon and
Weaver, 1949.]
4. Sciences that are influenced by the use of computer techniques. It
would be difficult to find a field of inquiry that has not been influenced by the
computer. The early concern of George Kistiakowski, that computers would
change scientific research by biasing the nature of the problems selected for
study, has not been justified. Computers make an entire range of research
problems more tractable. In a deeper sense, pointed out by Gorn, they serve
sciences by furthering information-oriented thinking.

There remains no doubt that all fields of inquiry (and many other pursuits)
have been affected by computers. It is also evident that virtually all disci¬
plines have benefited from both computer science and information science
and that few, if any, have suffered undesirable dislocations due to their
influence.
 COMPUTER SCIENCE AS THE
SCIENCE OF DISCRETE
MAN-MADE SYSTEMS

Joel Moses

I am pleased to have been given the opportunity to review Saul Gorn’s paper
on informatics and present some of my own views on computer science and
its relation to its neighboring disciplines. It is surprising to me that there
have been relatively few attempts to define such an important and fast¬
growing field as computer science and contrast its approaches with those of
mathematics, engineering, the physical sciences, and so forth.

STATIC AND DYNAMIC ASPECTS

We are well beyond the period (1930-1960) when people were fascinated by
the sheer fact that man-made artifacts could be general computing engines.
This fascination led to an overemphasis on the logical and dynamical aspects
of information processing. The fact that computers actually performed com¬
putations by manipulating numbers and other symbols, led some theoreti¬
cians to studies of the logical limits of various computing machines (e.g.,
finite-state machines, Turing machines), and others to philosophical explora¬
tion of the meaning of certain computational devices such as assignments of
values to variables. In the last two decades, the emphasis on the dynamics of
computation led to studies of how fast one can perform certain computations
such as multiplication or to what extent one can trade space for time in a
variety of algorithms.
While these studies have been interesting, their relevance to the practice
of computing has been limited except in special areas. I view the rise of
software engineering, led largely by Edsger Dijkstra, as a reaction to much
of the research on computational dynamics. The position I take here is
largely a systems view of computer science that is characteristic of the
undergraduate program in computer science at the Massachusetts Institute
of Technology. In this approach, large software systems are seen as rela-

157
158 JOEL MOSES

tively static objects that are worthy of serious study. Experience shows that
the cost of designing, modifying, and maintaining large systems grows non-
linearly with the size of the system. Furthermore, large systems that one
wishes to remain viable must be changed constantly as the environment in
which they operate changes. Methods for coping with the complexity that is
often present in large systems, and, hence, for easing the process of embed¬
ding changes in specifications of these systems, are among the key issues in
static analyses of systems.
Although I emphasize here the static aspects of computation, it is not my
intent to imply that other aspects, such as the dynamic or aesthetic, of the
field should be entirely ignored. Efficiency, or dynamic issues underlying a
given system, should be dealt with after the static structure has been de¬
fined. Making it easier for someone to use man-made systems involves a
variety of what are often aesthetic issues. Such issues are dealt with by
disciplines other than computer science, such as art, linguistics, and psy¬
chology. Computer scientists have increasingly been getting involved in
such work, often in collaboration with those in the other disciplines.
A view of computer science that emphasizes the static rather than the
dynamic aspects of computation yields distinctions different from those
Gorn proposed between computer science and its neighboring disciplines.
Computer science is clearly not a physical science. The physical sciences
are concerned with discovering the principles of the design of a single sys¬
tem, namely God’s. Computer science deals with principles for creating
new, large man-made systems and is not limited to studying computer sys¬
tems. God’s system is indeed complex, but new approaches for reducing
complexity that are absent in God’s original design are not of much interest
to physical scientists.

ABSTRACT ENGINEERING AND ABSTRACT MANAGEMENT

Software engineers who are, in the main, very close to the systems view
espoused here, seem to claim an affinity between computer science and
engineering. Clearly, they have a point, but the analogy ought not to be
stretched too far. Engineers certainly try to create new man-made systems;
they are, however, limited by the properties of the physical world. They
spend most of their time figuring out ways to circumvent limitations in
materials or mechanical and electrical devices. Computer scientists are in
the enviable position of being given primitives (i.e., instruction sets) that can
usually be considered to work perfectly, without the interference of the
demons that God has placed in the primitives with which engineers must
cope. If computer science is to be viewed as a branch of engineering, it
should be considered as abstract engineering, largely unconstrained by limi¬
tations of the physical world.
Physical scientists and engineers are making increasing use of computa¬
tion in their work. The reason for this is that computers can make highly
SCIENCE OF DISCRETE MAN-MADE SYSTEMS 159

accurate discrete approximations to the continuous models used in the phys¬

ical sciences and engineering. It has been clear for some time that the tech¬
niques associated with numerical analysis are not at the core of computer
science. Rather, the core of computer science deals with problems associ¬
ated with systems composed of discrete components (e.g., processors, pro¬
cedures, and instructions). That some such systems mimic the behavior of
solutions of nonlinear differential equations is interesting, but it does not
shed light on the fundamental system issues of complexity, robustness, mod¬
ifiability, and so forth. Certainly, modern airplanes and electric-power sys¬
tems are very complex systems. Their complexity, however, is derived in
large part from the need of these systems to perform effectively in the
physical world with its continuous phenomena. The study of such complex
systems is not of much interest to a computer scientist.
The goals of my systems view of computer science are analogous to the
goals of general systems theory. The weakness of the analogy is that general
systems theory uses continuous models and usually studies control issues in
continuous systems rather than information flow in discrete systems, which
is the domain of computer science.
Many have noted the analogy between computer systems and the human
organizations that are the domain of management science. Human organiza¬
tions can be viewed as systems composed of discrete components (i.e.,
people) with discrete information-flow. Managers have an obvious problem
in dealing with primitives that are as ill-understood as individual human
beings. Thus, computer science can be viewed as the study of abstract
management.

PURE MATHEMATICS

There is one field that is as abstract, in fact, more abstract than the picture I
have painted of computer science; that field is pure mathematics. Mathemat¬
ics missed out on being the science of man-made systems for several rea¬
sons. The four main areas of pure mathematics are algebra, analysis, geome¬
try, and topology; the last three of these all deal with continuous rather than
discrete models. Abstract algebra, to which I shall return later, has been
used as a tool in the other areas and usually not as a model of discrete
systems. The three areas in mathematics (i.e., elementary number theory,
combinatorics, and logic) that have been applied to discrete systems have
not been at the core of modern mathematics. These areas have, in fact, been
heavily used in the dynamical and logical models of computer science.
The second failing of modern mathematics was its general lack of interest
in issues of efficiency that are at the core of the dynamical view of computa¬
tion and of great importance in man-made systems. Thus, investigators in
computer science and operations research have had to develop new mathe¬
matics for discrete systems, with relatively little support from mathemati¬
cians.
160 JOEL MOSES

Mathematics could have become the underlying theory of man-made sys¬

tems as well as the theory of God-made systems. That it did not develop in
this fashion is a result of historical accidents as well as ideological decisions,
as Gorn has noted. Nor do we see engineers or management scientists devel¬
oping a powerful theory of discrete man-made systems, because they have
to cope with their primitives. Hence, if such a theory is to be developed in
the coming decades, computer scientists are in the best position to develop
it. In fact, they must develop it!

DISCRETE MAN-MADE SYSTEMS

I now proceed to discuss in general terms what I perceive are some of the
core issues arising in large, discrete man-made systems. The principle issue
is the control of complexity. Something is complicated, the dictionary says,
when it has a large number of interconnected parts. Thus, complexity is a
function of the static nature of a system. It is unfortunate that theoreticians
in computer science use the term complexity theory for the theory that deals
with such issues as the minimum time or space needed to solve specific
problems. Such a dynamic theory ought to have been called a theory of
difficulty, but, one supposes, such a term is not as exciting as complexity
theory. In any case, apparently difficult problems, such as factorization of
integers into primes, need not be complex and vice versa.
If complexity is measured by the number of interconnections, then the
obvious way to achieve simplicity is to reduce the number of interconnec¬
tions a system requires. There is a limit to such a reduction when the number
of components (procedures, people, transistors, musical notes, etc.) has to
be large in order to achieve the desired functionality. The real key to reduc¬
ing complexity in large systems is to introduce additional structure or regu¬
larity when this is possible. When the interconnections possess some struc¬
ture, one is capable of modeling the systems using abstractions. In a more
abstract setting, the number of components and their interconnections is
much reduced. For example, if one recognizes that the same process is being
performed in many parts of the system, one can introduce a single new
procedure. Similarly, if the same action is being performed on a number of
elements, one can organize the elements into a vector. In the first case, we
have a procedural abstraction, in the second, a data abstraction.
An instance of an organizational structure of relatively low overall com¬
plexity is that of a pyramid or tower of linguistic abstractions. For example,
computer systems are often organized as layers of languages: a language
(actually, a family with restricted interconnections) of electronic compo¬
nents, a microprogramming language, an assembly language, and a higher
level language. Application programs can also often be designed in such a
pyramidal fashion; unfortunately, this is not always possible. For example,
sorting-routines are in general too combinatorial in nature to permit such
restructuring. Nevertheless, I believe it would be wise to consider such
designs as objectives for large systems.
SCIENCE OF DISCRETE MAN-MADE SYSTEMS 161

TOP-DOWN DESIGN METHODS

Top-down design methods are currently popular. A problem is broken down

into a few components; each component is broken down again and so on.
Top-down design is very general—it can be applied to most problems with¬
out too much difficulty. It possesses, however, an inherent bias toward
increasing the overall complexity of the design. Unless one is exceedingly
careful, there is a tendency in top-down design to duplicate parts of the
design in the overall system. I am reminded of the air force general who
claimed that a database system consisting of several million lines of code
could not have been written unless they used top-down design. I assume that
millions of lines of code would have been saved had there been more coordi¬
nation in the project than is likely to exist with a top-down approach.
Considerations of efficiency also cause an increase in complexity in many
cases. If one knows how to solve a problem directly, one tends to avoid
going through channels. As a result, when one retires or goes on vacation,
one’s replacements may be unable to get the job done because they lack
knowledge of the special arrangements. Furthermore, one may be unable to
cope with new directives, because keeping up with all the special arrange¬
ments has become a full-time job in itself. Top-down design results in sys¬
tems, I believe, that tend to force such special arrangements and their result¬
ing increase in complexity.
The organization of large enterprises has many features in common with
the organization of large software systems. To a first order of approxima¬
tion, American firms are organized in a top-down fashion and Japanese firms
in a more abstract, layered fashion. American firms give the appearance of
great flexibility, as in all top-down designs. Actually, the complexity of such
organizations reduces their flexibility enormously. Japanese firms, with their
careful attention to cooperation and coordination, appear to have much less
flexibility. In fact, the much lower overall complexity that results from such
coordination greatly increases the Japanese organization’s actual flexibility.
We may conclude that the way to reduce the overall complexity of a system
is to introduce greater local coordination and more abstract links, with all of
the local increase in complexity that such a principle entails. The alternative
principle, practiced in American firms, of keeping local interaction as simple
and clear as possible, tends to increase global complexity due to effects akin
to the “tragedy of the commons.’’ [Hardin, 1968.]

THE COMPUTER IN COMPUTER SCIENCE

As can be seen from the preceding discussion, my view is that the science
underlying computer science has little to do with computers. Rather, it is the
systematic study of issues related to the design of discrete man-made sys¬
tems. Small systems are often of little interest in such studies, since one has
far less difficulty in designing them, except when circumventing the vagaries
of nature or in producing some pleasing effect.
162 JOEL MOSES

Gorn makes the distinction between knowledge-oriented disciplines and

action-oriented disciplines. Action-oriented disciplines involve the creation,
and, hence, the design, of systems for accomplishing certain actions. A
surprising number of Gorn’s knowledge-oriented disciplines have a large
component of design. Musical compositions and novels (as well as religious
texts and law books) are all man-made designs involving discrete compo¬
nents. There are several maxims that are said to hold in such disciplines:
One is that you should read before you write; another is that you cannot
teach design. I agree with the former but only partially with the latter. For
example, it may turn out that mankind will not discover powerful principles
underlying aesthetics. On the other hand, mankind will be in serious straits if
it does not discover powerful principles underlying the design of large man¬
made systems. Such principles ought to help composers as well as the de¬
signers of the next set of encyclopedias.

SYMBOLIC MANIPULATION

My technical field within computer science is called symbolic manipulation.

It is concerned with manipulation of algebraic formulas used in algebra or
calculus. The irony is that Gorn equates informatics or computer science
with symbolic expressions and their manipulation. A main lesson of the past
two decades of activity in symbolic manipulation is the critical importance of
proper abstract data structures. Abstract algebra had discovered these struc¬
tures and especially their interrelations as layers in a tower, which has, of
course, led to my interest in the static structure of systems. Algorithms that
manipulate data structures can be much more flexible (and less complex)
when the appropriate abstract structure is used. This is a point in which
computer scientists are only lately becoming interested in the context of data
abstraction.
The view I have espoused here is not widely held by the computer science
community at this time. In fact, the continuing hardware revolution, as
exemplified by the growth in personal computing, is creating increasing
interest in relatively small systems. Yet, there is no escaping the problems of
large systems. Hardware designers are interested in placing ever larger de¬
signs on a chip, and the newest microprocessors permit one to use large
programs. As a result, owners of personal computers will soon be using
much larger programs than heretofore. It is a natural tendency for a field to
work on the problems that are most amenable to a solution. Eventually, the
core issues posed by large systems will have to be faced once again. The
sooner this reality is recognized, the better off the field of computer science
will be.
 PARADIGMS OF
INFORMATION ENGINEERING

Peter Wegner

Information science (computer science) is a young discipline striving to gain

the kind of respectability possessed by physics, mathematics, and engineer¬
ing. It is therefore not surprising that computer scientists have modeled their
research on paradigms in the tradition of the experimental, mathematical,
and engineering disciplines. Definitions of computer science that embody
each of these traditions are reviewed. Attention is then focused on engineer¬
ing paradigms, and in particular on the paradigms of software engineering
and the emerging subdiscipline of knowledge engineering. It is predicted that
knowledge engineering will emerge as a central subdiscipline in the 1990s
with paradigms, such as the knowledge-graph paradigm, that will emphasize
interactive man-machine cooperation in the management, learning, and use
of knowledge.
Paradigms are described by Thomas Kuhn as “oustanding achievements
that serve as a model for research by a community of researchers.” [Kuhn,
1962a and 1970.] He asserts that normal science in a mature discipline is
generally based on a paradigm that prescribes acceptable research and in¬
fluences our interpretation of phenomena. Scientific revolutions such as the
transition from Ptolemaic to Copernican astronomy or from Newtonian to
relativistic physics result in a change of paradigm. The term paradigm will be
used here to denote a model for undertaking and evaluating research even
when there is no associated outstanding achievement.
Information science currently has several paradigms that coexist with and
complement each other. The coexistence of multiple paradigms is, in Kuhn’s
view, a characteristic of pre-science rather than normal science. This view
suggests that computer science might not yet be sufficiently mature to sup¬
port a single paradigm. However, an alternative explanation is that informa¬
tion science is not a single discipline like physics, but a collection of disci¬
plines like the physical sciences and that the flourishing of many paradigms
is a sign of health and vigor rather than of immaturity.

163
164 PETER WEGNER

PARADIGMS AND IDEOLOGY

In his paper, Gorn examines the roles of ideology, methodology, and sociol¬
ogy in the development of disciplines. He points out the close relation be¬
tween ideologies (systematic bodies of concepts) and paradigms. He dis¬
tinguishes between knowledge-oriented activities such as mathematics,
action-oriented activities such as farming, and activities that transform
knowledge into action, such as education and engineering. He suggests that
sciences cannot flower independently of technology and in particular that in¬
formatics (computer-and-information science) could not begin to flower be¬
fore the technology of computing was developed. Informatics was initially
action-oriented (practical) and has achieved a fusion of action-oriented and
knowledge-oriented activities as it has matured. This balance is one of the
strengths of computer science in that practical results suggest relevant
theory, and theoretical results can increase the effectiveness of action. Gorn
warns that a divorce of action-oriented from knowledge-oriented activities
could lead to disaster.
Gorn’s choice of ideology, methodology, and sociology as characterizing
attributes of disciplines is unconventional but suggestive. These attributes
are independent of particular disciplines and are applicable in characterizing
any discipline or human activity. The ideology of a discipline determines its
paradigms and its criteria for evaluating the quality and relevance of contri¬
butions to the discipline. Its methodology is the principles that underlie the
tools and techniques used in pursuing the goals determined by an ideology.
Its sociology includes social interactions among practitioners of the disci¬
pline and the social impact of its concepts and products on society. Thus, the
ideology, methodology, and sociology associated with a discipline may be
thought of as its paradigms, underlying principles, and impact on people.
Ideologies are associated with young people and young disciplines. Com¬
puter science became an academic discipline in the mid-1960s with the emer¬
gence of the first departments of computer science and the publication of a
comprehensive undergraduate curriculum. [ACM Curriculum Committee,
1968.] The search for an identity led to a number of alternative definitions of
computer science, associated with different paradigms forjudging the value
of research. In Gorn’s terminology, these definitions express alternative
ideologies.

1. Computer science is the study of phenomena related to computers.

[Newell, Perlis, and Simon, 1967.]
2. Computer science is the study of algorithms. [Knuth, 1968.]
3. Computer science is the study of information structures. [ACM Cur¬
riculum Committee, 1968; Wegner, 1970.]
4. Computer science is the study and management of complexity.
[Dijkstra, 1972.]
PARADIGMS OF INFORMATION ENGINEERING 165

The first definition reflects an empirical view, since it asserts that com¬
puter science is concerned with the study of classes of such man-made
phenomena as computers, programming languages, algorithms, and data
structures. The second and third definitions reflect a mathematical tradition,
since algorithms and information structures are two abstractions that deter¬
mine different paradigms for modeling the phenomena of computer science.
The fourth definition takes the viewpoint of engineering and reflects the
great complexity of problems of information engineering encountered in
developing complex hardware-software systems.
Each of these definitions has motivated valuable contributions, and they
should be regarded as complementary rather than mutually exclusive. The
first was dominant in the 1950s and is experiencing a resurgence with the
increasing fashionability of empirical computer science. [Feldman and
Sutherland, 1979.] The second led to the flowering of research in the areas of
analysis of algorithms and computational complexity in the late sixties and
early seventies. [Arden, 1980a.] The third has motivated the study of seman¬
tic models of programming languages, also in the late sixties and early
seventies. The fourth has led to software engineering, which emerged as a
subdiscipline in the late 1960s and became a dominant research area in the
mid-1970s.
The first view of computer science is action-oriented, the second and third
are knowledge-oriented, and the fourth (software engineering) represents a
synthesis of both approaches. Thus, the dominant paradigm was action-
oriented in the 1950s, knowledge-oriented in the late 1960s and early 1970s,
and evolved as a fusion of the two in the late 1970s and 1980s. Computer
science was action-oriented in the period immediately following the birth of
computer technology, as there were no theories to support the new technol¬
ogy. The development of a theoretical foundation led to a brief period of
dominance of theoretical approaches, followed by the current period in
which the ideal, not always realized, is research that combines theory and
practice.
These four definitions span a variety of views of computer science. We
must ask whether this variety is a manifestation of the confusion of a young
discipline or whether a variety of paradigms is an inherent (permanent)
feature of computer science. My view is that variety is both inherent and
intrinsically healthy (many flowers should continue to bloom). The informa¬
tion sciences will blossom into a broad spectrum of mathematical, experi¬
mental, and engineering subdisciplines. By the year 2000, many universities
may have a school of information sciences with several departments sharing
a common set of core requirements, just as schools of engineering currently
include departments of electrical, mechanical, and civil engineering.
Schools of information sciences will contain information engineering as a
major component; information engineering will, in turn, contain software
engineering and knowledge engineering as two subdisciplines. Software en¬
gineering is concerned with managing the complexity of software, while
166 PETER WEGNER

knowledge engineering is concerned with the computer-based management

of knowledge. Software engineering was born in the late 1960s and has
within the short span of 15 years generated a cohesive body of knowledge
that combines theory and practice. Knowledge engineering, in the sense that
this term is used here, is currently in the process of being born. Its growth
will be intimately bound up with the technology of powerful graphics-based
personal computers which, by increasing the bandwidth of man-machine
communication by several orders of magnitude, provide a technological base
for extending man’s mental abilities in much the same way that mechanical
machines have extended man’s physical abilities.
The remaining sections of this paper examine the paradigms and ideology
of these two branches of information engineering.

THE EVOLUTION OF SOFTWARE ENGINEERING

During the period 1950-1980, there have been radical changes in the eco¬
nomics of information processing. The cost per executed instruction of com¬
puter hardware has decreased by a factor of two every two or three years,
while the complexity and total cost of computer applications has increased
by several orders of magnitude. The cost of software as a proportion of total
computing cost has increased from under 20 per cent in 1960 to over 80 per
cent in 1980.
Programming languages such as FORTRAN, originally developed for nu¬
merical problems with several hundred lines of code, proved to be inade¬
quate for large applications, with real-time requirements and hundreds of
thousands of lines of code, concerned with controlling the operation of
ships, airplanes, banks, or chemical plants. Large software projects in the
1960s and 1970s frequently failed to meet schedules, were subject to enor¬
mous cost-overruns, and sometimes had to be abandoned because their
complexity became unmanageable.
In the late 1960s, it was recognized that there was a software crisis and
progress in software technology required the development of systematic
techniques for managing the complexity of large software systems. [Naur,
Randell, and Buxton, 1976.] The discipline of software engineering was born
with the aim of providing a technological foundation for the development of
software products analogous to that provided by conventional engineering
for the development of physical products.
Software engineering exemplifies the fusion of action-oriented and knowl¬
edge-oriented activities. It is motivated by the economic considerations of
making software cheaper and more reliable. This contrasts with the schol¬
arly motivations traditionally associated with empirical and mathematical
paradigms. However, economic motivations are not necessarily inferior to
scholarly motivations as a basis for research. On the contrary, economically
PARADIGMS OF INFORMATION ENGINEERING 167

motivated paradigms provide stronger incentives for the fusion of action-

oriented and knowledge-oriented activities than purely scholarly motiva¬
tions and should, therefore, be regarded as a source of strength. Software
engineering has motivated research on program specification and
verification and on modularity and abstraction. It provides scope for the
creative practical application of theoretical ideas and is a prime example of
the synergy that can be generated by the fusion of theory and practice.
Software engineering, like other branches of engineering, is concerned
with developing a technology for the cost-effective production of a particular
kind of economically valuable product. It differs from other branches of
engineering because of the unique nature of software products. Physical
products like cars or television sets have a nonnegligible unit production
cost, requiring both labor and raw materials. In contrast, software costs are
concentrated entirely in producing the initial prototype, with negligible costs
in producing additional copies. Software does not wear out, although it may
become obsolete. Reliability is determined by logical features, such as cor¬
rectness and robustness, rather than by the physical endurance of raw mate¬
rials.
Software is a conceptual rather than a physical product. Constructing
software is in many respects more like constructing a mathematical proof
than building a house or a television set, since we are concerned with the
logical properties of the program rather than with the physical constraints of
materials. However, the fact that software products are judged by their
usefulness in problem-solving rather than by an abstract correctness crite¬
rion requires engineering as well as mathematical standards to be applied in
their construction. Further discussion of the nature of the software problem
and research directions in software engineering can be found in a recent
volume on software technology. [Wegner et al, 1979.]
The paradigms of software engineering are those of conventional engi¬
neering modified to take into account the fact that software is a conceptual
rather than a physical product. Many of the terms used in software engineer¬
ing are modeled on those of engineering. The term software is modeled on
the term hardware and reflects the fundamental insight that any behavior
realized by software may in principle be realized by hardware. The funda¬
mental notion of a software life cycle, which provides a basis for modeling
software systems, is borrowed from the engineering notion of a system life
cycle. The term software factory is used to describe systematic production
techniques for software components and application generators.
The ideology of software engineering is motivated by economic consider¬
ations and is rooted in that of engineering. Software engineering’s methodol¬
ogy for managing the complexity of computer systems, an outgrowth of this
ideology, is concerned with systematic principles for the construction, main¬
tenance, and enhancement of composite software systems from software
components, such that cost is a linear rather than an exponential function of
program size. The problems of management and technology transfer in a
168 PETER WEGNER

rapidly changing field present formidable sociological challenges. Mecha¬

nisms for the systematic introduction of new technologies must be tried that
minimize the pain of obsolescence and encourage a cooperative attitude
toward technological innovation.

ADA—A CASE STUDY IN TECHNOLOGICAL INNOVATION

The interplay of ideology, methodology, and sociology in software engineer¬

ing may be illustrated by considering the development and technological
exploitation of the programming language ADA. ADA was developed in the
late 1970s in response to the software crisis in order to reduce the cost and
improve the reliability of software. The motivation (ideology) for ADA's
development was economic, but it has stimulated worthwhile research in
software technology, language design, and system implementation in its
pursuit of practical objectives. ADA was designed to support a new software
methodology for constructing large programs from modular components. Its
use will require a new approach to programming very different from that of
current assembly-language or FORTRAN programming. The problems of
retraining and technology transfer for the large and growing number of po¬
tential ADA programmers are not only technical but also political and
sociological.
The development of ADA is a self-conscious, well-documented case
study of a large-scale project involving the creation, dissemination, and
effective use of a new body of knowledge. Each stage of its development
was surrounded by public debate among leading programming-language ex¬
perts and representatives of the user community. The evolution of the lan¬
guage requirements during 1975-1978 and of the language design during
1977-1980 is well documented. Currently, there is public debate about how
the novel concepts of ADA should be taught so that practicing programmers
can internalize new methods of program design and problem-solving. ADA
illustrates both the transformation of an ideology into a body of knowledge
based on a new methodology and the sociological concerns that must be
addressed in the process of technology transfer.
The history of programming languages has given rise to a number of
distinct ideologies and methodologies, each associated with different
paradigms of problem-solving and methods of programming. FORTRAN,
COBOL, PASCAL, and APL each have a characteristic programming ideol¬
ogy that is fervently advocated by their respective user communities. Every
successful programming language spawns a dedicated community of users
who would rather fight than switch, illustrating that dedication to an existing
paradigm may cause resistance to a new paradigm.
The ADA Joint Program Office, which is coordinating the development
and introduction of ADA, is keenly aware that acceptance and cost-effective
use of ADA requires a new ideology, methodology, and sociology on the
PARADIGMS OF INFORMATION ENGINEERING 169

part of its users. It is sponsoring not only language design and implementa¬
tion but also the study of methodology, education, and technology transfer.
For example, I am developing guidelines for education and technology trans¬
fer that will include recommendations on how to teach novel language fea¬
tures; structure ADA courses for managers, programmers, and other con¬
stituencies; and introduce ADA into industrial, military, and university
programming environments. [Wegner, 1982.] This task is difficult since there
are no guidelines for developing such guidelines. (Who will guide the
guides?) My approach is to provide information about current education and
technology-transfer efforts and draw attention to issues without necessarily
resolving them.
ADA has, within the short period of five years, generated a subculture of
tens of thousands of devoted adherents throughout the world. Hundreds of
companies are involved in a variety of ADA-related activities ranging from
implementation to education. The rapid spread of ADA at a time when it is
not yet implemented and its methodology is not yet tested illustrates Thomas
Kuhn’s assertion that inadequacy of a current paradigm provides fertile
ground for new paradigms.

THE INFORMATION REVOLUTION

The parallelism between software engineering and conventional engineering

is part of a wider parallelism between the information revolution and the
industrial revolution. The industrial revolution, which occurred in the period
1750-1950, was concerned with harnessing energy to serve man. The infor¬
mation revolution, which started around 1950 and is currently in full swing,
is concerned with harnessing information to serve man. The industrial revo¬
lution was concerned with developing machines to replace manual labor,
while the information revolution is concerned with developing machines that
replace mental labor. [Machlup, 1962.] There are striking similarities be¬
tween industrial and software technologies that stem from the fact that both
are concerned with enhancing and amplifying man’s ability to control the
environment. There are also fundamental differences that stem from the
differences between energy and information and differences in the nature of
the products produced by manual and mental labor.
The industrial revolution resulted in radical changes in technology and the
economics of production. It caused changes in life style from a predomi¬
nantly rural to a predominantly urban society and a shift in political power
from a landed aristocracy whose power was based on the ownership of land
to an industrial (capitalist or socialist) establishment whose power was based
on controlling the mechanisms for industrial production.
The information revolution is causing changes in technology and life style
as radical as those caused by the industrial revolution. The computer indus¬
try has, during the last twenty years, grown from insignificant beginnings to
170 PETER WEGNER

become one of the largest employers of labor and of capital. The nature of
computer applications is changing from primarily numerical applications,
which compute a numerical result, to data management and embedded com¬
puter applications, whose objective is to control a larger system whose
purpose is not primarily computational. Computer control is playing a larger
and increasingly important role in the management of society. Power is
shifting from the industrial establishment to those who control the dissemi¬
nation and distribution of information.
Computers are playing an increasing role not only in the management of
society but also in the management of knowledge. After many false starts,
artificial intelligence is finally reaching a level that allows building expert
systems whose performance is competitive with human experts. Personal
computers are becoming cheaper and more powerful, to the point where
everyone soon will be able to afford a pocket computer that executes a
million instructions a second, has a large memory bank extensible with
cassettes, and supports graphical and voice communication. The combina¬
tion of greater power and greater accessibility will entirely change the rela¬
tion between man and computers, so that computers will no longer be
merely problem-solving tools. They will become an extension of man’s intel¬
lect and serve to amplify intellectual capability in the same sense that con¬
ventional machines amplify man’s physical ability. The new role of com¬
puters will give rise to new paradigms for managing and using knowledge
and to a new subdiscipline of computer science, which I will refer to as
knowledge engineering.

KNOWLEDGE ENGINEERING

Knowledge engineering may be defined as the application of systematic

techniques to the management and use of knowledge. It is, in this sense, as
old as knowledge itself. Euclid’s Elements is an example of a magnificent
piece of knowledge engineering that provided a basis for managing geometri¬
cal knowledge, while the classification techniques of Linnaeus are an impor¬
tant example of knowledge engineering in botany and biology. Many of the
milestones in the development of science are as important for their contribu¬
tions to the management of knowledge as for their contributions to knowl¬
edge itself.
Widespread availability of powerful graphics-based personal computers
will cause fundamental changes in our methods of managing, learning, and
using knowledge. It will result in new ways of representing and organizing
knowledge and in the conscious development of paradigms and methodology
in an area that has traditionally been considered an art rather than a science.
It is predicted that by the 1990s, knowledge engineering will be as important
a subdiscipline of computer science as software engineering is today. Some
ot the characteristics of the emerging field of (computer-based) knowledge
engineering follow.
PARADIGMS OF INFORMATION ENGINEERING 171

Computer-based knowledge engineering depends on the representation of

knowledge by information structures inside a computer. The principle that
knowledge as well as numbers can be represented in a computer was recog¬
nized right at the outset and led to work in artificial intelligence, natural-
language translation, and information retrieval in the 1950s. However, the
advent of cheap powerful graphics-based personal computers, which may be
carried in a briefcase or a pocketbook and used on a day-by-day basis as an
extension of the human intellect, will entirely change the relation between
people and computers. Computer-based knowledge engineering could not
reach critical mass before the 1980s, because of inadequate technology.
Knowledge engineering bears the same relation to the management of
knowledge that software engineering bears to the management of software.
An item of knowledge, like an algorithm, is an inherently conceptual object
that can be given a concrete representation by an information structure and
manipulated, used, or displayed by a computer. The creation of computer¬
ized knowledge structures representing substantial bodies of knowledge
requires techniques for managing information complexity similar to those
required for a large program. The common ancestry of software engineering
and knowledge engineering as branches of information engineering is
reflected in their sharing certain methodological principles. The information
revolution is likely to spawn many different kinds of information engineer¬
ing, just as the industrial revolution generated many different kinds of physi¬
cal engineering disciplines. The different kinds of information engineering
will share with software engineering and knowledge engineering the idea of
representing a class of conceptual objects by concrete information struc¬
tures and the need to manage complexity when structures become large and
may evolve over time.
The term knowledge engineering was introduced by Feigenbaum in the
context of artificial intelligence and defined as “the art of bringing the tools
and principles of artificial intelligence to bear on application problems re¬
quiring the knowledge of experts for their solution.’’ [Feigenbaum, 1977.]
This definition views knowledge engineering as the art of representing
knowledge so that it can be used by computers to perform intelligent tasks.
The present view of knowledge engineering is broader, since it includes
building knowledge structures to aid human understanding. Knowledge en¬
gineering for human understanding is motivated by an ideology (paradigm)
different from that which motivates the development of expert systems. Its
goal is to amplify human intelligence rather than substitute computer intelli¬
gence for human intelligence. Its methodology involves educational technol¬
ogy, cognitive science, and human-factors research. The technology of man¬
aging the modular presentation of complex knowledge structures has some
of the flavor of software engineering but requires consideration of human
factors associated with animation, user interaction, multiple windows, and
other techniques for increasing the effectiveness of man-machine communi¬
cation.
Knowledge representations should facilitate display for the benefit of
172 PETER WEGNER

users, including multiple views and other forms of redundancy, rather than
efficiency and precision for the benefit of computers. Whereas knowledge
structures for computer understanding must be very detailed and precise,
knowledge structures for human understanding are concerned, not with the
precise specification of a computational task, but with organizing knowledge
for human readers who possess considerable contextual understanding and
are capable of conceptualizing at a level far above that of the computer.
Restructuring existing knowledge so that it is more accessible to humans
involves more than putting existing knowledge repositories such as the
Library of Congress on computers and accessing them through information-
retrieval systems. It involves restructuring existing knowledge so that it can
be flexibly presented in different formats for different contexts of use. The
technology for such restructuring is not well understood, but its nature can
be illustrated by considering recent developments in computerized printing
technology and computer-based learning.
Computers are revolutionizing printing technology and allow high-quality
text to be quickly and cheaply produced. Word-processing systems provide
authors with much greater control over production, layout, and modification
of text. Soon computers will be used not only for writing and printing books
but also for reading them. Book-size computers with flat panel displays will
make electronic books a reality. The greater bandwidth of man-machine
interfaces will qualitatively change the nature of man-machine communica¬
tion and make communication of knowledge by reading computer books
more effective than conventional communication by reading hard-copy
books.
Whereas hard-copy books consist of a linear sequence of pages, materi¬
als intended to be read on a computer may have a graph structure with
different entry points for readers with different backgrounds. Multiple win¬
dows allow the reader to pursue several lines of thought simultaneously or
view a given object at several levels of detail. Interactive responses by the
user can be used by the computer to tailor the mode of graph traversal to the
interests and skill level of the student. Each mode of the graph structure can
include dynamically animated pictures, texts, and programs. For example,
the mathematician may wish to animate the development of a proof, while
the computer scientist may wish to animate the process of program develop¬
ment and execution.
An electronic book represents a family of different hard-copy books that
could be obtained by printing out nodes of the graph structure in a particular
linear order for particular kinds of students. It is conjectured that flexibility
in adapting the pace and order of presentation of information to the student,
combined with the power of animation (possibly augmented by voice input
and output) can, if properly used, enormously increase the student’s capac¬
ity to absorb and understand both elementary and advanced knowledge.
Knowledge graphs that may be entered at different points and traversed in
different ways represent a paradigm for knowledge engineering that imposes
PARADIGMS OF INFORMATION ENGINEERING 173

a modular, interactive, discipline on both creators (authors) and users (stu¬

dents). Knowledge graphs are a basic representation not only for electronic
books, but also for computer games such as Adventure, which derive their
fascination from the fact that they allow players to explore new graph-
structured worlds. We do not yet have much experience with building large
knowledge graphs, since the hardware technology to support effective use of
such graphs is only now being developed. Some features of such graphs are
briefly described.
Knowledge graphs should have a domain-independent interconnection
structure that facilitates several modes of graph traversal, such as browsing,
retrieval, learning, reference, authoring, and so forth. Each node will have a
domain-dependent internal structure containing objects such as programs,
when representing knowledge about programming, and proofs, when repre¬
senting knowledge about mathematics. Creators and users of a graph will
have available to them a domain-independent set of operations for navigat¬
ing in the graph and domain-dependent operations for manipulating objects
in each domain. The ZOG system is probably the best known current ex¬
ample of a general-purpose system of this kind. [Robertson, McCracken,
and Newell, 1981.]
This discussion only scratches the surface of potential developments in
knowledge engineering. We must wait for at least a decade to know whether
its potential will be realized and to gain a better understanding of its
paradigms. However, I believe that during the next decade, knowledge en¬
gineering will begin to realize its potential for amplifying man’s intellectual
reach and that within twenty years, personal computers will be as indispens¬
able to scholarly research as libraries are to the scholars of today.
Computer-based learning is likely to become an important subfield of
knowledge engineering. Its paradigms, methodology, and sociology are
briefly discussed.

COMPUTER-BASED LEARNING

A new computer-based educational technology based on the knowledge-

graph paradigm will emerge in the next decade. Better man-machine com¬
munication will increase the effectivness of computer-aided educational
technology to the point where it dominates technologies based on traditional
books. Teachers may find that computer-aided supplements to classroom
teaching will, during the next decade, become both more effective and more
accessible than conventional books.
The new educational technology will have both a technical and a social
impact on the process of textbook writing. Technical computer writing style
is likely to differ from that for traditional textbooks, being more modular and
more interactive. Organization of large knowledge domains as a graph struc¬
ture of text modules is a challenge comparable to that of writing a traditional
174 PETER WEGNER

text on the same subject matter but will require authors to organize the
material in new ways. The modular approach has the disadvantage that it
may violate the natural continuity of the subject matter but the advantage
that it requires authors systematically to decompose knowledge into man¬
ageable modules.
Creators and readers of a computer textbook form a social community
whose members can communicate directly with each other via a computer
message system. Authors can incrementally make the text available, receive
instantaneous feedback from readers, and rapidly respond to such feedback.
Man-machine communication may be used not only for machine display of
knowledge, but also for communication among its community of creators
and users. Such social interaction will permeate all work in knowledge en¬
gineering and may fundamentally affect the sociology of all disciplines by
providing a new mode of communication among scholars.
Computers will not only help students to learn more effectively, but also
help authors to write more effectively. Conventional printing technology
permits only very costly enhancement of a book by means of new editions.
Computer technology permits continuous incremental enhancement after
development of the book has been completed, thereby permitting im¬
provements in quality and flexible adaptation to changing requirements that
would previously have been impossible.
The advantages of incremental enhancement can be illustrated by drawing
an analogy between the life cycle of a program and a book. Studies have
shown that 80 per cent of the effort of supporting a program over its life
cycle is in maintenance and enhancement. With conventional printing tech¬
nology, the only form of maintenance and enhancement is printing a second
edition, which is time-consuming and expensive. By allowing cheap incre¬
mental maintenance and enhancement, computer printing technology could
completely change the role of authors in the life cycle, allowing them to play
a much more active role in both the production and enhancement process.
A computer support system for a community of authors and students
involved in creating, disseminating, and learning a body of knowledge will
be called an educational environment. An educational environment should
support not only individual students but also communication among the
community of students and authors. An educational environment should
provide interactive feedback to students using the system and interactive
feedback to authors concerning the effectiveness of the system. It should be
concerned not only with technical issues of knowledge representation but
also with social issues of communication among a community of users with
different goals. As the man-computer interface becomes more powerful and
computers are increasingly used for communication among humans, social
issues and human factors for effective communication will become increas¬
ingly important.
An educational environment must provide an author environment that
helps authors create and enhance text modules, a student environment that
PARADIGMS OF INFORMATION ENGINEERING 175

allows students to learn effectively and receive adequate interactive feed¬

back, and a testing environment that provides statistics concerning the effec¬
tiveness of learning and allows hypotheses to be tested. We can think of the
environment as consisting of a database of text modules with different but
overlapping views provided for authors, students, and testing.
Experience with such computer-aided instruction systems as PLATO
[Bitzer, 1976] is relevant to the development of a computer-based education
environment. But progress in personal-computer technology and developing
environments as portable collections of software tools (such as UNIX [Bell
Laboratories, 1978]) makes it possible to create educational environments
with much greater power than PLATO.
Computer-based learning is a subfield of knowledge engineering that pro¬
vides a concrete illustration of its paradigms, methodology, and potential
social impact. Its paradigms are concerned with fundamentally increasing
the capacity of man to master knowledge, both quantitatively and qualita¬
tively. Its methodology involves educational techniques, software meth¬
odology, cognitive science, and artificial intelligence. Its sociology involves
synergy among a community of authors, teachers, and students in devel¬
oping and using an open-ended computer-based educational system.

CONCLUSION

Gorn’s concern with the ideology, methodology, and sociology of informa¬

tion science reflects an unconventional but worthwhile perspective on the
creation, diffusion, and use of knowledge in the computer field. The help¬
fulness of this perspective is illustrated by applying it to subdisciplines like
software engineering and large-scale case studies like ADA. However, the
impact of information science on our methods of learning, working, and
thinking, illustrated here by our discussion of knowledge engineering, is an
even more important topic, which Gorn has not addressed. Information
science has a special, intimate relation to knowledge, both because subdisci¬
plines like artificial intelligence are concerned with the mechanistic modeling
of knowledge and because knowledge engineering provides a tool for manag¬
ing knowledge that offers our only hope for controlling the knowledge explo¬
sion. Moreover, computers provide a new dimension for communication
among a community of scholars that could fundamentally change the sociol¬
ogy of creating and using knowledge in all academic disciplines.
.
 A PRAGMATIST REPLIES

Saul Gorn

In the spectrum of comments on my paper, the one that was furthest re¬
moved from my point of view was that of Joel Moses. He remarks on the
irony of his individual interest being symbolic manipulation when he cannot
accept my making symbolic manipulation a central issue of computer sci¬
ence.
I gather that he wants to identify the field as being concerned with the
analysis, design, and management of large, complex, discrete, man-made
systems. He would, therefore, expect it to include certain chunks of cyber¬
netics, especially of systems science, systems engineering, decision science,
and management science. The involvement of computers, on the one hand,
and symbol manipulation, on the other, may be absolutely essential in han¬
dling problems but are not, from his point of view, the central issue. To him,
a part of what I have called informatics that is more characteristic of the area
he is discussing would be software engineering. Peter Wegner in his com¬
ments also seems to consider this to be the most characteristic area of
informatics; I will comment on this later.
Now, whether the area described by Moses will actually crystallize into
what I have called a discipline will, of course, depend on such social phe¬
nomena as the development of a single new ideology and a few paradigms. In
his discussion of what could be considered an outline of such an ideology, I
found a few of his attitudes a bit puzzling. First of all, what led him to think
that I called the arts and crafts knowledge-oriented disciplines? Secondly,
Moses claims that it is the static structure of the large, discrete, man-made
systems with which he is concerned, when it is their maintenance and
robustness and modifiability in a constantly changing environment that is on
his mind. I can only assume that Moses believes it is the stability of the
system that is the central issue. Finally, and most important, I am puzzled
by his claim to be pragmatic while making a sharp distinction between en¬
gineering for God’s system (i.e., the physical world) and engineering for
man-made systems. Moses’s assumption seems to be that there is only one
world out there, with only one proper perception of it—a point of view I
would call monoideologicab, I think such an assumption is held by many
materialists, idealists, positivists, and “unified-scientists” and results in
rather complicated methodological positions.
177
178 SAUL GORN

However, there are billions of possible perceptions and their possible

outlines beating on our senses and even more on the technological exten¬
sions of our senses that we have developed, whether they depend on very
short-range or very long-range sampling in the human time-scale. We must
choose at any moment which to outline and perceive and in so doing, must
shut out the others—hence, the different disciplines. Is this not also man’s
system as well as God’s? On the other hand, having chosen what to perceive
from our variety of methods of interpretation, our physical and biological
interpretive behavior intervenes. Is this not also God’s system and not just
man’s?
From this pragmatic point of view, the distinction between the two worlds
is a red herring drawn across the trail. Plato (the extreme idealist) also felt
that there is only one world out there, and we only see its shadow on the wall
of our cave. I feel that whether there is only one world out there or not,
different people, by extending their perceptions and motor controls in differ¬
ent dimensions, are viewing different shadows in the different disciplines.
My type of pragmatism is essentially polyideological, with the university
structures of Western culture as a kind of pantheon. If, indeed, the stability
of man-made systems is Moses’s concern, I wonder how stable the area he
outlines can be.
I believe that, by and large, Alan Perlis, on the contrary, understands me
very well, and I even agree with him that the name informatics is weak. I
have found, of course, that my personally preferred expression, mechanical
pragmatics, fails to arouse understanding, let alone enthusiasm, although the
more philosophic in the audience understand cybernetic pragmatism as my
point of view. In my local environment, computer and information science
served very well, especially in view of our sharing activities in cognitive
science.
I certainly agree with Perlis that the electronic computer had a striking
effect that was critical in the development of informatics. The computer
(and, still more, the computer network) has changed our visualization of the
historical fluidity of symbol-manipulation systems from the viscous to the
effervescent, from being apparent only to the time-accelerating minds of
philosophers and historians to being obvious to the frequent user. In short,
the electronic computer itself was certainly a time-collapsing catalyst to our
understanding of symbol manipulation but is still only half the story; under¬
standing and designing and using symbolic systems being the other and older
half.
Perlis says that the computer scientist does not deal differently with infor¬
mation than the scientist in any other field does. This I deny. I hold that the
computer scientist’s semantic interpretation of symbols belonging to other
fields is essentially more muted and less real than that of a person in any of
the other fields, because it remains strictly at the metalevel. A person in
another field views its symbols as representing a living reality and the appro¬
priate laws, natural or ethical. He or she will be more sensitive than the
A PRAGMATIST REPLIES 179

computer scientist to both the axiomatic principles and the characteristic

methodological arguments justifying the methods, especially the paradigms
of his or her field. This is why designers of expert systems in the knowledge
engineering discussed by Wegner are teams of informaticians and people in
the appropriate disciplines, who describe, analyze, and program the pro¬
tocols that those in the discipline express when they think aloud (thereby
revealing the particular logical methods belonging to the discipline’s
paradigms). This is what I called dredging up the basic concepts from the
mass ego of that discipline. (See my discussion in my paper in this volume of
the professional aspects of informaticians and their relation to education.)
Wegner discusses this also in his section on computer-based learning. Evi¬
dently Feigenbaum, in the paper quoted by Wegner, maintains that such an
activity demands the involvement of people in that discipline as well as
computer people.1
In studying the design and use of symbolic systems, it is important to
study many, and not merely those that might be considered good but also
those that might be called bad, if only to be able to recognize the symptoms
of ill effects when they occur. For example, it is quite possible that the
phenomenon called information overload in people might not only be not
alleviated but actually catastrophically aggravated by the use of computers.
This is a possibility that Wegner should also consider in his enthusiastic
discussion of knowledge engineering and computer-based learning.
The effervescence caused by computers and other extenders of our sen¬
sors and effectors should not only extend our sense of power but also in¬
crease our humility; for these devices by which we feel we have extended
our perceptions and capabilities are so prosthetic that we can, in their use,
acquire no more grace than a dog walking on its hind legs or a man walking
on stilts, while our capacity for catastrophic misjudgment is tremendously
amplified. Misuses of computers, like misuses of atomic energy, space tech¬
nology, genetics, medications, and the pollution of knowledge generally can
easily lead to human disaster. This is, of course, an old and often-repeated
message: Pandora’s box, Frankenstein, The Sorcerer’s Apprentice, Dr.
Faustus, Paradise Lost, Icarus and Daedalus, the tower of Babel, Dr. Jekyll
and Mr. Hyde, and in a hidden way, Prometheus.
I would, therefore, hesitate to imply in the name to be given to the disci¬
pline that processors alone are its subject. The substance being processed
and transformed is to be equally investigated in order to provide a picture of
professional possibilities to the rest of humanity, who, even more than we
informaticians, must evaluate and choose, for good or ill.
Vladimir Zwass, unlike Perlis, proposes that perhaps the study of the

JIn this connection, it is interesting to go back to the Proceedings of the Western Joint Com¬
puter Conference 1961, whose theme was Extending Man’s Intellect and look at the papers by
Herbert Simon, Edward Feigenbaum, and Julian Feldman. [Western Joint Computer Confer¬
ence, 1961.]
180 SAUL CORN

computer as “the active object under investigation” is not identical to study¬

ing the“purposes of computer operations, much as the two studies may ov¬
erlap” . I agree that such areas might, indeed, be separated, but 1 think that
it would be unfortunate if they were, as 1 have stated in my paper. I return to
this question later. However, I disagree with some of the reasons for, and
modes of, separation that Zwass suggests.
First of all, I do not agree that the study of the manipulation of symbolic
expressions includes formal logic and mathematics. It includes the study of
syntactic methods employed in their notational systems, and only some of
these syntactic methods have invariant mathematical and logical signifi¬
cance. But by and large, mathematicians and logicians would consider such
pragmatic symbolic properties to belong to their methodological questions
and, hence, strictly speaking, to be outside the range of their subject proper.
For example, a discussion of the clumsiness of a notational system is not
considered a mathematical or logical question. Again, courses in analysis do
not spend time proving that the formal differentiation process for analytic
expressions, whatever the notation happens to be, must conclude without
the possibility of continuing indefinitely, even though the method of proof is
an example of a mathematical method applied to their notational systems.
Similarly, Herbrand’s syntactical attack on the study of first-order logic
needs an argument to show that the particular notation used is irrelevant as
far as the logic is concerned, because the properties discussed are notation-
ally invariant. Mathematics or logic is concerned with the common seman¬
tics of the variety of notations, and not with their syntax. And these disci¬
plines are certainly not interested in those syntactic properties that vary with
the notation, such as the connectivity of subexpressions, which I call prag¬
matic accidents of the notational system.
On the other hand, a significant portion of the theoretical aspects of
informatics uses mathematical methods and, therefore, is an example of
applied mathematics, just as the methodological discussion of mathematical
and logical notation is applied informatics. The interesting fact is that some
of these mathematical methods in informatics actually serve to keep action-
oriented aspects tied to the knowledge-oriented. For example, the tech¬
niques of formal verification of programs or their logically equivalent ma¬
chines can be viewed as interlocking a description of the machine in the form
of a state diagram at a cogently chosen level of detail with a prescription of
the program in the form of a flowchart at the same level.
When Zwass says that “Computer science is concerned with the limits
and methods of processing symbolic information by a synthetic processor
and with synthesizing effective processors,” we seem to be agreeing, al¬
though I would not insist on the word synthetic and, hence, would include
more than the sciences of the artificial. However, when Zwass goes on
to interpret a synthetic processor as among those called computers, and
stresses the effectiveness of their algorithms, it becomes clear that I have not
A PRAGMATIST REPLIES 181

succeeded in conveying the idea of interpreting processor as abstracting the

common essence of machines, programs, and symbolic processes. Thus, the
principle of logical equivalence of hardware and software no longer appears
to be a basic tautology, and the design paradigm applying equally to the
architecture of processors and the synthesis and analysis of algorithms does
not come across. (See, for example, the mathematical method that I have
just described in the preceding paragraph.) Zwass can, therefore, misinter¬
pret me as defining this scholarly discipline by its relation to a single artifact.
It is exactly the fear of this kind of misinterpretation that is the basis of my
departure from the view of the field in the letter by Newell, Perlis, and
Simon to which Zwass refers. [Newell, Perlis, and Simon, 1967.] The misin¬
terpretation is there in spite of Zwass’s qualification that “the computer is
not a certain physical implement, but an ensemble of well-defined capa¬
bilities.”
Thus, my view of computer science is broader than Zwass gives it credit
for being, while my view of information science is not so all-embracing as he
thinks. On the other hand, the picture Zwass gives of a separate information
science seems to me to cover all the cognitive sciences, which I am not
convinced will crystallize into one discipline. For I believe they are coexten¬
sive with what I would be tempted to name general ideology, returning to the
original sense of that word.
In the areas of applications, Zwass presents an interesting classification of
the sciences by the type of relation they have with computer science. I
would like to see this classification extended to include action-oriented disci¬
plines and the professions.
Whereas Perlis thinks that I have overemphasized information science
and Zwass thinks that I have forced it into a marriage with computer sci¬
ence, Charts Pearson and Vladimir Slamecka agree somewhat with my har¬
nessing them together. They hold, however, that the result should include
more semiotic studies and not force the marriage with information engineer¬
ing. I am in general agreement with Pearson and Slamecka on the first point,
although I do not know how to set the boundary in order not to include all of
the cognitive sciences. In particular, I also agree that the variety of meanings
of meaning should be considered in informatics. In talking about semantics,
I have been using the word meaning in the sense of the object symbolized
and have let the word pragmatics carry the others, where, as Pearson and
Slamecka point out, its meaning is much, much broader. Also other mood
effects should be studied besides the indicative and the imperative, and even
there we need more application studies of deontic, temporal, and dynamic
logic. Furthermore, I am glad Pearson and Slamecka brought up the fact that
the syntactic types noun and verb are not linguistic universals; it adds em¬
phasis to my stressing the importance of pragmatic effects of language. I am
also glad that they mentioned Peirce’s concept of abduction, which can be so
important in modeling goal-seeking behavior in artificial intelligence. How-
182 SAUL GORN

ever, it was my intention to emphasize that each discipline has its own flavor
of reasoning and these flavors may be a major concern of methodology; 1
have mentioned this in connection with expert systems.
Pearson and Slamecka are also correct about my not intending to restrict
attention to the digital in symbol manipulation. I have always felt that analog
machines should also be involved. They are better suited and faster in imitat¬
ing by feedback the adaptive and conditioning type of learning, because they
do not proliferate memories to be retrieved. Robotic sensors and receptors
can obviously use them.
When Pearson and Slamecka object to my saying that the symbology of
mathematics and physics may be formal, they forget, on the one hand, that
the intuitionists in mathematics do not accept formalist views and, on the
other, that rational mechanics is considered part of physics in the United
States but part of mathematics in Europe.
Pearson and Slamecka find that harnessing knowledge and action is re¬
lated to pragmatics merely in “doing technology, not to the pragmatics
studied by a triadic science.” But does not the action force the selection of
such using and interpreting behavior as is relevant to the symbology? In this
connection, I was surprised to see Helmholtz listed as an action-indifferent
scientist. His study of the adaptation time required in using prismatic lenses
was a direct study of the pragmatic effects of perception on action and a
model for psychologists.
We are agreed, however, that the professional aspects of informatics go
well with the theoretical, though we may argue about how separate they are
and for how long. Not only do I insist, as Pearson and Slamecka do, that
pragmatic questions should be an object of study within the field, but I
submit that professional aspects must also be included in such studies. It is
very much to the point, for example, that a contextfree language cannot be
interpreted by a finite-state machine but also needs a pushdown store. To
me, this is not merely a theorem in automata theory and formal languages
but also a fact of mechanical pragmatics; for it clearly states a relation
between a set of symbolic expressions and its mechanical user or inter¬
preter.
As for the possibility of a number of professional aspects having their
pilot studies within informatics and then graduating into separate disciplines
with related ideologies but different paradigms, I find Pearson and Slamec-
ka’s description and examples very well put. Except for the VLSI (very-
large-scale integration) example, they belong to the area of knowledge¬
engineering discussed by Wegner.
Wegner remarks early in his comments that knowledge-oriented, action-
oriented, and professional aspects of informatics have different paradigms;
and I have to agree that such radically different orientations must differ in
their paradigms. I believe, however, that a shared ideology is enough to keep
them in a single community. I therefore disagree with Thomas Kuhn that the
mark of a single discipline (physical science, in his case) is a single paradigm;
A PRAGMATIST REPLIES 183

a single ideology suffices, and major revolutions in a discipline are changes

in ideology, in which case the paradigms must follow suit. After all,
paradigms are social decisions on the courses of action that are proper for
the given ideology. The community may agree completely on what is worth
perceiving and yet vary in deciding what to do about it, without necessarily
losing touch with one another.
In the case of informatics, Wegner describes this variety, and I agree with
him that it is a sign of health, although I doubt that its members form the neat
time-packages he describes. For that matter, although software engineering,
as Wegner describes it, fits neatly into informatics, I consider its scope too
narrow. Its title, already a decade old, is unfortunate because it tends to
negate the ideological identification of hardware and software. Will the sys¬
tem design of software be so very different from that of hardware in a world
of data distributed in networks and of microprocessors assembled in VLSI
components? Will methodological arguments produce different paradigms? I
would guess that they would not precisely because of the logical-equivalence
idea that I have already described.
As for these areas and the ones spinning off knowledge engineering to
form a school of information sciences, Wegner seems to agree with Pearson
and Slamecka, though he stresses different areas. Various areas of applied
mathematics have formed such alliances in the past, and very successful
ones, as Wegner, Pearson, and Slamecka know better than most. But the
spread of applications of mathematics became too varied for all to fit into
anything smaller than a university. The spread of applied informatics will be
even broader, because of its pragmatic aspects. I would think that institutes
smaller than complete universities would be built around closely allied
ideologies and paradigms. The cognitive sciences alone would have to be as
ambitious as most schools of education, covering as they do the survey of
ideologies and their maintenance. And is not the idea of an institute of
knowledge engineering a description in modern dress of a school of educa¬
tion, emphasizing modern educational techniques?
.
SECTION 3
 INTELLECTUAL ISSUES
IN THE HISTORY OF
ARTIFICIAL INTELLIGENCE

Allen Newell

Science is the quintessential historical enterprise, though it strives to pro¬

duce at each moment a science that is ahistorical. With a pa'Ssion bordering
on compulsion, it heeds the admonition that to ignore the past is to be
doomed to repeat it. Science has built its main reward system around discov¬
ering and inventing, notions that are historical to the core. Thus, writing
about science in the historical voice comes naturally to the scientist.
Ultimately, we will get real histories of artificial intelligence (henceforth,
AI), written with as much objectivity as the historians of science can muster.
That time has certainly not come. We must be content for a while with
connections recorded in prefaces, introductions, citations, and acknowledg¬
ments—the web that scientists weave in their self-conscious attempt to
make their science into a coherent historical edifice. So far, only a few
pieces, such as Machines Who Think, provide anything beyond that, and
they still have no deliberate historiographic pretensions. [McCorduck,
1979.]
This essay contributes some historical notes on AI. I was induced to put
them together originally in response to a request by some of our graduate
students in computer science for a bit more historical perspective than is
usual in their substantive fare. It is to be viewed as grist for the historian’s
mill but certainly not as serious history itself. The attempt to define and
document all of what I put forward is beyond my resources for the moment.
This essay’s claim to accuracy, such as it is, rests on my having been a

I thank Elaine Kant and Stu Card for comments on an earlier draft and Paul Birkel and Marc
Donner for leading me to write the paper. Note: This research was sponsored in part by the
Defense Advanced Research Projects Agency (DOD), ARPA Order No. 3597, monitored by the
Air Force Avionics Laboratory under Contract F33615-78-C-1551. The views and conclusions
contained in the paper are those of the author and should not be interpreted as representing the
official policies, either expressed or implied, of the Defense Advanced Research Projects
Agency or the United States government.

187
188 ALLEN NEWELL

participant or an observer during much of the period. As is well known to

historians, the accuracy of the participant-observer is at least tinged with
bias, if not steeped in it. The situation is worse than that; I am not just a
participant but a partisan in some of the history here, including parts still
ongoing. Reader beware.

HOW IS THE HISTORY OF A SCIENCE TO BE WRITTEN?

Human endeavors are indefinitely complex. Thus, to write history requires

adopting some view that provides simplification and homogenization. The
standard frame for the history of science is in terms of important scientific
events and discoveries, linked to and by scientists who were responsible for
them. This assumes that scientific events declare themselves, so to speak. In
many respects this works, but it does so best when the present speaks
clearly about what concepts have won out in the end, so that we can work
backward through the chain of antecedents, adding only a few dead-ending
branches to flesh out the story.
With fields in an early state—and AI is certainly one—critical events do
not declare themselves so clearly. Additional frameworks are then useful.
Obvious ones of general applicability are proposed theories and research
methodologies; neither is very satisfactory for AI. The theoretical ideas put
forth have, especially when successful, been embedded in computer systems
(usually just as programs but sometimes including special hardware). Often,
the systems speak louder than the commentary. Indeed, a common com¬
plaint of outsiders (and some insiders) is that there is no theory in AI worthy
of the name. Whether true or not, such a perception argues against taking
theories as the unit in terms of which history is to be written. As for research
methodology, AI as a whole is founded on some striking methodological
innovations, namely, using programs, program designs, and programming
languages as experimental vehicles. However, little additional methodolog¬
ical innovation has occurred within the field since its inception, which makes
for lean history.
Similarly, the more sophisticated units of historical analysis, such as the
paradigms of Kuhn or the research programmes of Lakatos, provide too
course a grain. [Kuhn, 1962a; Lakatos, 1970.] It can be argued that AI has
developed and maintained a single paradigm over its short lifetime, or at
most two. Similarly, it has contained at most a small handful of research
programmes. But units of analysis work best with enough instances for
comparative analysis or for patterns to emerge. There are certainly too few
paradigms for an internal history of AI. The same is probably still true of
research programmes as well, though it would be of interest to attempt such
a description of AI.
Useful frameworks for historical analysis can often be based on the or¬
ganization of subject matter in a field. AI proceeds in large part by tackling
INTELLECTUAL ISSUES IN THE HISTORY OF Al 189

one task after another, initially with programs that can accomplish them
crudely, followed gradually by successive refinements. Game-playing,
theorem-proving, medical diagnosis—each provides a single developmental
strand that can be tracked. Thus, a history of AI as a whole could be written
in terms of the geography of tasks successfully performed by AI systems.
Almost orthogonal to this task-dimension is that of the intellectual func¬
tions necessary for an intelligent system—representation, problem-solving
methods, recognition, knowledge acquisition, and so forth—what can be
termed the physiology of intelligent systems. All these functions are required
in any intellectual endeavor of sufficient scope, though they can be realized
in vastly different ways (i.e., by different anatomies), and tasks can be found
that highlight a single function, especially for purposes of analysis. Thus, a
history can also be written that follows the path of increased understanding
of each function and how to mechanize it. Both of these structural features
of AI, and perhaps especially their matrix, provide potentially fruitful
frameworks for a history. Their drawback is just the opposite from the ones
mentioned earlier, namely, they lead to histories that are almost entirely
internal, shedding little light on connections between AI and neighboring
disciplines.
I settle on another choice, which I will call intellectual issues. It is a
sociological fact of life that community endeavors seem to polarize around
issues—fluoridation versus ban fluoridation, liberal versus conservative.
Such polarizing issues are not limited to the purely political and social arena
but characterize scientific endeavors as well—heliocentrism versus geocen¬
trism, nature versus nurture. Intellectual issues are usually posed as
dichotomies, though occasionally three or more positions manage to hold
the stage, as in the tussle between capitalism, socialism, and communism.
Intellectual issues are to be distinguished from issues in the real world of
action. No matter how complex and ramifying the issues of individual free¬
dom and state control that lie behind a fluoridation campaign, the passage or
defeat of an ordinance banning fluoridation is a concrete act and is properly
dichotomous. But with nature versus nurture, the dichotomy is all in the eye
of the beholder, and the real situation is much more complex (as is pointed
out ad nauseum). The tendency to polarization arises from the way people
prefer to formulate intellectual issues.
Scientifically, intellectual issues have a dubious status at best. This is true
even when they do not have all the emotional overtones of the previous
examples. Almost always, they are defined only vaguely, and their clarity
seldom improves with time and discussion. Thus, they are often an annoy¬
ance to scientists just because of their sloganeering character. Some time
ago, in a conference commentary entitled You Can’t Play Twenty Questions
with Nature and Win, I myself complained of the tendency of cognitive
psychology to use dichotomies as substitutes for theories (e.g., serial ver¬
sus parallel processing, single-trial versus continuous learning). [Newell,
1973 b.]
190 ALLEN NEWELL

Intellectual issues surely play a heuristic role in scientific activity. How¬

ever, I do not know how to characterize it, nor am I aware of any serious
attempts to determine it, though some might exist. Of course, large numbers
of scientists write about issues in one way or another, and almost all scien¬
tists of an era can recognize and comment on the issues of the day. Were this
not true, they could hardly be the issues of the particular scientific day.
From a historical and social standpoint, of course, intellectual issues have a
perfectly objective reality. They are raised by the historical participants
themselves, and both the existence of intellectual issues and the activity
associated with them can be traced. They enter the historical stream at some
point and eventually leave at some other.
Whether intellectual issues make a useful framework for a scientific his¬
tory seems to me an entirely open question. Such a history does not at all
substitute for histories based on events and discoveries, laid down within a
framework drawn from the substantive structure of a field. Still, ever since
that earlier paper in 1973,1 have been fascinated with the role of intellectual
issues. Recently, I even tried summarizing a conference entirely in terms of
dichotomies. [Newell, 1980a.] Withal, I try it here.

THE INTELLECTUAL ISSUES

I will actually do the following: I will identify, out of my own experience and
acquaintance with the field, all of the intellectual issues that I believe have
had some prominence at one time or another. Although I will take the field of
AI as having its official start in the mid-1950s, the relevant intellectual issues
extend back much earlier. We surely need to know what issues were extant
at its birth. I will attempt to put a date both on the start of an issue and on its
termination. Both dates will be highly approximate, if not downright specu¬
lative. However, bounding the issues in time is important; some issues have
definitely gone away and some have come and gone more than once, though
transformed each time. I will also discuss some of the major features of the
scientific scene that are associated with a given issue. I will often talk as if an
issue caused this or that. This is in general illegitimate. At best, an issue is a
publicly available indicator of a complex of varying beliefs in many scientists
that have led to some result. Still, the attribution of causation is too conven¬
ient a linguistic practice to forego.
Table 1 lays out the entire list of intellectual issues. In addition to the
short title ol the issue, expressed as a dichotomy, there is an indication of an
important consequence, although this latter statement is necessarily much
abbreviated. The issues are ordered vertically by date of birth and within
that by what makes historical sense. All those born at the same time are
indented together, so time also moves from left to right across the figure;
except that all the issues on hand when AI begins in 1955 are blocked
togethei at the top. Issues that show up more than once are multiply repre-
 Table 1. The Intellectual Issues of Al

1640-1945 Mechanism versus teleology: settled with cybernetics

1800-1920 Natural biology versus vitalism: establishes the body as a machine
1870- Reason versus emotion and feeling #1: separates machines from men
1870-1910 Philosophy versus the science of mind: separates psychology from
philosophy
1910-1945 Logic versus psychologic: separates logic from psychology
1940-1970 Analog versus digital: creates computer science
1955-1965 Symbols versus numbers: isolates AI within computer science
1955- Symbolic versus continuous systems: splits AI from cybernetics
1955-1965 Problem-solving versus recognition #1: splits AI from pattern
recognition
1955-1965 Psychology versus neurophysiology #1: splits AI from cybernetics
1955-1965 Performance versus learning #1: splits AI from pattern recognition
1955-1965 Serial versus parallel #1: coordinate with above four issues
1955-1965 Heuristics versus algorithms: isolates AI within computer science
1955-1985 Interpretation versus compilation: isolates AI within computer
science
1955- Simulation versus engineering analysis: divides \I
1960- Replacing versus helping humans: isolates AI
1960- Epistemology versus heuristics: divides AI (minor); connects
with philosophy
1965-1980 Search versus knowledge: apparent paradigm shift within Al
1965-1975 Power versus generality: shift of tasks of interest
1965- Competence versus performance: splits linguistics fiom AI
and psychology
1965-1975 Memory versus processing: splits cognitive psychology from
AI
1965-1975 Problem-solving versus recognition #2: recognition rejoins
AI via robotics
1965-1975 Syntax versus semantics: splits linguistics from AI
1965- Theorem-proving versus problem-solving: divides AI
1965- Engineering versus science: divides computer science,
including AI
1970-1980 Language versus tasks: natural language becomes central
1970-1980 Procedural versus declarative representation #1: shift
from theorem-proving
1970-1980 Frames versus atoms: shift to holistic representations
1970- Reason versus emotion and feeling #2: splits AI from
philosophy of mind
1975- Toy versus real tasks: shift to applications
1975- Serial versus parallel #2: distributed AI (Hearsay-like
systems)
1975- Performance versus learning #2: resurgence
(production systems)
1975- Psychology versus neuroscience #2: new link to
neuroscience
1980- Serial versus parallel #3: new attempt at neural
systems
1980- Problem-solving versus recognition #3: return of
robotics
1980- Procedural versus declarative representation #2:
PROLOG

191
192 ALLEN NEWELL

sented in the table, according to the date of rebirth, and labeled #1, #2, and
so forth. When the ending date is not shown (as in Reason versus Emotion
and Feeling #7: 1870- ), then the issue still continues into the present.
The issues are discussed in historical order, that is, according to their
order in the table. This has the advantage of putting together all those issues
that were animating a given period. It has the disadvantage of mixing up lots
of different concepts. However, since one of the outcomes of this exercise is
to reveal that many different conceptual issues coexisted at any one time, it
seems better to retain the purely historical order.

Mechanism versus Teleology: 1640-1945

We can start with the issue of whether mechanisms were essentially without
purpose. This is of course the Cartesian split between mind and matter, so
we can take Descartes as the starting point. It is an issue that can not be
defined until the notion of mechanism is established. It is and remains a
central issue for AI, for the background of disbelief in AI rests precisely with
this issue. Nevertheless, I place the ending of the issue with the emergence
of cybernetics in the late 1940s. If a specific event is needed, it is the paper
by Rosenblueth, Wiener, and Bigelow, which puts forth the cybernetic
thesis that purpose could be formed in machines by feedback. [Rosenblueth,
Wiener, and Bigelow, 1943.] The instant rise to prominence of cybernetics
occurred because of the universal perception of the importance of this
thesis. (However, the later demise of cybernetics in the United States had
nothing whatsoever to do with any change of opinion on this issue.) AI has
added the weight of numbers and variety to the evidence, but it has not
provided any qualitatively different argument. In fact, from the beginning,
the issue has never been unsettled within AI as a field. This is why I charac¬
terize the issue as vanishing with cybernetics. It does remain a live issue, of
course, in the wider intellectual world, both scientific and nonscientific,
including many segments of cognitive science. Above all, this issue keeps AI
in perpetual confrontation with its environment.
Intelligence presupposes purpose, since the only way to demonstrate in¬
telligence is by accomplishing tasks of increasing difficulty. But the relation
is more complex the other way around. While purpose could hardly be
detected in a device with no intelligence, that is, with no ability at all to link
means to ends, no implication follows about the upper reaches of intelli¬
gence. Animals, for instance, are obviously purposive yet exhibit strong
limits on theii intelligence. Thus, settling the question of artificial purpose
does not settle the question of artificial intelligence. The continuation of this
basic controversy throughout the entire history of AI over whether intelli¬
gence can be exhibited by machines confirms this separation. Yet, histori¬
cally it is not right to posit a separate issue of mechanism versus intelligence
to contrast with mechanism versus teleology. No such distinction ever sur¬
faced. Instead, there is an underlying concern about the aspects of mentality
INTELLECTUAL ISSUES IN THE HISTORY OF Al 193

that can be exhibited by machines. This shows itself at each historical mo¬
ment by denying to machines those mental abilities that seem problematic at
the time. Thus, the argument moves from purpose in the 1940s to intelli¬
gence in the 1950s. With the initial progress primarily in problem-solving, we
occasionally heard in the 1960s statements that machines might solve prob¬
lems but they could never really learn. Thus, the basic issue simply endures,
undergoing continuous transformation.

Natural Biology versus Vitalism: 1800-1920

A critical issue for AI that had come and gone long before AI really began is
the issue of vitalism—do living things constitute a special category of en¬
tities in the world, inherently distinct from inanimate physical objects. As
long as this issue was unsettled, the question of whether the mind of man
was mechanical (i.e., nonspecial) was moot. It is difficult to conceive of
concluding that the animate world does not generally obey the laws of the
physical world but that the mind is an exception and is entirely mechanical.
Thus, only if vitalism has been laid to rest for our bodies can the issue be
joined about our minds.
The vitalist controversy has a long and well-chronicled history. Retro¬
spectively, it appears as an inexorable, losing battle to find something spe¬
cial about the living, though the issue was joined again and again. Organic
matter was just a different kind of matter from inorganic matter—an issue
laid to rest finally with the synthesis of urea, an indisputably organic mate¬
rial, from inorganic components in 1828 by Wohler. Organisms had their
own inherent internal heat—an issue laid to rest in the work of Bernard by
the mid-1800s. For our purposes, the starting and ending dates of the issue
are not critical. Vitalism’s last champion may be taken to be the embryolo¬
gist Hans Driesch at the turn of the century, who proposed that organisms
develop only by virtue of nonmaterial vital principles, called entelechies.
[Driesch, 1914.] Issues almost never die, of course, as the continued exist¬
ence of the Flat Earth Society should remind us. Nevertheless, no substan¬
tial intellectual energy has been focused on vitalism in more than fifty years.
That the human body is a physical machine, operating according to under¬
stood physical laws and mechanisms, sets the stage for considering the
mechanistic nature of thought and intelligence.

Reason versus Emotion and Feeling #1: 1870-

The basic separation of the heart from the head occurred long ago and is a
fundamental part of Christian folk psychology. It is background. What con¬
cerns us is the ascription of reason (cold logic) to machines and the belief
that a machine could have no heart—no feelings or emotions—to ever con¬
flict with its reason. I do not seem to find any good way to fix the initiation of
this issue. The striking characteristic of the golem of Rabbi Loew in 1580
194 ALLEN NEWELL

seemed to have been literal-mindedness, not heartlessness. And nineteenth-

century artificial humans seemed to combine all the human attributes, as did,
for instance, Frankenstein’s constructed monster. [Shelley, 1818.] But by
the twentieth century, certainly in R.U.R. (Rossum’s Universal Robots), we
clearly have the intelligent robot, who is without soul, hence, without emo¬
tions or independently felt wants. [Capek, 1923.] So I have split the latter
two dates and taken 1870 as the start.
The relevance of this for AI is in providing a basis for separating machines
from humans that is different from the issue of purpose. Although a birth¬
right issue of AI, it does not play a major role. That the issue is there can be
seen clearly enough in the paper on “Hot Cognition” by Abelson, which put
forth some proposals on how to move machine intelligence in the direction
of having affect. [Abelson, 1963.] The lack of prominence stems in part, no
doubt, from the strong engineering-orientation of AI, which emphasizes use¬
ful mental functions (e.g., problem-solving and learning). In agreement with
this, Abelson is one of the few social psychologists associated with AI, and
the paper was given at a psychology conference. Thus, this issue remains in
the background, waiting to become prominent at some future time.

Philosophy versus The Science of Mind: 1870-1910

For science as a whole, the separation from philosophy and the acceptance
of empiricism as a fundamental tenet occurred centuries ago. For psychol¬
ogy, this occurred very recently, in the last decades of the nineteenth cen¬
tury. Indeed, psychology celebrates the establishment of the first experi¬
mental laboratory (Wundt’s in Leipzig) in 1879. It was not an especially
difficult passage for psychology, given the rest of science as a model. It can
be considered complete by the rise of behaviorism, say, by Watson’s classic
paper. [Watson, 1913.] Thus, this issue emerged and vanished before AI
began. The residue was a continuing tradition in philosophy concerned with
mind, which was completely distinct from work in psychology and, even
more so, from technology. This issue ensured that when AI did emerge,
which happened instantly on computers becoming sufficiently powerful,1 it
would be without more than peripheral involvement of the philosophy of
mind.

Logic versus Psychologic: 1910-1945

We continue to lay out the issues—and their resolutions—that were in effect

at the birth of AI. This issue concerns whether symbolic logic was to be
taken as revealing how humans think or whether humans use some sort of
unique psychologic. It surely started out with logic identified with

' A case can be made that serious AI started as soon as computers attained 4K of random-access
primary memory.
INTELLECTUAL ISSUES IN THE HISTORY OF Al 195

thought, as Boole’s classic monograph entitled The Laws of Thought tes¬

tifies. [Boole, 1854.] But logic was rapidly transformed from an explication
of the possible varieties of thinking to a device for probing the foundations
of mathematics. We can take the Principia Mathematica of Whitehead and
Russell as marking the completion of this transformation. [Whitehead and
Russell, 1910-1913.] The effect was to separate logic from psychology (and
also from the philosophy of mind, although that is a more complex story).
Modern logic, of course, was integrally involved in the development of
the digital computer, and, thus, it enters into the history of AI. But logic did
not enter AI at all as the logic of thought; that separation remained. Logic
was part of the underlying technology of making mechanisms do things. In
fact, it was precisely the split of logic from thought that set logic on the path
to becoming a science of meaningless tokens manipulated according to for¬
mal rules, which, in turn, permitted the full mechanization of logic.
Thus the issue was really settled by 1910, and the status in the first half of
the century was that psychologic was not a significant item on the agenda of
any science. This, of course, was due to behaviorism’s restriction of psy¬
chology’s agenda. I have placed a date of 1945 for the ending of this issue;
this is really an ending of the phase of separating logic from thought. The
nerve-net model of McCulloch and Pitts can be used to mark this, along with
the work of Turing on which it depended. [Turing, 1936; McCulloch and
Pitts, 1943.] They attempted to show that physical systems that echo the
structure of the brain could perform all computations, which is to say, all
logical functions. Whether this is seen as saying more about the brain or
more about logic can be argued; in either case, it brought them back into
intimate contact. We might think that the ending of one phase of the issue
(the stable separation of logic from thought) should initiate a new phase,
namely, a new controversy over the exact nature of the connection. But it
did not happen that way. Rather, the issue was not discussed, and basic
questions about the mechanization of mind took the form of other issues.
The reason that happened cannot be explored here. In part, it comes from
the shift with AI from the characterization of the brain in computational
terms to the digital computer, where logic played a completely technical and
engineering role in describing sequential and combinational logic circuits.

Analog versus Digital: 1940-1970

When computers were first developed in the 1940s, they were divided into
two large families. Analog computers represented quantities by continuous
physical variables, such as current or voltage; they were fast, operated
simultaneously, and had inherently limited accuracy. Digital computers rep¬
resented quantities by discrete states; they were slow, operated serially, and
had inherently unlimited accuracy. There was a certain amount of skirmish¬
ing about which type of computer was better for which type of job. But the
technical opinion-leaders maintained a view of parity between the two
196 ALLEN NEWELL

families—each for its own proper niche. Inevitably, there arose hybrid com¬
puters, which claimed to have the best of both worlds: digital control and
memory coupled with analog speed and convenience.
It was all over by 1970. The field of computers came to mean exclusively
digital computers. Analog systems faded to become a small subpart of elec¬
trical engineering. The finish was spelled not just by the increased speed and
cost-efficiency of digital systems, but by the discovery of the Fast Fourier
Transform, which created the field of digital signal processing and thus
penetrated the major bastion of analog computation. The transformation of
the field is so complete that many young computer scientists hardly know
what analog computers are.
The main significance of this issue, with its resolution, was to help create
the discipline of computer science and separate it from electrical engineer¬
ing. Its effect on AI lies mostly in the loss of an analytical point of view, in
which the contrast between analog and digital computation is taken as the
starting point for asking what sort of information-processing the nervous
system does. An admirable example of this point of view can be seen in the
notes for von Neumann’s Silliman Lectures, published posthumously, [von
Neumann, 1958.] This style of analysis belongs to the world of cybernetics
and not to that of AI. I doubt if many young AI scientists have read von
Neumann’s little book, though it was highly regarded at the time, and von
Neumann was one of the towering intellects of the computer field.

Symbols versus Numbers: 1955-1965

We now come to the first of the issues that characterizes AI itself, as op¬
posed to the background against which it emerged. The digital-computer
field defined computers as machines that manipulated numbers. The great
thing was, its adherents said, that everything could be encoded into num¬
bers, even instructions. In contrast, scientists in AI saw computers as ma¬
chines that manipulated symbols. The great thing was, they said, that every¬
thing could be encoded into symbols, even numbers. The standard measure
of a computation at the time was the number of multiplications it required.
Researchers in AI were proud of the fact that there were no multiplications
at all in their programs, though these programs were complex enough to
prove theorems or play games. The issue was actively pursued as a struggle
over how the computer was to be viewed. However, it was joined in an
asymmetric way. The bulk of the computer field, and all its responsible
opinion-leaders, simply adopted the view that computers are number
manipulators. There was no attempt to argue against the view that com¬
puters are symbol manipulators. It was just ignored, and the standard inter¬
pretation maintained. Researchers in AI, on the other hand, were actively
eng&ged in promoting the new view, considering the standard one to be a
radical misreading of the nature of the computer and one that provided a
significant barrier to the view that computers could be intelligent.
The result ot this clash of views was to isolate AI within computer sci-
INTELLECTUAL ISSUES IN THE HISTORY OF Al 197

ence. AI remained a part of computer science, but one with a special point of
view that made it somewhat suspect, indeed somewhat radical. This isola¬
tion is important historically, for it has affected the professional and disci¬
plinary organization of the two fields. It derives ultimately, no doubt, from a
basic divergence of views about whether computers can or cannot exhibit
intelligence. This overarching issue, of course, continued to be important on
its own, as witnessed by the debates that occurred throughout the 1950s on
whether machines could think. But the more specific issues that it spawned
also had independent lives.
The issue of symbols versus numbers did not arise until after the first AI
programs came into existence, circa 1955. Before that time, programs were
classified as numerical versus nonnumerical. This latter class was a miscel¬
lany of all the things that processed data types other than numbers—
expressions, images, text, and so forth.2 This included the few game-playing
and logic programs but much else as well. The symbols-versus-numbers
issue emerged only when a positive alternative became formulated, that is,
symbolic manipulation. This was not a synonym for nonnumerical process¬
ing, for it laid the groundwork for the separation of image- and text¬
processing from AI. Indeed, the work on machine translation, which started
in the early 1950s, was initially considered as one strand in the development
of intelligence on machines. [Locke and Booth, 1957.] But that effort be¬
came concerned with text and not symbols and developed its own identity as
computational linguistics. (All of this, of course, was before text processing
in its current meaning emerged—an event that bore no significant relation to
the development of computational-linguistics.)
I have placed the ending of this issue at about 1965, although I do not have
a significant marker event for its demise. The issue is certainly not alive now
and has not been for a long time. In part, this is due to the prominence of
many nonnumerical data types in computer science generally, such as text
and graphics. These make the characterization of computers as number
manipulators no longer ring true. In part, it is due to the shift within theoret¬
ical computer science to algebraic and logical formalisms, with the concur¬
rent retreat of numerical analysis from its early dominant role. In part, of
course, it is due to the success of AI itself and the demonstrations it brought
forward of the symbolic character of computation. It is tempting to say that
the cause was simply the growth of scientific understanding—but such rea¬
sons do not fare well in historical accounts. In any event, my recollection is
that the symbols/numbers issue was no longer prominent by the late 1960s,
though a little historical digging might place it five years later.

Symbolic versus Continuous Systems: 1955-

An important characterization of a science, or an approach within a science,

is the class of systems it uses to construct its theories. Classical physics, for

2The concept of data type did not arrive in clear form until much later.
198 ALLEN NEWELL

instance, viewed systems as being described by systems of differential equa¬

tions. Given a new phenomenon to be explained, a physicist automatically,
without a thought, used differential equations to construct his or her theory
of that phenomenon. Mathematical psychology in the 1950s and 1960s could
be characterized by its acceptance of Markov processes as the class of
systems within which to seek theories of particular phenomena.
The issue is within what class of systems should a description of intelli¬
gent systems be sought. On one side were those who, following the lead of
physical science and engineering, adopted sets of continuous variables as
the underlying state descriptions. They adopted a range of devices for ex¬
pressing the laws—differential equations, excitatory and inhibitory net¬
works, statistical and probabilistic systems. Although there were important
differences between these types of laws, they all shared the use of continu¬
ous variables. The other side adopted the programming system itself as the
way to describe intelligent systems. This has come to be better described as
the class of symbolic systems, that is, systems whose state is characterized
by a set of symbols and their associated data structures. But initially, it was
simply the acceptance of programs per se as the theoretical medium.
Adopting a class of systems has a profound influence on the course of a
science. Alternative theories that are expressed within the same class are
comparable in many ways, but theories expressed in different classes of
systems are almost totally incomparable. Even more, the scientist’s intui¬
tions are tied strongly to the class of systems he or she adopts—what is
important, what problems can be solved, what possibilities exist for theoret¬
ical extension, and so forth. Thus, the major historical effect of this issue in
the 1960s was the rather complete separation of those who thought in terms
of continuous systems from those who thought in terms of programming
systems. The former were the cyberneticians and engineers concerned with
pattern recognition; the latter became the AI community. The separation has
been strongly institutionalized. The continuous-system folk ended up in elec¬
trical-engineering departments; the AI folk ended up in computer-science
departments. (It must be remembered that initially computer-science depart¬
ments were almost exclusively focused on software systems and almost all
concern with hardware systems was in electrical-engineering departments.)
I believe this issue largely explains one peculiar aspect of the organization
of the science devoted to understanding intelligence: By almost any account,
pattern recognition and AI should be a single field, whereas they are almost
entirely distinct. By now, in fact, due to another important historical twist,
many people in computer science work in pattern recognition. But if such
people also know traditional pattern recognition, they are seen as interdisci¬
plinary.
Another interesting implication is buried here. The issue is not properly
dichotomous, for there exist other classes of systems within which to search
for intelligent systems. One obvious candidate is logic.3 Were there not

3In fact, there are additional possibilities. [Newell, 1970.]

INTELLECTUAL ISSUES IN THE HISTORY OF Al 199

scientists who believed that logic was the appropriate class of systems? And
if not, why not? First, by logical systems is meant the class of systems that
do logical operations, such as AND, OR, NOT, and so forth.4 This is the
class corresponding to the logic level in the hierarchy of computer struc¬
tures. The logic level is located between the circuit level and the program
(symbol) level. All three levels are equally comprehensive and provide three
possibilities for ways of describing intelligent systems. Indeed, circuit and
program levels correspond exactly to the continuous and symbol positions
of the issue under discussion. Now, in fact, in the early days, there were
attempts to build logic machines and discuss the behavior of systems di¬
rectly in terms of logic circuits. The classical neural networks of McCulloch
and Pitts were an effort at modeling the neural system at the logic level.
[McCulloch and Pitts, 1943.] But all these efforts rapidly died out and were
all but gone by the mid-1960s. My own guess about why this happened is that
the hierarchy of computer levels indicated quite clearly what to do with a
logic level—namely, compose a higher level system. But this implied simply
reproducing existing program-level systems, at least without some new or¬
ganizational ideas at the program level. But the logic level provided no such
ideas, nor could it. Thus, there was nowhere to go. In fact, the history of
these efforts seems quite obscure, and tracing the demise of logic as a
system language for intelligent systems would be a substantial, though re¬
warding, undertaking.

Problem-Solving versus Recognition #1: 1955-1965

An interesting issue grew up in association with the continuous/symbolic

split. Those thinking within the framework of continuous systems concen¬
trated on pattern recognition as the key type of task for machines to do—
character recognition, speech recognition, and visual-pattern recognition.
They also often concentrated on learning (as noted in the following para¬
graphs), but it was almost always a recognition capability that was being
learned. The Perceptron of Rosenblatt can be taken as paradigmatic here.
[Rosenblatt, 1958.] Contrariwise, those thinking within the framework of
symbolic systems concentrated on problem-solving as the key type of task
for machines to do—game-playing, theorem-proving, and puzzle-solving.
This separation of tasks reinforced the split between these groups. To the
AI community, the intellectual depth of the tasks performed by the pattern-
recognition systems seemed relatively trivial compared with the problem¬
solving tasks done by the programming systems. But just because of that, a
myth grew up that it was relatively easy to automate man’s higher reasoning
functions but very difficult to automate those functions man shared with the
rest of the animal kingdom and performed well automatically, for example,

4 It might also mean the class of theorem-proving systems using logical calculi; but this is really
a subclass of symbol systems.
200 ALLEN NEWELL

recognition. Thus, work on recognition was at the foundation of the problem

of intelligence, whereas work on problem-solving was an add-on.
The symbolic/continuous split and the problem-solving/recognition split
are organically related. Each task is the one most easily approached in terms
of the class of systems adopted. However, that does not make the two
intellectual issues the same. Scientists can hold quite different attitudes
about the two splits, and the two issues can become uncoupled in a different
era under different conditions. Both these issues emerged in the late 1950s
concurrently with the birth of AI. By 1965 the two fields of AI and pattern
recognition had separated rather completely and taken up distinct, relatively
permanent institutional roles. The conflict could be considered to have
reached a resolution. However, it was to become unstuck again almost im¬
mediately.

Psychology versus Neurophysiology #1: 1955-1965

Strongly coordinated with the issues of symbolic versus continuous systems

and problem-solving versus recognition was another, conceptually distinct
issue, namely, whether AI would look to psychology or to neurophysiology
for inspiration. That human intelligence was to be both guide and goad to
engineering intelligent systems was clear. However, this did not discrimi¬
nate between psychology and neurophysiology. As is well known, these two
disciplines speak with entirely separate, though not necessarily contradic¬
tory, voices. In general, those concerned with continuous systems and pat¬
tern recognition looked to neurophysiology; those concerned with symbolic
systems and problem-solving (i.e., AI) looked to psychology. Evidence of
the exclusive attention of early AI to psychology (in contradistinction to
biology) is amply provided by the two major sets of readings of those years.
[Feigenbaum and Feldman, 1963; Minsky, 1968.] By 1965, this issue was no
longer a live one, and the cast for AI was set.
The split between neurophysiology and psychology did not dictate the
split between symbolic and continuous systems; if anything, it was the other
way around. Neurophysiology, of course, was linked to continuous vari¬
ables, with its signals, networks, and geometry. But experimental psychol¬
ogy was not linked at all to symbolic systems. The dominant class of systems
in psychology at the time was that of stimulus/response (S/R) systems, an
abstract form of inhibition-and-excitation network. The only alternatives
were the continuous fields of Gestalt theory or the pseudo-hydraulic systems
of Freudian psychology (both only vaguely defined, though that is irrelevant
here). In fact, the class of symbolic systems was discovered within AI and
imported into psychology. [Newell and Simon, 1976a; Newell, 1980b.]
Thus, the choice of psychology by AI was made because the class of systems
that AI took to work with, that is, programming systems, led to psychologi¬
cally, not physiologically, revealing tasks.
Neurophysiology played a key role in keeping continuous systems from
INTELLECTUAL ISSUES IN THE HISTORY OF Al 201

suffering the same fate as logic systems. Whereas with logic systems there
was nowhere to go except toward program-like organizations, with continu¬
ous systems there was the brain to model. We need not demand an answer to
what the higher organization would be, we could just take as guide the brain
as revealed in current neurophysiological work. It is true, of course, that in
the late 1940s and early 1950s, the discrete approximation to the nervous
system (neurons as digital threshold devices) promised to provide neuro¬
physiological inspiration for the class of logic systems. But under a barrage
of criticism, even the engineers came to accept the nervous system as too
complex to be modeled by logic-level systems, which is to say, its con¬
tinuities had to be taken seriously. Thus, without any source of inspiration,
logic-level systems faded away as a separate language for modeling intelli¬
gence, but continuous systems remained.

Performance versus Learning #1: 1955-1965

Yet another issue can be identified that is coordinated with the issue of
symbolic versus continuous systems. AI concentrated on creating perform¬
ance systems, that is, systems that performed some task demanding intelli¬
gence. Cybernetics and pattern-recognition research concentrated on creat¬
ing systems that learned. Indeed, another subfield grew up that called itself
self-organizing systems. [Yovits, Jacobi, and Goldstein, 1962.] In practice,
self-organizing systems largely overlapped with the work in pattern recogni¬
tion and it had common roots in cybernetics. But self-organizing systems
took the problem of learning as the central focus rather than the problem of
recognition. For instance, within self-organizing systems, there was consid¬
erable interest in embryology, even though it had little to do with recognition
at the time.
Through the early 1960s, all the researchers concerned with mechanistic
approaches to mental functions knew about each other’s work and attended
the same conferences. It was one big, somewhat chaotic, scientific happen¬
ing. The four issues I have identified—continuous versus symbolic systems,
problem-solving versus recognition, psychology versus neurophysiology,
and performance versus learning—provided a large space within which the
total field sorted itself out. Workers of a wide combination of persuasions on
these issues could be identified. Until the mid-1950s, the central focus had
been dominated by cybernetics, which had a position on two of the issues—
using continuous systems and orientation toward neurophysiology—but no
strong position on the other two. For instance, cybernetics did not concern
itself with problem-solving at all. The emergence of programs as a medium
of exploration activated all four of these issues, which then gradually led to
the emergence of a single composite issue defined by a coordination of all
four dimensions. This process was essentially complete by 1965, although I
do not have any marker event. Certainly by 1971, at the second International
Joint Conference on Artificial Intelligence in London, it was decided that
202 ALLEN NEWELL

henceforth the conference would not accept pure pattern-recognition pa¬

pers, an act which already reflected an existing state of affairs.

Serial versus Parallel #1: 1955-1965

It is worth noting for future reference that most pattern-recognition and self¬
organizing systems were highly parallel network structures. Many, but not
all, were modeled after neurophysiological structures. Most symbolic-
performance systems were serial programs. Thus, the contrast between se¬
rial and parallel (especially highly parallel) systems was explicit during the
first decade of AI. The contrast was coordinated with the other four issues I
have just discussed. However, I do not recollect it playing nearly as active a
role as any of the other four, so I have simply added it on as a comment.

Heuristics versus Algorithms: 1955-1965

These issues were not the only ones that emerged in the first decade of AI’s
existence, nor the most important. A candidate for the most important initial
issue was AI’s development of heuristic programs in contradistinction to
algorithms. Algorithms were taken to be programs that guaranteed that they
would solve a problem or solve it within given time bounds. Good programs
were algorithmic, and if not, the fault lay with the programmer, who had
failed to analyze his or her problem sufficiently—to know what the program
should do to solve this problem. Heuristic programs, on the other hand,
were programs that operated by means of heuristic rules of thumb-
approximate, partial knowledge that might aid in the discovery of the solu¬
tion but could not guarantee to do so. The distinction implied that intelligent
problem-solving could be attained by heuristic programs. For a short while,
one name for the field of AI was heuristic programming, reflecting, in part, a
coordination with such subfields as linear programming and dynamic pro¬
gramming (which were also just then emerging).
An important effect of this issue was to isolate AI within computer sci¬
ence but along a different dimension than the issue of symbols versus num¬
bers. Heuristic programming indicates a commitment to a different course
than finding the best engineering solution or mathematical analysis of a
problem. According to the standard engineering ethos, the proper use of the
computer requires the engineer or analyst to exert his or her best intellectual
efforts studying the problem, find the best solution possible, and then pro¬
gram that solution. Providing a program with some half-baked, unanalyzed
rules seemed odd at best and irrational, or even frivolous, at worst. A good
example ot this tension can be found in the work of Wang, whose theorem¬
proving program performed much better than the LOGIC THEORIST.
[Newell, Shaw, and Simon, 1957; Wang, I960.] The thrust of Wang’s posi¬
tion was that much better theorem-provers could be built if appropriate
results in mathematical logic were exploited. The defense by the AI commu-
INTELLECTUAL ISSUES IN THE HISTORY OF Al 203

nity stressed finding how humans would solve such problems, in effect deny¬
ing that the fullest analysis of experimental tasks was the object of the
investigation. Another important example was the MACSYMA project to
construct an effective computer system for physicists and engineers to do
symbolic manipulation of mathematical expressions. Although this work
grew out of two prior efforts in AI, it was cast by its leaders as “not part of
AI,” but, rather, as part of an area of computer science called symbolic
manipulation, which took a thoroughgoing engineering and analytical at¬
titude. [Slagle, 1963; Moses, 1967.]
I have put the demise of the issue at the mid-1960s; the issue gradually
ceased to be discussed, though the distinction continues to be made in
textbooks and introductory treatments. Once the field was underway, with
lots of AI systems to provide examples, the point at issue became transpar¬
ent. Moreover, the distinction has difficulty in being transformed into a
technical one, because it is tied to features external to the procedure itself,
namely, to the problem that is supposed to be solved and the state of knowl¬
edge of the user of the procedure.

Interpretation versus Compilation: 1955-1985

A third issue served to separate AI from the rest of computer science, in

addition to the issues of symbols versus numbers and heuristics versus
algorithms. AI programs were developed in list-processing languages, which
were interpretive, whereas the mainstream of language development was
moving irrevocably toward the use of compilers. Prior to the mid-1950s,
programming languages beyond assemblers were interpretive. The major
turning point in compilers, FORTRAN, was developed in the mid-1950s,"
and it determined the direction of programming-language development
(though, of course, not without some controversy). Speed of execution was
the consideration uppermost in the minds of the programming fraternity. In
contrast, AI took the interpretive character of its languages seriously and
declared them to be necessary for attaining intelligent systems. This was
epitomized by the use of full recursion, but it penetrated throughout the
entire philosophy of language design, with the attractive idea of putting
intelligence into the interpreter.
This separation of AI programming from mainline high-level language
programming, which started immediately at the birth of AI, has persisted to
the present. Its effects go much deeper than might be imagined. This separa¬
tion has played a major role in determining the heavy AI involvement in
interactive programming, which contrasts with the minimal involvement of
the central programming-languages, with their adherence to the compile-

5 In fact, the first report of FORTRAN at a scientific meeting occurred at the same session as the
first report of a list-processing language. [Backus et al., 1957; Newell and Shaw, 1957 ]
204 ALLEN NEWELL

and-run operating philosophy. Just for fun, I have indicated the end of this
issue in 1985, on the assumption that the coming generation of powerful
personal computers will finally force all languages to come to terms with full
dynamic capabilities in order to permit interactive programming. But this is
pure conjecture, and the separation may now be wide enough to require a
generation to heal.
The grounds for this issue can be traced to demands for efficiency on the
one hand versus demands for flexibility on the other; perhaps the issue
should have been so labeled. For instance, the main programming commu¬
nity in the late 1950s also had a strong negative reaction to list-processing,
because of its giving up half the memory just to link the actual data together.
But, although the general efficiency issue was always on the surface of
discussions, the total situation seems better described in terms of distinct
structural alternatives, that is, interpreters versus compilers, list structures
versus arrays, and recursion versus iteration.

Simulation versus Engineering Analysis: 1955-

One issue that surfaced right from the start of AI was whether to make
machines be intelligent by simulating human intelligence or by relying on
engineering analysis of the task. Those who were primarily trying to under¬
stand human intelligence inclined naturally to the simulation view; those
who were primarily engineers inclined to the pure task-analysis view. The
principle was frequently invoked that we do not build a flying machine by
simulating bird flight. On the simulation side, there was more than one
position. The majority took the view that casual observation and casual
introspection was the appropriate approach—that is, the human was a
source of good ideas, not of detail. A few, usually with strong psychological
interests or affiliations, took the view that actual experimental data on hu¬
mans should be examined.
This issue seems never to have produced any important crises or changes
of direction in the field; however, it has probably decreased the amount of
mutual understanding. There seems to be little movement in a scientist’s
position on this issue. Each investigator finds his or her niche and stays
there, understanding only superficially how those with different approaches
operate. The position adopted probably reflects fairly deep attitudes, such as
determine whether a scientist goes into an engineering discipline or a social/
behavioral discipline in the first place. This is to be contrasted with many
fields where methods are effectively neutral means to ends, to be used by all
scientists as the science demands. There is little indication of diminution of
this issue over the years, although starting in the 1970s, there has been some
increase in the general use of protocols to aid the design of AI systems, even
when there is no psychological interest.
This completes the set of new issues that arose coincident with the birth
of AI. Five of them—symbolic versus continuous systems, problem-solving
INTELLECTUAL ISSUES IN THE HISTORY OF Al 205

versus recognition, psychology versus neurophysiology, performance ver¬

sus learning, and serial versus parallel—separated AI from other endeavors
to mechanize intelligence. But the goal of mechanizing intelligence bound all
of these enterprises together and distinguished them from the greater part of
computer science, whose goal was performing tasks in the service of man¬
kind. Three issues—symbols versus numbers, heuristics versus algorithms,
and interpreters versus compilers—clustered together to make AI into a
relatively isolated and idiosyncratic part of computer science. Finally one-
simulation versus engineering—was purely internal to AI itself.

Replacing versus Helping Humans: 1960-

An issue that surfaced about five years after the beginning of AI was whether
the proper objective was to construct systems that replace humans entirely
or to augment the human use of computers. The fundamentally ethical di¬
mension of this issue is evident. Yet, it was not overtly presented as an issue
of social ethics but, rather, as a matter of individual preference. An inves¬
tigator would simply go on record one way or another, in the prefaces of his
or her papers, so to speak. Yet, there was often an overtone, if not of ethical
superiority, of concordance with the highest ideals in the field. Those whose
inclinations were toward AI did not so much meet this issue head on
as ignore it. Indeed, it was perfectly possible to take the view that work
in AI constituted the necessary exploration for man/computer symbiosis.
[Licklider, I960.]
A relatively weak issue such as this could not really become established
unless man/machine cooperation offered technical possibilities and chal¬
lenges as exciting as constructing intelligent machines. Thus, the beginning
of this issue coincides with the appearance of interesting interactive sys¬
tems, such as SKETCHPAD, which had an immense influence on the field.
[Sutherland, 1963.]
Artificial intelligence scientists have had a relatively large involvement in
the development of user/computer interaction throughout the history of
computer science; for example, in time-sharing in the 1960s and 1970s, in
making languages interactive in the 1970s, and in developing personal ma¬
chines in the early 1980s. One explicit justification given for this involvement
was that AI itself needed much better programming tools to create intelligent
programs—a reason quite independent of the issue presented heie. How¬
ever, it is not possible to untangle the relations between them without some
rather careful historical analysis.
Many of those who opted for working in user/computer cooperation
tended not to become part of AI as the latter gradually evolved into a field.
However, as I have already noted, it was entirely possible to work in both AI
and user/computer cooperation. Still, the net result was an additional factor
of separation between those in AI and those in neighboring parts of com¬
puter science.
206 ALLEN NEWELL

Epistemology versus Heuristics: 1960-

It is easy to distinguish the knowledge that an intelligent agent has from the
procedures that might be necessary to put that knowledge to work to exhibit
the intelligence in action.6 The initial period in AI was devoted almost exclu¬
sively to bringing into existence modes of heuristic processing worthy of
consideration. In 1959, John McCarthy initiated a research position that
distinguished such study sharply from the study of appropriate logical for¬
malisms to represent the full range of knowledge necessary for intelligent
behavior. [McCarthy, 1959.] This study was clearly that of epistemology—
the study of the nature of knowledge. It bore kinship with the subfield of
philosophy by the same name, although, as with so many other potential
connections of AI and philosophy, the orientation of the two fields is highly
divergent, although the domain of interest is nominally the same.
There has been little controversy over this issue, although the two poles
led to radically different distributions of research effort. Work on epistemol¬
ogy within AI has remained extremely limited throughout, although recently
there has been a substantial increase. [D. G. Bobrow, 1980.]

Search versus Knowledge: 1965-1980

In the first years of AI, through the early 1960s, AI programs were character¬
ized simply as highly complex programs, without any particular notion of
common structure. For instance, the field was also called complex informa¬
tion processing as well as heuristic programming. By 1965, however, it had
become clear that the main AI programs used the same fundamental tech¬
nique, which became known as heuristic search. [Newell and Ernst, 1965.]
This involves the formulation of the problem to be solved as combinatorial
search, with the heuristics cast in specific roles to guide the search, such as
the selection of which step to take next, evaluation of a new state in the
space, comparison of the present state to the posited goal-state, and so on.
As the scope of AI programs seemed to narrow, there arose a belief in some
AI scientists that the essence of intelligence lay not in search, but in large
amounts of highly specific knowledge, or expertise. This issue was well
enough established by the mid-1970s to occasion the declaration that a
paradigm shift in AI had already occurred, the original paradigm having been
heuristic search with little knowledge of the task domain and the new
paradigm being knowledge-intensive programs. [Goldstein and Papert,
1977.]
It may be doubted that these changes amounted to an actual paradigm
shift. What clearly did happen was a major expansion of AI research to

Said this way, the connection of this issue to the competence/performance issue discussed
later would seem to be overwhelming. However, the research programmes associated with the
two issues have never made common cause.
INTELLECTUAL ISSUES IN THE HISTORY OF Al 207

explore systems that included substantial domain-specific knowledge. The

subfield currently called expert systems, which includes many of the at¬
tempts at constructing applied AI systems, emerged in the mid-1970s in part
as a result of this emphasis. However, it became clear that heuristic search
invariably continued to show up in these programs. Whenever it did not, the
problems being solved by the AI system were extremely easy relative to the
knowledge put into the system.
It is useful to see that two types of searches are involved in intelligence.
The first is the search of the problem space, that is, heuristic search, which is
combinatorial. The second is the search of the system’s memory for knowl¬
edge to be used to guide the heuristic search. This memory search is through
a pre-existing structure that has been constructed especially for the purpose
of being searched rapidly; it need not be combinatorial. Both types of
searches are required of an intelligent system, and the issue of search versus
knowledge helped to move the field to a full consideration of both types. The
net result was not so much a shift in the paradigm as a broadening of the
whole field. This had become clear enough to the field so that by 1980
the issue can be declared moot.

Power versus Generality: 1965-1975

Another way to characterize the major early AI programs is that they took a
single well-defined difficult task requiring intelligence and demonstrated that
a machine could perform it. Theorem-proving, chess and checkers playing,
symbolic integration, IQ-analogy tasks, and such management-science tasks
as assembly-line balancing—all these fit this description. Again, there was a
reaction to this. Although AI could do these sorts of tasks, it could not do the
wide range of presumably trivial tasks we refer to as having common sense.
The need was for generality in AI programs, not power.
This call had been issued early enough. [McCarthy, 1959.] However, it
was really not until the mid-1960s that a significant shift occurred in the field
toward the generality and commonsense side. This gave rise to using small
constructed puzzles and artificial problems to illustrate various components
of everyday reasoning. A typical example was the monkey-and-bananas
task, patterned after simple tasks solved by Kohler’s chimpanzee, Sultan.
Whereas such problems would have seemed insignificant in the early years,
they now became useful, because the goal of research was no longer power,
but understanding how commonsense reasoning could occur.
By 1975, this shift had run its course, and new concerns for working with
relatively large-scale real problems took over with the development ot ex¬
pert systems already mentioned. As could have been expected, the end of
this period of emphasis did not mean a shift back to the original issue.
Although expert systems tackled real problems and, hence, were obviously
powerful, they did not achieve their power by the heuristic-search tech¬
niques of the early years; instead they used large amounts of domain-specific
knowledge (coupled, sometimes, with modest search).
208 ALLEN NEWELL

However, as is usual in the history of science, work on powerful AI

programs never stopped; it only diminished and moved out of the limelight.
By 1975, highly successful chess programs emerged, built on heuristic-
search principles, with an emphasis on large amounts of search—a million
positions per move in tournament play—and good engineering. Thus, intel¬
lectual issues shift the balance of what gets worked on but rarely shut off
alternative emphases entirely.

Competence versus Performance: 1965-

The Chomskian revolution in linguistics also started in the late 1950s. It was,
along with AI, just one of many similar and interrelated developments in
engineering, systems, and operational analysis. Although each of these de¬
velopments had a particularly intense significance for some particular field,
for example, linguistics or computer science, they all formed a common
interdisciplinary flux. Gradually, these activities sorted themselves into sep¬
arate subfields or disciplines, developing opposing positions on the issues
previously laid out, as we have seen for AI vis-a-vis cybernetics and pattern
recognition.
In many ways, linguistics was a special case. It was already a well-formed
discipline, and the revolution was at the heart of the discipline, not in some
peripheral aspect that could have split off and aligned with other intellectual
endeavors. Furthermore, only very few linguists participated in the general
flux that was occurring in the world of engineering and applied mathematics.
Linguistics was culturally and organizationally quite distinct, having strong
roots in the humanities. In fact, it probably made an immense difference that
Noam Chomsky became affiliated with the Massachusetts Institute of Tech¬
nology (MIT).
It was not until the mid-1960s that issues emerged that determined rela¬
tions between linguistics and other subfields and disciplines. A principal
issue was the distinction between competence and performance, which was
moved to a central position in the new linguistics by Chomsky. [Chomsky,
1965.] Linguistic competence was the general knowledge a speaker had of
the language, in particular, of the generative grammar of the language. Per¬
formance was the actual production of utterances, which could be affected
by many additional factors, such as cognitive limits, states of stress, or even
deliberate modifications for effect. The distinction made useful operational
sense for linguistics, because there were two sources of evidence about
human-language capabilities, the actual utterance and the judgment of gram-
maticality—a sort of recall/recognition difference, although that analogy
was never exploited.
This distinction might seem innocuous from the standpoint of science
history, that is, purely technical. In fact, it served to separate quite radically
the sciences concerned primarily with performance, namely AI, computa¬
tional linguistics, cognitive psychology, and psycholinguistics, from linguis-
INTELLECTUAL ISSUES IN THE HISTORY OF Al 209

tics proper. Linguistics itself declared that it was not interested in perform¬
ance. More cautiously said, competence issues were to have absolute
priority on the research agenda. But the effect was the same: Work in any of
the performance fields was basically irrelevant to the development of lin¬
guistics. There could be a flow from linguistics to these other fields, and,
indeed, there was an immense flow to psycholinguistics, but there could not
be any significant flow in the other direction.7
A more effective field-splitter would be hard to find. It has remained in
effect ever since, with the competence/performance distinction being ex¬
tended to other domains of mentality. This has certainly not been the only
significant cause of the separateness of AI from linguistics. There are impoi-
tant isolating differences in method, style of research, and attitudes toward
evidence. Many of these other issues share substance with the competence/
performance distinction and affect the separation between psychology and
linguistics much more than that between AI and linguistics. Thus, perhaps
these issues can be left to one side.

Memory versus Processing: 1965-1975

During the immediate postwar decades, the mainstream of individual human

psychology was strongly influenced by the general ferment of engineering,
system, and operational ideas (as I have previously termed it). This involved
human factors and information theory in the early 1950s; and signal-
detection theory, control theory, game theory, and AI in the mid-1950s. As
with linguistics in the period of 1955-1965, all these ideas and fields seemed
to mix while matters sorted themselves out. By the mid-1960s, psychology
had focused on memory as the central construct in its view of man as an
information processor. Short-term memory and the visual iconic store com¬
bined to provide an exciting picture of the interior block-diagram of the
human mental apparatus (what would now be called the architecture). This
settled what the main lines of investigation would be for the field; the marker
event for this conviction is Neisser’s book, Cognitive Psychology. [Neisser,
1967.]
This settlement is important for the history of AI, because AI’s influence
on psychology in the 1955-1965 period was primarily in the area of problem¬
solving and concept formation. With psychology opting for memory struc¬
ture, psychology and AI went fundamentally separate ways. Although the
work on problem-solving remained a common concern, it was a sufficiently
minor area in psychology, so that it exerted only a modest integrating effect.
AI itself during this period had little interest in memory structure at the
block-diagram level. Psychologically relevant research on memory by AI
researchers did exist but moved out of AI into psychology; for example, the

7This is not the whole story of the relations of linguistics with other fields; for example, there
have been important contacts with logic and philosophy.
210 ALLEN NEWELL

work on EPAM (Elementary Perceiver and Memorizer). [Simon and Feigen-

baum, 1964.]
In the second half of the 1960s came another major advance in cognitive
psychology, namely, the discoveries of how to infer basic processes from
reaction times. [Neisser, 1963; Sternberg, 1966.] This insight promised even
greater ability to dissect human cognitive processes and confirmed the basic
choice of psychology to analyze the block-diagram level of cognition. This
insight also broadened the analysis from just memory structure to the stages
of information-processing. In this respect, it might seem better to call the
issue under discussion one of system levels: AI focusing on the symbolic
level and psychology focusing on the architecture,8 that is, the equivalent of
the register-transfer level. However, the concern with memory so dominates
the years prior to 1965, when this issue was being sorted out, that it seems
preferable to label it memory versus processing.
Long-term memory has been absent from the previous account. During
this period, AI was certainly concerned about the structure of long-term
memory, under the rubric of semantic memory. This would seem to provide
common ground with psychology, yet initially it did not do so to any great
extent. Two factors seem to account for this. First, in psychology, the new
results, hence the excitement, all involved short-term memories. The estab¬
lished theory of learning, interference theory, against which these new ideas
about memory made headway, assumed a single memory, which was in
essence long-term memory. Second, the memory that psychology consid¬
ered was episodic—learning what happened during an episode, such as
learning what familiar items were presented at a trial. This stood in marked
contrast with semantic memory, which appeared to be a timeless organiza¬
tion of knowledge. Only gradually did the psychologically relevant work on
semantic memory by a few investigators capture any significant attention
within cognitive psychology. The seminal publication of Anderson and
Bower s Human Associative Memory can be taken as a marker of the begin¬
ning of this attention. [Anderson and Bower, 1973.]

Problem-Solving versus Recognition #2: 1965-1975

In 1965, AI took back the problem of recognition that had become the
intellectual property of the pattern-recognition community. This can be
marked rather precisely by the work of Roberts on the recognition of three-
dimensional polyhedra. [Roberts, 1965.] The essential features were two:
First, recognition was articulated, that is, the scene had to be decomposed
or segmented into subparts, each of which might need to be recognized to be
a different thing. Thus, the result of recognition was a description of a scene
rather than just an identification of an object. But a description is a symbolic

Although the term architecture is just now coming into common use in psychology.
 211
INTELLECTUAL ISSUES IN THE HISTORY OF Al

structure that has to be constructed, and such processes were quite outside
the scope of the pattern-recognition techniques of the time, though exactly
of the sort provided by AI. Second, a major source of knowledge for making
such recognitions came from adopting a model of the situation (e.g., it con¬
sists only of polyhedra). This made recognition processes strongly inferen¬
tial, again fitting in well with work in AI, but not with work in pattern
recognition.
By the late 1960s, work on vision was going on throughout AI, but the
transformation went further than just vision. Three laboratories (at MIT,
Stanford, and the Stanford Research Institute) started major efforts in robot¬
ics. Vision was to be coupled with arms and motion and in at least one AI
center (Stanford), with speech. The entire enterprise was radically different
in its focus and problems from the research in pattern recognition that was
still going on in parallel in departments and research centers of electrical
engineering. In fact, there was little actual controversy to speak of. Both
groups simply did their thing. But likewise, there was no substantial rap¬
prochement.

Syntax versus Semantics: 1965-1975

The Chomskian revolution in linguistics was strongly based on theory. Built

around the notions of generative and transformational grammar, it posited
three distinct components (or modules) for phonology, syntax, and seman¬
tics, each with its own grammar. The initial emphasis was on syntax, with
work on semantics much less well developed.9 Despite cautions from the
competence/performance distinction, the inference was clear from both the
theory and practice of linguistics—syntactic processing should occur in a
separate module independently of semantic processing. Indeed, what com¬
putational linguistics there was in association with the new linguistics in¬
volved the construction of programs for syntactic parsing.
In the late 1960s, a reaction to linguistics arose from within the AI and
computational linguistics communities. It took the form of denying the sep¬
aration of syntax and semantics in the actual processing of language. The
initial analysis of an utterance by the hearer was as much a question of
semantics as of syntax. Language required an integrated analysis by the
hearer and, hence, by the theorist. This reaction can be marked by the work
of Quillian, whose introduction of semantic nets was a device to show how
semantic processing could occur directly on the surface structure ot the
utterance (though presumably in conjunction with syntax). [Quillian, 196 .]
This reaction was grounded more broadly in the assertion of the impor¬
tance of processing considerations in understanding language, the very thing

9There was work on phonology, but the domain lay outside the range of interest of AI and. in
fact, of psychology as well.
212 ALLEN NEWELL

denied by the competence/performance distinction. It sought to put process¬

ing considerations into the mainstream of linguistic studies, the latter being
owned, so to speak, by the linguistics community. One result, as might have
been expected, was to compound the separation between linguistics, on the
one hand, and computational linguistics and AI, on the other. Another was
to create a stronger independent stream of work on language in AI with its
own basis.

Theorem-Proving versus Problem-Solving: 1965-

Theorem-proving tasks have always been included in the zoo of tasks

studied by AI, although the attention these tasks received initially was
sporadic. However, some logicians and mathematicians worked on theorem¬
proving in logic, not just as another task, but as the fundamental formalism
for understanding reasoning and inference. In the last half of the 1960s, with
the development of a logical formalism called resolution, this work in
theorem-proving took center stage in AI. [Robinson, 1965.] It seemed for a
time that theorem-proving engines would sit at the heart of any general AI
system. Not only was their power extended rapidly during this period, but a
substantial amount of mathematical analysis was carried out on the nature of
theorem proving in the predicate calculus. Even further, theorem-proving
programs were extended to handle an increasing range of tasks, for example,
question-answering, robot-planning, and program-synthesis.
A consequence of this success and viewpoint was that theorem-proving
was taken to be a fundamental category of activity distinct from other prob¬
lem-solving, with its own methods and style of progress. A good indicator
of this is Nilsson’s AI textbook, which divides all problem-solving methods
of AI into three parts: state-space search, problem-reduction (i.e., subgoals),
and predicate-calculus theorem-proving. [Nilsson, 1971.] It is not clear
whether this issue has been laid to rest by now or not. As recounted in the
following section, under the procedural/declarative issue, theorem-proving
has become much less central to AI since the mid-1970s. But theorem¬
proving and problem-solving still remain distinct research strands.

Engineering versus Science: 1965-

Computer science is torn by a fundamental uncertainty over whether it is an

engineering or science discipline. There is no doubt about the engineering
side; computer science designs and creates artifacts all the time. The doubt
exists on the nature of the science involved. Computer science certainly
studies intellectual domains that are not part of other disciplines. The ques¬
tion is whether or not they have the character of a science. However, the
dichotomy need not be accepted: A third alternative is that the unique intel¬
lectual domain of computer science is part of mathematics. Computer sci¬
ence would then join other engineering specialties, such as control theory
INTELLECTUAL ISSUES IN THE HISTORY OF Al 213

and information theory, which have their own characteristic mathematical

development.
Much rests on the putative outcome of this issue: What should computer
science be like in the future? Should departments of computer science be
part of the college of engineering or the college of arts and sciences? What
status should be accorded to various subdisciplines in computer science?
Can a thesis involve just a design? And more. The start of this issue coin¬
cides with the creation of departments of computer science in the mid-1960s,
which served to raise all these questions. Whether the issue will ever be laid
to rest is unclear, but it is certainly unlikely while the whole field grows
dynamically, with a continuing flood of new and destabilizing notions.
Artificial intelligence participates along with the rest of computer science
in the uncertainties over whether it is an engineering or science discipline.
However, the issue for AI has its own special flavor. AI participates with
many disciplines outside computer science in the attempt to understand the
nature of mind and intelligent behavior. This is an externally grounded
scientific and philosophic goal, which is clearly not engineering. Thus, the
nature of the science for AI is not really in doubt as it is for the rest of
computer science. However, this does not end the matter, for interactions
occur with other issues. For instance, to the extent that we are oriented
toward helping humans rather than replacing them, we may not wish to
accept the understanding of the nature of mind as a scientific goal, but only
as a heuristic device.
The orientation toward engineering or science can have major conse¬
quences for how a field devotes its energies. Currently, for example, an
important divergence exists in the subfield of computer vision. Should the
nature of the environment be studied to discover what can be inferred from
the optic array (a scientific activity); or should experimental vision systems
be constructed to analyze the data they generate within the framework of the
system (an engineering activity)? That both activities are legitimate is not in
question; which activity gets the lion’s share of attention is in dispute. And
there is some indication that an important determiner is the basic engineer¬
ing/science orientation of a given investigator.

Language versus Tasks: 1970-1980

The 1970s saw the emergence of concerted efforts within AI to produce

programs that understand natural language, amounting to the formation of a
subfield, lying partly in AI and partly in computational linguistics. The key
markers are the works of Woods and Winograd. [Woods, 1970; T. Wino-
grad, 1971]. This issue had been building for some time, as we saw in the issue
of syntax versus semantics.
The emergence of such a subfield is in itself not surprising. Natural lan¬
guage is clearly an important, even uniquely important, mental capability. In
addition to AI, there existed another relevant field, computational linguis-
214 ALLEN NEWELL

tics, concerned generally with the application of computers to linguistics.

Neither is it surprising that this subfield had almost no representation from
linguistics, although, of course, linguistics was of obvious central rele¬
vance.10 The syntax/semantics issue, which had reinforced the separation of
linguistics from AI, was a primary substantive plank in the programme of the
new subfield.
What is interesting was the creation of another attitude within a part of
AI, which can be captured by the issue of language versus tasks. Studying
the understanding of language was seen as a sufficient context for investigat¬
ing the nature of common sense. An important discovery was how much
knowledge and inference appeared to be required to understand even the
simplest sentences or short stories. Thus, the very act of understanding such
stories involved commonsense reasoning and, with it, the essence of general
human intelligence. Programs could be interesting as AI research, so the
attitude went, without doing any other task in addition to understanding the
presented language input. The effect of this strategic position was to sepa¬
rate the work in natural-language processing from the tradition in AI of
posing tasks for programs to do, where the difficulty could be assessed. The
issue did not occasion much discussion, although its effects were real
enough. The issue was masked by the fact that understanding by itself was a
difficult enough task for AI research to make progress on. No one could
object (and no one did) to not adding what seemed like an irrelevant second
difficult task for the system, which would simply burden the research en¬
deavor.

Procedural versus Declarative Representation #1: 1970-1980

Recall that resolution theorem-proving flourished in the late 1960s and bid
fair to become the engine at the center of all reasoning. In fact, it took only a
few years for the approach to come up against its limitations. Despite in¬
creases in power, relative to prior efforts, theorem provers were unable to
handle any but trivial tasks. Getting from logic to real mathematics—seen
always as a major necessary hurdle—seemed as far away as ever.
The reaction to this state of affairs became known as the procedural/
declarative controversy. Theorem provers were organized as a large homo¬
geneous database of declarative statements (clauses in resolution), over
which an inference engine worked to produce new true statements to add to
the database. This was the essence of a declarative representation of knowl¬
edge and its attractions were many. Its difficulty lay in the costs of process¬
ing. The inference engine treated all expressions in the database alike or,
more precisely, without regard for their semantics. There also seemed no

10Among the contributors to the first conference on Theoretical Issues in Natural Language
Processing, a series that became the forum tor this subfield, I can identify only one mainstream
linguist. [Schank and Nash-Webber, 1975.]
INTELLECTUAL ISSUES IN THE HISTORY OF Al 215

way for a theorem prover to be given information about how to solve prob¬
lems. These two features added up to a major combinatorial explosion. The
remedy—the procedural side of the issue—lay (so it was claimed) in encod¬
ing information about the task in procedures. Then knowledge would be
associated directly with the procedures that were to apply it; indeed, the
procedures would embody the knowledge and, thus, not have to be inter¬
preted by another inference engine. This would permit the appropriate guid¬
ance for problem-solving and, thus, keep the combinatorial explosion under
control.
There are irremediable flaws in both sides of the argument whethei
knowledge should be coded in procedural or declarative form, just as there
are irremediable flaws in both sides of the argument whether a program is
heuristic or algorithmic. Both procedural and declarative representations are
necessary to make any computation at all happen. In consequence, argu¬
ments over the issue were largely inconclusive, although they produced the
closest thing to a public issue-controversy in AI’s short history. However,
the effect on the course of AI research was enormous. First, work on
theorem-proving shrank to a trickle, with what remained mostly devoted to
nonresolution theorem-proving. Second, so-called planning languages
emerged as a result—PLANNER, QA4, CONNIVER, POPLAR, and so
forth. [Bobrow and Raphael, 1974.] These programming-language systems
were intended to provide a vehicle for writing the sorts of domain-
dependent, procedure-oriented theorem provers called for in the debate.
While that did not quite happen, these languages in themselves provided a
major conceptual advance in the field. The effects of this issue had about run
their course by 1980.

Frames versus Atoms: 1970-1980

In a paper that circulated widely before it was published in the mid-1970s,

Marvin Minsky raised the issue about the size of representational units in an
intelligent system. [Minsky, 1975.] Knowledge should be represented in
frames, which are substantial collections of integrated knowledge about the
world, rather than in small atoms or fragments. The basic issue is as old as
the atomistic associationism of British empiricism and the countering com¬
plaints of the Gestaltists. How are the conflicting requirements for units of
thought and contextual dependence to be reconciled9
This issue had hardly surfaced at all in the first decade of AI. List struc¬
tures, the basic representational medium, were in themselves neither atom¬
istic nor wholistic but adaptable to whatever representational constructs the
designer had in mind.11 But the coming to prominence of resolution-theorem-

11 This is because list structures approximate general symbolic systems. The neutrality is easily
confirmed in the continued and universal use of list-processing languages to realize systems o
all kinds along this dimension.
 216 ALLEN NEWELL

proving in the late 1960s brought with it as a side effect the clause as the unit
of representation. The clause was a primitive assertion that could not be
broken down into a conjunction of other assertions—primitive predicates P,
negations of primitive predicates ~P, disjunctions P or Q, implications P
implies Q, and so forth. The total knowledge of the system was to be repre¬
sented as the conjunction of clauses—that is, to use the old Gestaltist
phrase, as an And-sum of separate bits of knowledge.
Thus, the issue of size of representational unit grew out of the same
ground as the procedural versus declarative controversy, and, indeed, it was
articulated by the same group at MIT who had made most of the latter issue.
As is always the case, concern was, in fact, widespread but had been subor¬
dinated to other concerns. [Abelson, 1973; Norman, 1973; Schank, 1973.]
Minsky was the first one to give clear voice to the concern. The effect of the
paper was dramatic, despite the fact that the paper itself was entirely specu¬
lative and discursive. Throughout AI, the concept of the frame as the appro¬
priate data structure was widely embraced. By 1980, frame systems were an
established part of AI, and a very substantial fraction of the work in knowl¬
edge representation was involved in such systems.
Much follows on this development (in conjunction with the procedural/
declarative issue)—the rise of substantial research effort in knowledge
representation and the strengthening of renewed ties with philosophy.
[Brachman and Smith, 1980.] These efforts conjoin with those of AI epis¬
temology, discussed earlier. They raise some new issues, such as the rela¬
tion of philosophic work on meaning to directly inspired computational mod¬
els. But these issues have not yet jelled enough to be included in their own
right.

Reason versus Emotion and Feeling #2: 1970-

Philosophy has a long-standing concern with the mechanization of mind.

Indeed, under the rubric of the mind/body problem, it can be said almost to
own the problem, it having been bequeathed to philosophy by Descartes. In
its genesis, AI had very little involvement with philosophy, beyond the
background awareness that comes from participation in the general intellec¬
tual culture. No philosophers of mind were involved and no technical philo¬
sophical issues were dealt with. A glance at the content of the two fields
provides one obvious clue. The phenomena attended to in philosophy are
sensations as subjective experiences—raw feels, to use a bit of philosophic
jargon. A typical article is entitled “The Feelings of Robots.” [Ziff, 1959.]
Thus, though AI and philosophy of mind ostensibly deal with the same
pioblem, in fact they go after largely distinct phenomena.12

Another example is the problem of induction, where philosophy is concerned with the cer¬
tainty ot induction and AI is concerned with performing the inductions. [Newell, 1973c.]
INTELLECTUAL ISSUES IN THE HISTORY OF Al 217

The issue has not been especially active, but it has been raised.
[Gunderson, 1971.] It is argued that performance functions (i.e., those func¬
tions AI currently deals with, called program-receptive functions) can be
mechanized; but that sentient functions (i.e., feelings, called program-
resistant functions) cannot. Whether this will ever grow to a substantial
controversy is hard to tell at this point. It is certainly available as a reserve
position that can serve to separate AI from the philosophy of mind. It adds to
the general background concern, discussed in the first occurrence of this
issue, of the absence of emotion and feeling in the development of intelligent
systems.

Toy versus Real Tasks: 1975-

As noted in the power/generality issue, the field took a shift in the mid-1960s
away from powerful programs toward programs that could exhibit common
sense. Further, as noted in the language/tasks issue, this line further trans¬
muted to being concerned with understanding via the understanding of natu¬
ral language. Concomitantly, programs were often built to work on small
simple illustrative tasks or environments, usually puzzles or made-up
situations.
By the mid-1970s some systems had been developed that worked with real
tasks that had substantial intellectual content, to judge from their role in the
real world. The initial such system can be taken to be DENDRAL, which
determined the structural formula for chemical molecules, given the data on
the mass spectrogram.13 [Lindsay, Buchanan, Feigenbaum, and Lederberg,
1980.] DENDRAL began in the late 1960s and grew in power throughout the
early 1970s. It was joined in the mid-1970s by several systems that per¬
formed competently in real medical-diagnosis tasks, of which MYCIN was
the paradigm. [Shortliffe, 1974.] This was the immediate locus of expert
systems, which, as previously noted, grew up as part of the general empha¬
sis on knowledge in contrast to search. With it grew an attitude that AI in
general should no longer work on small illustrative, artificial tasks but that it
was time to work on real tasks. The simple artificial tasks came to be called
toy tasks, not just because the term conveys the contrast between childish
and grown-up pursuits, but also because stacking children’s blocks had be¬
come a favorite illustrative task environment.
The tension between basic research and application exists in all sciences
at all times. Sciences sometimes build institutional structures to contain the
tension. As we saw in the issue of science versus engineering, computer
science has kept its basic and applied components mixed together in a single
discipline, thus exacerbating the tension. The tension was, in fact, especially

13The other system often mentioned similarly is MACSYMA, the highly sophisticated program
at the Massachusetts Institute of Technology for doing symbolic mathematics. As mentioned
earlier, it had deliberately removed itself from being an AI program.
218 ALLEN NEWELL

severe for AI during the decade of the 1970s. The climate in Washington was
not benign for basic research in general, and there was sustained pressure
from AI’s primary government funding agency (DARPA—Defense Ad¬
vanced Research Projects Agency) to make AI pay off. That said, however,
the distinction between toy versus real tasks is not solely the distinction
between basic and applied research. Tasks taken from the real world and
performed by intelligent humans as part of their working lives carry a prima
facie guarantee of demanding appropriate intelligent activity by systems that
would perform them. It can be argued that such tasks are the appropriate
ones for AI to work on, even if the goal is basic research. Thus, the toy-
versus-real-tasks issue stands ambiguously for both meanings—basic versus
applied and irrelevant versus relevant basic science.

Serial versus Parallel #2: 1975-

By the mid-1970s, computer science had for some time been seriously ex¬
ploring multiprogramming and multiprocessing. These provided the ground¬
work for considering parallel systems for doing AI. A major instigation
occurred with the development of the Hearsay-II model of speech under¬
standing. [Lesser and Erman, 1977.] Hearsay-II comprised a number of
knowledge sources (acoustic, phonetic, phonological, lexical, syntactic, se¬
mantic, and pragmatic), each working concurrently and independently off a
common blackboard that contained the current working state about the ut¬
terance and each contributing their bit to the evolving recognition and react¬
ing to the bits provided by the others.
The Hearsay-II structure was certainly a parallel one, but it was at a level
of parallelism quite different from earlier network models, namely, a modest
number (tens) of functionally specialized processes. Furthermore, individual
processes remained fundamentally symbolic (even though lots of signal¬
processing was inherent in the speech-recognition task). Hearsay-II was
only one of several efforts to pursue the notion that an intelligent system
should be thought of in terms of communicating subprocesses rather than as
an individual serial machine. A metaphor arose for thinking about an intelli¬
gent system—the scientific community metaphor—which took the opera¬
tion of science, with its notion of cooperation, publication, experiment,
criticism, education, and so forth, as the appropriate model for intelligent
activity. Gradually, a group of people emerged interested in working on
distributed AI.

Performance versus Learning #2: 1975-

As noted earlier, learning was generally associated with work on pattern

recognition. With the split between problem-solving and recognition, work
on learning within AI declined. As always, it never stopped entirely. Indeed,
such is the basic fascination with learning processes, and with the belief that
INTELLECTUAL ISSUES IN THE HISTORY OF Al 219

they hold the key to intelligence, that each learning program that was con¬
structed received substantial attention.14 [Samuel, 1959; D. A. Waterman,
1970; Winston, 1970; Sussman, 1975.] However, each learning system was
relatively idiosyncratic, with its own interesting lessons, so that the whole
did not add up to a coherent effort for the field.
A reversal of this state of affairs developed by the late 1970s. It was
triggered by the spread of a class of programming systems, called produc¬
tion, or rule-based systems, which are used for both constructing expert
systems and analyzing human cognition. [Waterman and Hayes-Roth, 1978.]
To appreciate their role in the resurgence of work on learning, we must take
a step back. To create a learning system requires solving two research
problems. First, a space of potential performance programs must be created,
in which learning will constitute moving from one program to another,
searching for programs with better performance. If the space of programs is
too vast and irregular, then learning is, in effect, automatic programming,
and it becomes extremely difficult. If the space is too limited, then learning is
easy, but the performance programs are of little significance. Determining
the right space is, thus, a critical research activity. Second, given the space,
it is still necessary to design an interesting learning system, for the space
only lays out the possibilities. Thus, inventing the learning system is also a
critical research activity. A major reason why early AI learning-systems
seemed so idiosyncratic was that each made unique choices on both these
dimensions. Most important, doing research on learning was doing a double
task and taking a double risk.
A production system is composed entirely of a set of if-then rules (if such
and such conditions hold, then execute such and such actions). At each
instant, the rules that hold are recognized, and a single rule is selected to
execute. In such a system, the natural space of performance programs con¬
sists of subsets of if-then rules, and the primitive act of learning is to add a
new rule to the existing set (or sometimes to modify an existing rule in some
simple way, such as by adding another condition). This space of perform¬
ance programs is neither too limited nor too open, since it is easy to restrict
the rules to be learned to a special class. As a consequence, the first research
choice is essentially made for the researcher, who can then concentrate on
constructing an interesting learning program. Moreover, learning programs
will have much in common, since they now use similar spaces of per¬
formance programs. Indeed, this is just what happened in the late 1970s
as researchers began to construct a wide variety of small learning systems,
all built around variants of the production-system formalism. [Michalski,
Carbonell, and Mitchell, 1983.] It must be realized, of course, that such fo¬
cusing of effort does not remove the collective risk. If production systems

14Some other systems were built, which might have been viewed as learning systems, but,
instead, were taken simply to be performance programs in specialized task environments, for
example, induction programs.
220 ALLEN NEWELL

are the wrong program organization to be exploring, then the entire field is
moving down an unproductive path.

Psychology versus Neuroscience #2: 1975-

AI would appear to be at the mercy of the immense gulf that continues to

separate psychology and the biology of the brain. As each field continues to
progress—which both do dramatically—hopes continually spring up for new
bridging connections. No doubt at some point the permanent bridge will be
built. So far, although each increment of progress seems real, the gap re¬
mains disappointingly large.
It is possible that AI has a major contribution to make to this by exploring
basic computational structures at a level that makes contact with neural
systems. In the early instance of psychology versus neurophysiology (which
was before the term neuroscience had been coined), that possibility seemed
quite remote. The theoretical structures that did make contact with
neurophysiology were remote from the computational structures that preoc¬
cupied AI researchers. Then the split occurred, with pattern recognition all
but moving out of computer science.
In the mid-1970s, a new attempt began to connect AI with neuroscience,
initiated by the work of David Marr. [Marr, 1976.] The emphasis remained
on vision, as it had been in the earlier period. But the new effort was
explicitly computational, focusing on algorithms that could perform various
low-level vision functions, such as stereopsis. Although Marr’s effort was
new in many ways, and based on specific technical achievements, most of
the global issues of the earlier time reappeared. This work has now ex¬
panded to a larger group, which calls its work, among other things, the new
connectionism, and promises to be a substantial subfield again, this time
within AI.

Serial versus Parallel #3: 1980-

The new wave of neuroscience-inspired AI contains, of course, a commit¬

ment to highly parallel network structures. The issue of serial versus parallel
merits a separate entry here to maintain a clear contrast with the distributed
AI effort, which defined the second wave of concern with parallel systems.
In this third phase, the degree of parallelism is in the millions, and computing
elements in the network have modest powers; in particular, they are not
computers with their own local symbols. In the new structures, computation
must be shared right down to the roots, so to speak. The interaction cannot
be limited to communicating results of significant computations. Further¬
more, the communication media between elements are continuous signals,
and not just bits. However, unlike the earlier work, these new computational
systems are not to be viewed as neural nets; that is, the nodes of the network
are not to be put in one-to-one correspondence with neurons, but, rather,
with physiological subsystems of mostly unspecified character.
INTELLECTUAL ISSUES IN THE HISTORY OF Al 221

Problem-Solving versus Recognition #3: 1980-

Robotics has returned to AI after having left it for most of the 1970s. Perhaps
it is unfortunate to call the issue problem-solving versus recognition, since
recognition is only one aspect of robotics. The main sources of the new wave
of effort are external to AI—industrial robotics plus the concern of the
decline in American productivity and the trade position of the United States
vis-a-vis Japan and West Germany. The initial growth of industrial robotics
took place largely outside of AI as a strictly engineering endeavor. As a
result, the initial growth tended to minimize the intelligence involved, for
example, sensory-motor coordination. One component of the new associa¬
tion of robotics with AI is the coupling of significant amounts of vision with
manipulators, reflecting the continued advance of vision capabilities in AI
throughout the 1970s. (Touch and kinesthetic sensing is increasingly impor¬
tant, too, but this does not build so strongly on prior progress in AI.) Impor¬
tantly, along with industrially motivated aspects, there is also a revival of
basic research in manipulation and movement in space and over real ter¬
rains.
It might seem that this is just another purely technical progression. But
with it has returned, as night follows day, the question of the relation of AI
and robotics as disciplines, just as the question was raised in the issue of
problem-solving versus recognition during the late 1960s. Is robotics a cen¬
tral part of AI or only an applied domain? Do graduate students in AI have to
understand the underlying science of mechanics and generalized coordinate
systems that are inherent in understanding manipulation and motion? Or is
that irrelevant to intelligence? Cases can be made either way. [Nilsson,
1982.]

Procedural versus Declarative Representation #2: 1980-

In the late 1970s, a new programming system called PROLOG emerged,

based on resolution-theorem-proving and constituting, in effect, a continua¬
tion of the effort to show that declarative formulations can be effective.
[Kowalski, 1979.] The effort is based primarily in Europe, and it is a vigor¬
ous movement. The attack is not occurring at the level of planning lan¬
guages, but at the level of LISP itself. Over the years, LISP has established
itself as the lingua franca of the AI community. Even though various other
programming systems exist, for example, rule-based systems of various fla¬
vors, practically everyone builds systems within a LISP programming envi¬
ronment. The planning languages (PLANNER, CONNIVER, etc.), which
showed how to effect another level of system organization above LISP, have
not proved highly effective as a replacement, and they receive only modest
use. As already noted, their contribution has been primarily conceptual.
Thus, although the original attack on theorem-proving was in terms of the
planner languages, the modern counterattack is at the level of LISP. By
being centered in Europe, with very little attention paid currently to
222 ALLEN NEWELL

PROLOG in the major AI centers in the United States, the issue takes on
additional coordinated dimensions. The outcome is far from clear at this
juncture.

DISCUSSION

It should be clear by now why I entered the caveats about historical accu¬
racy at the beginning. Each of the issues raises serious problems of charac¬
terization and historical grounding. No attempt has been made to define an
intellectual issue, so that some modestly objective way could be found to
generate a complete set of issues, for example, by placing a grid over the
literature of the field. Several additional issues might weil have emerged, and
some of those presented here might not have made the grade. Thus, the
population of issues exhibited must be taken, not just with a pinch of salt,
but soaked in a barrel of brine. Similar concerns attend dating the issues and
my interpretation of them; nevertheless, some comments about the total
picture seem worthwhile.

What Is Missing?

I do know why some issues did not make it. Three examples will illustrate
some reasons. The first is the broad but fundamental issue of the ethical use
of technology and the dehumanization of people by reduction to mechanism.
This issue engages all of technology and science. It seems particularly acute
for AI, perhaps, because the nature of mind seems so close to the quick. But
the history of science reminds us easily enough that at various stages as¬
tronomy, biology, and physics have seemed special targets for concern.
There has been continued and explicit discussion of these issues in connec¬
tion with AI. [Taube, 1961; Weizenbaum, 1976; McCorduck, 1979.] I have
not included them in the list of intellectual issues because they do not, in
general, seem to affect the course of the science. Where some aspect does
seem to do so, as in the issue of helping humans or replacing them, it has
been included. However, the broader issue certainly provides a thematic
background against which all work goes on in the field, increasing its ambi¬
guity, and the broader issue undoubtedly enters into individual decisions
about whether to work in the field and what topics to select.
The second example involves Hubert Dreyfus, who has been a persistent
and vocal critic of AI. [Dreyfus, 1972.] He has certainly become an issue for
the field; however, this does not necessarily produce an intellectual issue.
Dreyfus’s central intellectual objection, as I understand him, is that the
analysis of the context of human action into discrete elements is doomed to
failure. This objection is grounded in phenomenological philosophy. Unfor¬
tunately, this appears to be a nonissue as far as AI is concerned. The an¬
swers, refutations, and analyses that have been forthcoming to Dreyfus’s
INTELLECTUAL ISSUES IN THE HISTORY OF Al 223

writings have simply not engaged this issue—which, indeed, would be a

novel issue if it were to come to the fore.
The third example involves the imagery controversy, which has been
exceedingly lively in cognitive psychology. [Kosslyn, Pinker, Smith, and
Shwartz, 1979.] The controversy is over the nature of the representations
used by humans in imagining scenes and reasoning about them. There is no
doubt about its relevance to AI—the alternatives are a classical dichotomy
between propositional (symbolic?) representations and analog ones. Thus,
at heart, it is a variant of the issue of analog-versus-digital representation,
which has received mention. But for reasons that are quite obscure to me,
the imagery issue has received hardly any interest in the AI community,
except where that community also participates in cognitive psychology. As
things stand at the moment, this would be an issue for cognitive science, but
it is not one for AI.
Though enumerating intellectual issues exposes a certain amount of the
history of a field, even if only from particular viewpoints, some important
parts can be missed. These seem to be endeavors that were noncontroversial
or where the controversies were merely of the standard sort—of what prog¬
ress had been made, what subfields should get resources, and so forth. Thus,
work on program synthesis and verification goes unnoticed. Also, the major
effort in the 1970s to construct speech-understanding systems is barely
noticed. Perhaps this is not a valid point about the basic historical scheme
but reflects only the unevenness of my process of generating issues. Cer¬
tainly, there were issues in speech-recognition research both in the 1960s,
when Bell Laboratories decided to abandon speech recognition as an inap¬
propriate task, and in the 1970s, when a substantial effort sponsored by
DARPA to construct speech-understanding systems was dominated by AI
considerations over speech-science considerations. Perhaps intellectual
issues are generated from all scientific efforts in proportion to the number of
scientists involved in them (or to their square?); all we need to do is look for
them.

Characteristics of the History

Turning to what is revealed in Table 1, the most striking feature, to me at

least, is how many issues there are. Looked at in any fashion—number
active at one time (fifteen on average) or total number of issues during AI’s
quarter-century lifespan (about thirty)—it seems to me like a lot of issues.
Unfortunately, similar profiles do not exist for other fields (or I do not know
of them). Perhaps the situation in AI is typical, either of all fields at all times
or of all fields when they are getting started. In fact, I suspect it is due to the
interdisciplinary soup out of which AI emerged. [See my paper “Reflections
on the Structure of an Interdiscipline” in this volume.] Many other related
fields were being defined during the same post-World-War-II era—
cybernetics, operations research, management science, information theory,
224 ALLEN NEWELL

control theory, pattern recognition, computer science, and general systems

theory. Even so, I do not see any easy way of pinning down a correct
interpretation of why there are so many issues.
Issues are not independent; they come in clusters, which are coordinated.
Researchers tend to fall into two classes, corresponding to one pole or
another on all issues in the cluster. Clusters that occur in this history are as
follows (where polarities of subissues have been reoriented, if necessary, to
make them all line up together, corresponding to the superordinate issue):

AI versus Cybernetics

Symbolic versus continuous systems

Problem-solving versus recognition
Psychology versus neuroscience
Performance versus learning
Serial versus parallel

AI versus Computer Science

Symbols versus numbers

Heuristics versus algorithms
Interpretation versus compilation
Replacing versus helping humans
Problem-solving versus theorem-proving

Problem-Solving versus Knowledge Search

Heuristics versus epistemology

Search versus knowledge
Power versus generality
Processing versus memory

Linguistics versus AI and Cognitive Psychology

Competence versus performance

Syntax versus semantics

Engineering versus Science

Engineering analysis versus simulation

Engineering versus science
Real versus toy tasks

Wholes versus Atoms

Procedural versus declarative representation

Frames versus atoms
INTELLECTUAL ISSUES IN THE HISTORY OF Al 225

A cluster might seem to define a single underlying issue, which can then
replace component issues. However, the fact that issues are coordinated
does not make them identical. Some scientists can always be found who are
aligned in nonstandard patterns. In fact, some of the clusters seem much
more consistent than others. Thus, the multiplicity of issues keeps the
scientific scene complex, even though, because of clustering, it appears that
it should be clear and simple. In fact, many of the groupings are more easily
labeled by how they separate fields than by any coherent underlying concep¬
tual issue.
Clustering of issues does seem to be a common occurrence; for instance,
a standard advanced text on learning in psychology begins with a list of
seven dichotomous issues that characterize learning theories. [Hilgard and
Bower, 1948 and 1975, pp. 8-13.] The first three—peripheral versus central,
habits versus cognitive structures, and trial-and-error versus insight—form
a coordinated cluster that characterizes stimulus/response theories versus
cognitive theories (to which could even be added tough-minded versus ten¬
der-minded, the contrast William James used to distinguish the two main
types of psychologists). One possible source for such coordinated clusters is
the attempt to find multiple reasons to distinguish one approach from an¬
other. The approach comes first and the issues follow afterward. Then the
issues take on an autonomous intellectual life and what starts as rationaliza¬
tion ends up as analysis.
A major role of the issues here seems to be to carve up the total scientific
field into disciplines. AI, computer science, logic, cybernetics, pattern rec¬
ognition, linguistics, and cognitive psychology—all these seem to be dis¬
criminated in part by their position on these various issues. The issues, of
course, only serve as intermediaries for intellectual positions that derive
from many circumstances of history, methodological possibilities, and
specific scientific and technical ideas. Still, they seem to summarize a good
deal of what keeps the different fields apart, even though the fields have a
common scientific domain.
Is the large burst of issues that occurred at the birth of AI just an artifact
of my intent to gather issues for AI? If the period just before AI began, say
from 1940-1955, were examined carefully, would many more issues be
added? The relevant question should probably be taken with respect to some
other field as a base. Would a burst like this be found for cybernetics, which
started in 1940-1945? My own suspicion is yes, but I have not tried to verify
it.
Perhaps then the situation of AI could turn out to be typical. We would
find a plethora of issues in any science if we would but look and count; the
list from Hilgard and Bower might serve as a positive indicator. However,
before rushing to embrace this view, some counterevidence should be exam¬
ined. An interesting phenomenon in this same postwar period was the emer¬
gence of several one-theorem fields. Game theory, information theory,
226 ALLEN NEWELL

linear programming, and (later) dynamic programming—all had a single

strong result around which the field grew.15 Certainly, each also provided a
novel formulation, which amounted to a class of systems to be used to
theorize about some field. But initially there was only one striking theorem
to justify the entire field. It gave these fields a curious flavor. My personal
recollection is that all these fields, while exciting, profound, and (sometimes)
controversial, had none of the complexity of issues that we find in Table 1.

Intellectual Issues and Progress

There is a natural temptation to use the history of intellectual issues to

measure progress, once it has been explicitly laid out. It is true that some
issues have vanished from the scene, such as symbols versus numbers; that
seems, perhaps, like progress. It is also true that other issues seem to recur,
such as problem-solving versus recognition; that seems, perhaps, like lack of
progress. Neither interpretation is correct, I think. Rather, the progress of
science is to be measured by the accumulation of theories, data, and tech¬
niques, along with the ability they provide to predict, explain, and control.
This story is not to be told in terms of such intellectual issues as populate
this paper. It requires attention to the detailed content, assertions, and prac¬
tice of the science itself. True, at the more aggregate level of the paradigms
of Kuhn or the programmes of Lakatos, whole bodies of theory and data can
become irrelevant with a shift in paradigm or programme. But on the scale of
the twenty-five years of AI research (1955-1980), the story is one of accumu¬
lation and assimilation, not one of shift and abandonment. It is not even one
of settling scientific questions for good.
What then is the role of intellectual issues in the progression of science?
To echo my earlier disclaimer, I can only conjecture. Intellectual issues
seem to me more like generalized motivators. They evoke strong enough
passions to provide the springs to action, but they are vague enough so that
they do not get in the way of specific work. They can be used to convey a
feeling of coherence among investigations in their early stages, before it is
known exactly what the investigations will yield.
Evidence for this is that issues do not really go away; they return and
return again. Repetition is abundant in Table 1. The model that suggests
itself immediately is the spiral—each return constitutes a refined version of
the issue. Though the issues are certainly not identical each time, it seems
difficult to construe the changes as any sort of progressive refinement; some
seem more like wandering (e.g., the serial/parallel issue). A more plausible
explanation (to me) is that intellectual issues reflect perennial unanswerable

Another field, general systems theory, also had a single idea around which to build—that
there are common laws across all levels of systems from the atomic through cellular through
societal through astronomical. But there was no central result available, only the system view,
and this field has been markedly less successful than others in its growth and health.
INTELLECTUAL ISSUES IN THE HISTORY OF Al 227

questions about the structure of nature—continuity/discontinuity, stasis/

change, essence/accident, autonomy/dependence, and so forth. Whenever
in the course of science one of these can be recognized in the ongoing stream
of work, an appropriate intellectual issue will be instantiated, to operate as a
high-level organizing principle for a while. To be sure, this picture does not
capture all that seems to be represented in our population of intellectual
issues. But it seems substantially better than viewing science as progres¬
sively resolving such issues.

CONCLUSION

Putting to one side questions about the accuracy of the particular set of
issues displayed in Table 1, of what use is a history of a scientific field in
terms of intellectual issues? To repeat once more: It cannot substitue for a
substantive history in terms of concepts, theories, and data; however, it
does seem to capture some of the flavor of the field in an era. It is clearly a
component of the paradigm of a field or of research programmes within a
field. And, let us confess it, intellectual issues have a certain spiciness about
them that makes them fun to talk and write about. Perhaps it is the sense of
touching fundamental issues. But perhaps it also echoes Bertrand Russell s
famous aphorism that dealing with intellectual issues has all the advantages
of theft over honest toil.
,
 METHODOLOGICAL LINKS
BETWEEN ARTIFICIAL INTELLIGENCE
AND OTHER DISCIPLINES

Margaret A. Boden

Whether our interest is in psychology, philosophy, or linguistics, there is no

question but that artificial intelligence (AI) has given us a new standard of
rigor, and a new appreciation of the importance of mental process. Linguis¬
tics already had rigor but not process; psychology had little of either; and
philosophy had less of each. AI provides a range of precisely definable
computational concepts, specifying various symbolic representations and
transformations, with which to conceptualize the mind. And the technology
of programming makes manageable a degree of theoretical complexity that
would overwhelm the unassisted human brain. So the inadequacy of theoret¬
ical approaches that fail to recognize the complexity of mental structure and
process is now evident, and psychology and the philosophy of mind have
been influenced accordingly.
One example of a class of empirical psychological work partly inspired
by AI ideas is microdevelopmental research [e.g., Karmiloff-Smith and In-
helder, 1975; Karmiloff-Smith, 1979], which studies the dialectical interplay
between the child’s action sequences and changing cognitive representations
(theories, models, heuristics, and choice-criteria). The specifics of action are
emphasized on the assumption that procedural details of performance (not
only its overall structure) give clues to the underlying competence. How¬
ever, the degree of procedural detail—though high relative to more tradi¬
tional forms of experimentation in psychology—is inadequate for expressing
a complete computational model of the psychological processes concerned.
It is not a straightforward matter to assess such studies in computational
terms, and we need to learn how to refine the theories and methodology of
these studies so as to facilitate such assessment.
This case exemplifies the general point that, if we ask whether AI has
given us new discoveries as well as a new approach, the reply might be that
it has not been so helpful to working psychologists as its supporters initially
hoped. There has been an increasing amount of computationally influenced
229
230 MARGARET A. BODEN

empirical research in cognitive, developmental, and educational psychol¬

ogy. But (with the arguable exception of vision) we have gained little new
insight into the actual details of mental life, as opposed to the sorts of
questions that it may be appropriate to ask.
Is this because psychologists have not yet learnt how to apply AI fruit¬
fully to further their research or because (as some critics claim) it is in
principle unsuitable for psychological modeling? This question raises a num¬
ber of methodological difficulties and conceptual unclarities in applying AI
ideas to other disciplines. Some of these involve commitments to basic
theoretical or philosophical issues and call for cooperative research by peo¬
ple in various specialties.
There is much disagreement—and not a little skepticism in some quar¬
ters—about the extent to which empirical psychological work should or can
be planned and assessed in the light of computational ideas. It is not even
agreed whether or not psychologists sympathetic to the computational ap¬
proach should seek to express their theories in programmable (or pro¬
grammed) terms, as opposed to merely bearing computational issues in mind
in their work. Some AI workers even believe that doing psychological exper¬
iments is not an intellectually justifiable exercise in our present state of
ignorance, arguing that we should concentrate on clarifying the range of
possible computational mechanisms before trying to discover which ones
are actually used by living creatures.
Correlatively, there is disagreement over the psychological relevance of
specific examples of work within AI. Some of this disagreement is grounded
not in detailed objections, but in broad philosophical differences over the
potential psychological relevance of facts about neurophysiology or hard¬
ware.
For instance, there are two streams of work within AI vision research,
each of which has spurred psychological experimentation. The theoretical
emphases of these two streams are different, and, to some degree, opposed.
One is focused on low-level computational mechanisms, while the other is
focused on higher level, top-down processes in scene analysis. The former
(especially in the work of David Marr and his group) takes account of psy¬
chological optics and neurophysiology in some detail. [Marr, 1982.] But the
latter considers optics only in very general terms and ignores neurophysiol¬
ogy on the principled ground that physiological (hardware) implementation
is theoretically independent of questions about computational mechanisms.
This last is a widely shared view in AI (in some quarters approaching the
status of a dogma), and one that has caused many physiologically minded
psychologists to doubt the usefulness of AI work. It is a position that is
correct in principle but possibly sometimes misleading in practice. In an
abstract theoretical sense, all computing devices are equivalent, just as all
programming languages are. But to ignore the varying computational powers
of distinct (electronic or physiological) hardware may be as stultifying in
practice as to try to use a single programming language for all programs.
LINKS BETWEEN Al AND OTHER DISCIPLINES 231

Differences between programming languages often matter: A computation

that can be expressed easily if we use the representational potential of one
language may be difficult, or even practically infeasible, if we rely on an¬
other. Clearly, further computationally informed work on neurophysiolog¬
ical mechanisms is needed. It may be that physiology is relevant to the
relatively peripheral processing but irrelevant at higher levels, but the pre¬
cise points at which we may expect physiology to have a casting vote are
controversial. (Some of Marr’s earlier work on the cerebellum, for instance,
is now attracting interest within AI.)
If we could prove that a particular computation simply could not be
carried out in real time by any existing cerebral mechanism, then the use of
alien computer hardware to effect it would be psychologically irrelevant.
However, our ignorance of both computational and neurophysiological con¬
straints preempts such proofs. Nor can we prove that only mechanisms like
those in our brains are capable of certain computations. The most that can
be claimed as physiological support of a programmed model is that it is
consonant at some significant level with neurophysiology.
This claim is made, for example, in support of a very recent advance in
the computational modeling of vision. [Hinton, 1981.] G. E. Hinton’s work
is focused on low-level, dedicated hardware, mechanisms that are capable of
cooperative computation or parallel processing. Although it is not a simula¬
tion of detailed neurophysiology, Hinton believes it to be a prime strength of
his model that it is compatible with what is known about nervous function.
For instance, his model relies on excitatory and inhibitory connections be¬
tween computational units on various levels that appear to have an analogue
in the nervous connectivity of our own visual system.
Critics of AI often complain that one program does not make a theory,
any more than one swallow makes a summer. That is, AI is accused of being
empirical in the sense in which much of medicine is—it achieves practical
results by methods it does not understand and which it, therefore, cannot
responsibly generalize. This is, indeed, a methodological shortcoming of
much AI work—but not of all. Thus, Hinton’s research is especially inter¬
esting because it provides not only an example of a program that achieves a
desired result (the perception of shape), but also a general proof that results
of this class can be computed by computational systems of this form that are
within specific size constraints. In brief, he has proved that many fewer
computational units are necessary for the parallel computation of shape than
we might initially have supposed. This proof lends some more physiological
weight to the model, since the human retina apparently has enough cells to
do the job.
Because Hinton’s model of vision uses a type of computation fundamen¬
tally different from that of traditional AI, it raises the question of just which
psychological phenomena AI can be used to illuminate and which it cannot.
Hinton’s results suggest that parallel-processing systems can perform shape
discriminations—such as recognition of an overall Gestalt—commonly be-
 MARGARET A. BODEN
232

lieved (even within AI) to require relatively high-level interpretative pro¬

cesses Hinton’s results suggest also that the way in which an object is
represented may be radically different depending on whether it is perceived
as an object in its own right or as part of some larger whole. This might
account for the phenomenological differences between perceptual experi¬
ences of which we are reminded by those philosophers (e.g., Dreyfus [1972])
who argue that AI is essentially unfitted to model human minds. In general,
commonly expressed philosophical criticisms of AI and cognitive psychol¬
ogy that assume serial processing may be invalidated by these recent devel¬
opments. , . . . . .
This would be doubly true if computational techniques ot this work on
vision can be generalized to other domains. Hinton believes, for example,
that his computational model of spatial relations enables motor control to be
understood in a new way, one that is significantly analogous to mechanisms
of muscular control in the human body. Phenomenologically influenced phi¬
losophers, as well as scientists concerned with the psychophysiology of
movement, commonly complain that AI does not—or even cannot model
the body. Many philosophers and psychologists argue that human intelli¬
gence is rooted in our embodiment as material beings situated in a material
world and see AI as, therefore, radically irrelevant. Most current computers
do not have bodies that can move in and manipulate the external world, and
even robots are currently very crude in their motor abilities. But Hinton s
preliminary work on motor control suggests an efficient way of computing a
jointed limb’s movements and pathway through space (a problem that can
be solved by traditional computing techniques only in a highly inefficient
manner).
Even where psychologists deliberately match experimental results against
theories expressed in programmed form (e.g., the work of Newell and Simon
on problem-solving), the psychological relevance of the computational
model is debatable. [Newell and Simon, 1972.] It is not always clear just
which aspects of a program we might plausibly expect to be open to empir¬
ical test. Some aspects are not intended to have any psychological reality but
are included merely to produce a program that will run. However, we cannot
be sure that none of these last have any psychological significance, since it is
a prime claim of AI that it can highlight procedural lacunae in our theories
and offer us new concepts with which to jump the gap. Nor is the methodol¬
ogy of protocol-matching unproblematic: What are we to conclude from the
fact that no behavioral protocol is observed to match a specific process
posited by the programmed theory or that some matching protocol is ob¬
served? These problems (which have analogous forms to trouble all experi¬
mental psychologists) have been discussed by both proponents and oppo¬
nents of AI, but there is no consensus about the extent to which they cast
doubt on a computational approach to empirical psychology.
Of the many people who would concede that certain aspects (at leas) )f
vision, language-use, and problem-solving might yield to an AI approac
LINKS BETWEEN Al AND OTHER DISCIPLINES 233

some may feel that social psychology, for instance, has nothing to gain from
computational insights [e.g., Gauld and Shotter, 1977]; this should not be too
hastily assumed, however. Work within AI on the structure of action and the
attribution of intentions is relevant to theoretical discussions in social psy¬
chology. In general, AI supports the view that there may be generative rules
underlying social interaction or that social perception is a structured inter¬
pretative activity. But although these ideas are essentially consonant with a
computational viewpoint, specifying them in a particular case is a notori¬
ously difficult matter.
A general account of what sorts of psychological phenomena are or are
not grist for the AI mill would, of course, be very useful. But firm intellectual
ground could be provided for such an account only by a systematic theory of
representation. Philosophical discussions of the nature of intentionality are
clearly relevant [e.g., Fodor, 1981; Dennett, 1978/?]. Some philosophers
[e.g., Searle, 1980] argue that AI cannot model genuine (biological) inten¬
tionality, although discussions in recent issues of the peer-commentary jour¬
nal Behavioral and Brain Sciences show this claim to be highly controver¬
sial. But even John Searle admits that it can provide a scientifically useful
metaphor for intentionality. This is why AI is potentially relevant to studies
that are normally thought of as being humanistically oriented, such as social
and clinical psychology. [Boden, 1972.] Given that representational pro¬
cesses in computer models can function as heuristically fruitful analogues of
representational processes in our minds, the problem remains of providing
an account of the range and efficacy of such processes.
Artificial intelligence has shown that distinct representational forms affect
and effect inference in significantly different ways. Hinton’s work previ¬
ously mentioned is one of many examples that addresses such issues. An¬
other is Saul Amarel’s [1968] comparison of solutions to the “missionaries
and cannibals” problem, grounded in six representations of increasing
power; and a third is Aaron Sloman’s [1978b] discussion of analog represen¬
tations, which are interpreted by exploiting the similarity between their own
structure and that of the thing represented. However, there is—as yet—
little systematic understanding of the power and limitations of different rep¬
resentations. Work in computational logic is pertinent if it can show whether
or not a certain type of representation or computational process is, in princi¬
ple, capable of modeling a specific type of knowledge or simulating a given
class of psychological process.
General results in the philosophy of science apply to Al-based psychology
no less than to noncomputational theories. Some such results provide for a
rebuttal of common criticisms of the computational viewpoint. For instance,
even were it to turn out that AI is not appropriate for modeling many psycho¬
logical phenomena, we should not forget the Popperian point that we would
still have learnt something by the enterprise. Science involves conjecture
and refutation, and it is an advance to know that a specific conjecture has
been empirically rejected. Nor should we forget that some tricky
234 MARGARET A. BODEN

methodological problems apply not only to Al-based psychology but to other

theories too. Thus, critics of AI often remark—rightly—that we cannot
conclude from the fact that a computer program achieves a result in a certain
way that the mind achieves it in the same way. This is a special case of the
general truth that if our theory fits the facts, it may not be the only one to do
so. Because of this, conclusive verification of any scientific theory is in
principle impossible.
Work in AI concerns the nature and functioning of knowledge, and we
may hope for an increasing degree of cooperation between AI researchers
and philosophical epistemologists. Traditional approaches to reasoning
(whether deductive, inductive, or probabilistic) are overidealized. They ig¬
nore epistemologically important features of intelligent inference, features
that apply to all finite minds and cannot be dismissed as mere psychologism
irrelevant to normative epistemology. AI offers richer and more rigorous
descriptions of various data and procedures that comprise knowledge and of
the computational constraints that necessitate this rich variety.
Current AI research into the logic of nonmonotonic reasoning and truth-
maintenance, for example, asks how a belief system can be organized to
cope with the fact that a proposition may intelligently be proved to be true
yet turn out later to be false. Traditional logicians may wince at this descrip¬
tion, but finite minds have to construct their knowledge under this epistemic
constraint. Closely related work on frames considers the ways in which
single exemplars or stereotypes can be used in a flexible fashion for intelli¬
gent (though fallible) reasoning. Current discussion of naive physics exam¬
ines the everyday (pretheoretical) understanding of concepts such as cause,
shape, thing, pathway, inside, fluid, and so forth, and should help clarify
traditional problems concerning concepts like these. [P. J. Hayes, 1979.]
As these examples suggest, AI calls for a closer relation between epis¬
temology and empirical science than is usually thought proper by philoso¬
phers. Work on nonmonotonic reasoning can correctly be described as a
logical enquiry and, in principle, could have arisen in a noncomputational
context. In practice, however, it is AI that has enabled us to recognize the
complexity of problems involved in formalizing everyday inference and has
extended traditional formal approaches by offering new (computational)
concepts suited to expressing epistemic matters. Developmental psychology
(both Piagetian and non-Piagetian) has much to say on what might be called
naive physics—as also do studies of the perceptuomotor basis of language
(such as the psycholexicology of George Miller and Philip Johnson-Laird
[1976]). Biological and physiological considerations are relevant in view of
the sensorimotor ground of our knowledge, and there is a growing recogni¬
tion of the extent to which the newborn baby is already equipped with
computational structures and procedures fitted to the interpretation of its
life-world. Some recent work in the philosophy of mind [Churchland, 1979]
similarly argues that epistemology cannot ignore our material and biological
LINKS BETWEEN AI AND OTHER DISCIPLINES 235

embodiment—but it suffers from a failure to consider the computational

point of view.
Thus, we need an interdisciplinary epistemology in which computational
insights are integrated with philosophical understanding and psychological
and biological knowledge. Indeed, the need for a genuine interdisciplinarity
is a prime lesson of the computational approach. Workers in AI have much
to learn from the insights of psychologists, linguists, physiologists, biolo¬
gists, and philosophers, who, in turn, can benefit from their computationally
informed colleagues. [Boden, 19816.] Mere intellectual communication
across the boundaries of these several disciplines is not enough. We also
need mutually cooperative research by people who (albeit specializing in one
area) have a familiarity with other fields and a commitment to their intellec¬
tual integration. This vision of cognitive science will require modifying cur¬
rent educational practices, so that students are no longer socially sepa¬
rated—and even intellectually opposed—by traditional academic labels.
Reference to education reminds us of the pragmatic, as opposed to the
methodological, implications of AI. I have in mind here not primarily the
many commercial and administrative applications of AI, though these will
radically affect our social relations and institutions. Rather, I mean the way
in which the spread of computer analogies of the mind may influence the way
people think about themselves and society. As I have argued elsewhere, AI
is not only not dehumanizing but—potentially—is positively rehumanizing.
[Boden, 1977, chap. 15.] There are at least two senses in which this is so.
First, the view of intelligence springing from AI is active and constructive
rather than passive and defeatist like that which all too commonly informs
current educational (and mental-testing) practices. For example, the AI-
grounded educational approach developed by Seymour Papert deliberately
fosters constructive self-criticism, so that children concentrate on the
specifics of how to get better at doing something rather than giving up in
despair at their lack of talent. [Papert, 1980.] Again, Al-based CAI (com¬
puter assisted instruction) focuses on the pupil’s active construction and
exploration of the relevant domain of knowledge. [Sleeman and Brown,
1982.] In this, it differs significantly from the mechanistic approach of tradi¬
tional teaching machines.
Second, because AI deals with representational systems, it has a concep¬
tual base that can admit discussion of human subjectivity. This is why, as I
remarked earlier, social and clinical psychology can make use of the compu¬
tational approach. In general, this approach is consonant with humanistic or
hermeneutic (interpretative) theories of psychology rather than with those
psychological theories, such as behaviorism, grounded in the objective natu¬
ral sciences. Correlatively, hermeneutic or intentionalist philosophies of
mind are closer in spirit to AI than most of their proponents believe.
This remains true even if we accept the claim of some philosophers (e.g.,
Searle) already mentioned that processes in computer programs are not
236 MARGARET A. BODEN

really representations and do not really possess intentionality but that these
terms as used by the computer scientist are parasitic on their use in the
human psychological context. The point is that the representational
metaphor (for such it is, in this view) is one that is suited to express psycho¬
logical phenomena (which alone are truly representational or intentional)
precisely because it is drawn from those parts of our everyday conceptual
scheme that concern these matters. For concepts to be fruitful in the theory
and methodology of an empirical psychology, it is not required that they be
interpreted as literal descriptions of the phenomena, just as we need not see
the atom as literally a solar system in order to benefit from the notion of
planetary electrons. So, whether computer programs specify representa¬
tional processes or merely “representational” ones, they are conceptually
close to hermeneutic forms of psychology rather than to those forms that
ignore subjectivity.
Educational projects within society at large are needed to alert people to
these facts, for most people associate computers with relatively stupid
brute-force programs (such as those used to calculate gas bills) and think of
them as machines and, therefore, as mechanistic. Most people fail to realize
that computational machines are radically different from noncomputational
machines and that they are not mechanistic in the sense that implies a denial
of subjectivity. The mistaken, though widespread, assumption that A1 mod¬
els of human beings are mechanistic in this sense may make people experi¬
ence a threat to—or even an undermining of—their personal autonomy and
moral responsibility. Behaviorism in psychology and the philosophy of mind
has been often, and justly, criticized for its underestimation or denial of
these psychological characteristics. But the computational approach, if
properly understood, is not open to such criticisms. To realize this is to
disarm the computational bogeyman.
 ARTIFICIAL INTELLIGENCE
Cognition as Computation

Avron Barr

The ability and compulsion to know are as characteristic of human nature as

our physical posture and our languages. Knowledge and intelligence, as
scientific concepts, are used to describe how an organism’s experience
seems to mediate its behavior. This paper discusses the relation between
artificial intelligence (AI) research in computer science and the approaches
of other disciplines that study the nature of intelligence, cognition, and
mind. The state of AI after twenty-five years of work in the field is reviewed,
as are the views of its practitioners about its relation to cognate disciplines.
The paper concludes with a discussion of some possible effects on our
scientific work of emerging commercial applications of AI technology, ma¬
chines that can know and can take part in human cognitive activities.

ARTIFICIAL INTELLIGENCE

Artificial intelligence is the part of computer science concerned with creating

and studying computer programs that exhibit behavioral characteristics we
identify as intelligent in human behavior—knowing, reasoning, learning,
problem-solving, understanding language, and so on. Since the field came
into being in the mid-1950s, AI researchers have designed dozens of pro¬
grams and programming techniques that support some sort of intelligent
behavior. Although there are many attitudes expressed by researchers in the
field, most are motivated in their work on intelligent computer programs by
the thought that this work may lead to a new understanding of mind:

AI has also embraced the larger scientific goal of constructing an information¬

processing theory of intelligence. If such a science of intelligence could be
developed, it could guide the design of intelligent machines as well as explicate
intelligent behavior as it occurs in humans and other animals. (Nilsson, 1980,
P- 2.)

Whether or not it leads to a better understanding of the mind, there is

every evidence that current work in AI will lead to a new intelligent
237
238 AVRON BARR

technology that may have dramatic effects on our society. Already, experi¬
mental AI systems have generated interest and enthusiasm in industry and are
being developed commercially. These experimental systems include programs
that: solve some hard problems in chemistry, biology, geology, engineering,
and medicine at human-expert levels of performance; manipulate robotic
devices to perform some useful sensory-motor tasks; and answer questions
posed in restricted dialects of English (or French, Japanese, etc.). Useful AI
programs will play an important part in the evolution of the role of com¬
puters in our lives—a role that has changed in our lifetime from remote to
commonplace and if current expectations about computing cost and power
are correct, is likely to evolve further from useful to essential.

The Origins of Artificial Intelligence

Scientific fields emerge as the concerns of scientists congeal around various

phenomena. Sciences are not defined, they are recognized. (Newell, 1973c,
p. 1.)

The intellectual currents of the times help direct scientists to the study of
certain phenomena. For the evolution of AI, the two most important forces
in the intellectual environment of the 1930s and 1940s were mathematical
logic, which had been under rapid development since the end of the
nineteenth century, and new ideas about computation. The logical systems
of Frege, Whitehead and Russell, Tarski, and others showed that some
aspects of reasoning could be formalized in a relatively simple framework.

The fundamental contribution was to demonstrate by example that the manipu¬

lation of symbols (at least some manipulation of some symbols) could be de¬
scribed in terms of specific, concrete processes quite as readily as could the
manipulation of pine boards in a carpenter shop ....
Formal logic, if it showed nothing else, showed that ideas—at least some
ideas—could be represented by symbols, and that these symbols could be
altered in meaningful ways by precisely defined processes. (Newell and Simon,
1972, p. 877.)

Mathematical logic continues to be an active area of investigation in AI, in

part because general-purpose, logico-deductive systems have been success¬
fully implemented on computers. But even before the advent of computers,
the mathematical formalization of logical reasoning shaped people’s concep¬
tion of the relation between computation and intelligence.
Ideas about the nature of computation, due to Church, Turing, and
others, provided the link between the notion of formalization of reasoning
and the computing machines about to be invented. What was essential in this
work was the abstract conception of computation as symbol-processing. The
first computers were numerical calculators that did not appear to embody
Al: COGNITION AS COMPUTATION 239

much intelligence at all. But before these machines were even designed,
Church and Turing had seen that numbers were an inessential aspect of
computation—just one way of interpreting the internal states of the ma¬
chine.

In their striving to handle symbols rigorously and objectively—as objects—

logicians became more and more explicit in describing the processing system
that was supposed to manipulate the symbols. In 1936, Alan Turing, an English
logician, described the processor, now known as the Turing machine, that is
regarded as the culmination of this drive toward formalization. (Newell and
Simon, 1972, p. 878.)

The model of a Turing machine contains within it the notions both of what can
be computed and of universal machines—computers that can do anything that
can be done by any machine. (Newell and Simon, 1976u, p. 117.)

Turing, who has been called the father of AI, not only invented a simple,
universal, and nonnumerical model of computation, but also argued directly
for the possibility that computational mechanisms could behave in a way
that would be perceived as intelligent.

Thought was still wholly intangible and ineffable until modern formal logic
interpreted it as the manipulation of formal tokens. And it seemed still to
inhabit mainly the heaven of Platonic ideals, or the equally obscure spaces of
the human mind, until computers taught us how symbols could be processed by
machines. A.M. Turing . . . made his great contributions at the mid-century
crossroads of these developments that led from modern logic to the computer.
(Newell and Simon, 1976u, p. 125.)

As Allen Newell and Herbert Simon point out in the “Historical Epi¬
logue” to their classic work Human Problem Solving, there were other
strong intellectual currents from several directions that converged in the
middle of this century in the people who founded the science of artificial
intelligence. [Newell and Simon, 1972.] The concepts of cybernetics and
self-organizing systems of Wiener, McCulloch, and others dealt with the
macroscopic behavior of locally simple systems. The cyberneticists in¬
fluenced many fields, because their thinking spanned many fields, linking
ideas about the workings of the nervous system with information theory and
control theory as well as with logic and computation. Their ideas were part
of the Zeitgeist, but, in many cases, the cyberneticists influenced early
workers in AI more directly—as their teachers.
What eventually connected these diverse ideas was, of course, the devel¬
opment of computing machines themselves, conceived by Babbage and
guided in this century by Turing, von Neumann, and others. It was not long
after the machines became available that people began to try to write pro¬
grams to solve puzzles, play chess, and translate texts from one language to
another—the first AI programs. What was it about computers that triggered
240 AVRON BARR

the development of AI? Many ideas about computing relevant to AI emerged

in early designs—ideas about memories and processors, systems and con¬
trol, and levels of languages and programs. But the single attribute of the
new machines that brought about the emergence of the new science was
their inherent potential for complexity, encouraging (in several fields) the
development of new and more direct ways of describing complex pro¬
cesses—in terms of complicated data structures and procedures with hun¬
dreds of different steps.

Problem solving behaviors, even in the relatively well-structured task environ¬

ments that we have used in our research, have generally been regarded as
highly complex forms of human behavior—so complex that for a whole genera¬
tion they were usually avoided in the psychological laboratory in favor of
behaviors that seemed to be simple ....
The appearance of the modern computer at the end of World War II gave us
and other researchers the courage to return to complex cognitive performances
as our source of data. ... a device capable of symbol-manipulating behavior at
levels of complexity and generality unprecedented for man-made mechanisms.

This was part of the general insight of cybernetics, delayed by ten years and
applied to discrete symbolic behavior rather than to continuous feedback sys¬
tems. (Newell and Simon, 1972, pp. 869-870.)

Computers, Complexity, and Intelligence

As Pamela McCorduck notes in her entertaining historical study of AI,

Machines Who Think, there has been a long-standing connection between the
idea of complex mechanical devices and intelligence. [McCorduck, 1979.]
Starting with the fabulously intricate clocks and mechanical automata of
past centuries, people have made an intuitive link between the complexity of
a machine’s operation and some aspects of their own mental life. Over the
last few centuries, new technologies have resulted in a dramatic increase in
the complexity we can achieve in the things we build. Modern computer
systems are more complex by several orders of magnitude than anything
man has built before.
The first work on computers in this century focused on the kinds of
numerical computations that had previously been performed collaboratively
by teams of hundreds of clerks, organized so that each did one small subcal¬
culation and passed his or her results on to the clerk at the next desk. Not
long after the dramatic success demonstrated by the first digital computers
with these elaborate calculations, people began to explore the possibility of
more generally intelligent mechanical behavior—could machines play chess,
prove theorems, or translate languages? They could, but not very well. The
computer performs its calculations following the step-by-step instructions it
is given—the method must be specified in complete detail. Most computer
scientists are concerned with designing new algorithms, new languages, and
Al: COGNITION AS COMPUTATION 241

new machines for performing tasks like solving equations and alphabetizing
lists—tasks that people perform using methods they can explicate. How¬
ever, people cannot specify in detail how they decide which move to make in
a game of chess or how they determine that two sentences mean the same
thing.
The realization that the detailed steps of almost all intelligent human
activity were unknown marked the beginning of artificial intelligence as a
separate part of computer science. AI researchers investigate different types
of computation, and different ways of describing computation, in an effort
not just to create intelligent artifacts, but also to understand what intelli¬
gence is. A basic tenet is that human intellectual capacity will best be de¬
scribed in the same terms as those invented to describe artificial intelligence
researchers’ programs. However, researchers are just beginning to learn
enough about those programs to know how to describe them scientifically—
in terms of concepts that illuminate the program’s nature and differentiate
among fundamental categories. These ideas about computation have been
developed in programs that perform many different tasks, sometimes at the
level of human performance, often at a much lower level. Most of these
methods are obviously not the same as those people use to perform the
tasks—some of them might be.

The Status of Artificial Intelligence

Many intelligent activities besides numerical calculation and information

retrieval have been accomplished by programs. Many key aspects of
thought—like recognizing people’s faces and reasoning by analogy—are still
puzzles; they are performed so unconsciously by people that adequate com¬
putational mechanisms have not been postulated. Some of the successes, as
well as some of the failures, have come as surprises. I will list here some of
the aspects of intelligence investigated in AI research and try to give an
indication of the state of progress.
There is an important philosophical point here that will be sidestepped:
Doing arithmetic or learning the capitals of all the countries of the world, for
example, are certainly activities that indicate intelligence in humans. The
issue here is whether a computer system that can perform these tasks can be
said to know or understand anything. This point has been discussed at length
[e.g., Searle, 1980 and appended commentaries] and will be avoided here by
describing the behaviors themselves as intelligent without commitment on
how to describe the machines that produce them.

Problem Solving

The first big successes in AI were programs that could solve puzzles and
play games like chess. Techniques, such as looking ahead several moves and
242 AVRON BARR

dividing difficult problems into easier subproblems, evolved into the funda¬
mental AI techniques of search and problem reduction. Today’s programs
play championship-level checkers and backgammon, as well as very good
chess. Another problem-solving program that performs symbolic evaluation
of mathematical functions has attained very high levels of performance and
is being widely used by scientists and engineers. Some programs can even
improve their own performance with experience.
As discussed in the following sections, the open questions in this area
involve capabilities that human players exhibit but cannot articulate, such as
the chess master’s ability to see the board configuration in terms of meaning¬
ful patterns. Another basic open question involves the original conceptuali¬
zation of a problem, called in AI the choice of problem representation.
Humans often solve a problem by finding a way of thinking about it that
makes the solution easy; AI programs, so far, must be told how to think
about the problems they solve (i.e., the space in which to search for the
solution).

Logical Reasoning

Closely related to problem- and puzzle-solving was early work on logical

deduction. Programs were developed that could prove assertions by manipu¬
lating a database of facts, each represented by discrete data-structures just
as they are represented by formulas in mathematical logic. These methods,
unlike many other AI techniques, could be shown to be complete and consis¬
tent. That is, given a set of facts, the programs theoretically could prove all
theorems that followed from the facts, and only those theorems. Logical
reasoning has been one of the subareas most persistently investigated in AI
research. Of particular interest are the problems of finding ways of focusing
on only the relevant facts in a large database and of keeping track of
justifications for beliefs and updating them when new information arrives.

Programming

Although perhaps not an obviously important aspect of human cognition,

programming itself is an important area of research in AI. Work in this field,
called automatic programming, has investigated systems that can write
computer programs from a variety of descriptions of their purpose: exam¬
ples of input/output pairs, high-level language descriptions, and even En¬
glish descriptions of algorithms. Progress has been limited to a few fully
worked-out examples. Automatic-programming research may result not only
in semiautomated systems for software development, but also in AI pro¬
grams that learn (i.e., modify their behavior) by modifying their own code.
Related work in the theory of programs is fundamental to all AI research.
Al: COGNITION AS COMPUTATION 243

Language

The domain of language-understanding was also investigated by early AI

researchers and has consistently attracted interest. Programs have been
written that retrieve information from a database in response to questions
posed in English, translate sentences from one language to another, follow
instructions or paraphrase statements given in English, and acquire knowl¬
edge by reading textual material and building an internal database. Some
programs have even achieved limited success in interpreting instructions
that are spoken into a microphone rather than typed into a computer. Al¬
though these language systems are not nearly so good as people are at any of
these tasks, they are adequate for some applications. Early successes with
programs that answered simple queries and followed simple directions, and
early failures at machine-translation attempts, have resulted in a sweeping
change in the whole AI approach to language. The principal themes of cur¬
rent language-understanding research are the importance of vast amounts of
knowledge about the subject being discussed and the role of expectations,
based on the subject matter and the conversational situation, in interpreting
sentences. The state of the art of practical language programs is represented
by useful “front ends” to a variety of software systems. These programs
accept input in some restricted form—they cannot handle some of the
nuances of English grammar and are useful for interpreting sentences only
within a relatively limited domain of discourse. Although there has been
very limited success in translating AI results in language- and speech¬
understanding programs into ideas about the nature of human language pro¬
cessing, the realization of the importance in language-understanding of ex¬
tensive background knowledge and the contextual setting and intentions of
the speakers has changed our notion of what language or a theory of lan¬
guage might be.

Learning

Certainly one of the most salient and important aspects of human intelli¬
gence is our ability to learn. However, this is an example of cognitive behav¬
ior that is so poorly understood that very little progress has been made in
achieving it in AI systems. Although there have been several interesting
attempts, including programs that learn from examples, from their own per¬
formance, or from advice from others, AI systems do not exhibit noticeable
learning ability.

Robotics and Vision

One area of AI research that is receiving increasing attention involves pro¬

grams that manipulate robot devices. Research in this field has looked at
244 AVRON BARR

everything from the optimal movement of robot arms to methods of planning

a sequence of actions to achieve a robot’s goals. Some robots “see” through
a television camera that transmits an array of information back to the com¬
puter. Processing visual information is another very active, and very
difficult, area of AI research. Programs have been developed that can recog¬
nize objects and shadows in visual scenes and even identify small changes
from one picture to the next, for example, for aerial reconnaissance. The
true potential of this research, however, is that it deals with artificial intelli¬
gences in perceived and manipulate environments similar to our own.

Systems and Languages

In addition to work directly aimed at achieving intelligence, the development

of new tools has always been an important aspect of AI research. Some of
the most important contributions of AI to the world of computing have been
in the form of spin-offs. Computer-systems ideas like timesharing, list pro¬
cessing, and interactive debugging were developed in the AI research envi¬
ronment. Specialized programming languages and systems, with features
designed to facilitate deduction, robot manipulation, cognitive modeling,
and so on, have often been rich sources of new ideas. Most recent among
these has been the plethora of knowledge-representation languages. These
are computer languages for encoding knowledge and reasoning-methods as
data structures and procedures, developed over the last five years to explore
a variety of ideas about how to build reasoning programs. Terry Winograd’s
article, “Beyond Programming Languages,” discusses some of his ideas
about the future of computing, inspired in part by his research on AI.
[Winograd, 1979.]

Expert Systems

Finally, the area of expert or knowledge-based systems has recently

emerged as a likely area for useful applications of AI techniques.
[Feigenbaum, 1977.] Typically, the user interacts with an expert system in a
“consultation dialogue,” just as he would interact with a human expert in a
particular area: explaining his problem, performing suggested tests, and ask¬
ing questions about proposed solutions. Current experimental systems have
achieved high levels of performance in consultation tasks like chemical and
geological data-analysis, computer-system configuration, completing in¬
come tax forms, and even medical diagnosis. Expert systems can be viewed
as intermediaries between human experts who interact with the systems in
knowledge acquisition mode and human users who interact with the systems
in consultation mode. Furthermore, much research in this area of AI has
focused on endowing these systems with the ability to explain their rea¬
soning, both to make the consultation more acceptable to the user and to
Al: COGNITION AS COMPUTATION 245

help the human expert locate the cause of errors in the system’s reasoning
when they occur.
Because I am most familiar with this area of AI research, and because its
imminent commercial applications are, I feel, indicative of important
changes in the field, much of the ensuing discussion of the role of AI in the
study of mind will refer to expert-systems research. The fact that these
systems represent vast amounts of knowledge obtained from human ex¬
perts; are used as tools to solve hard problems using this knowledge; can be
viewed as intermediaries between human problem-solvers; must explain
their thought processes in terms that people can understand; and are worth a
lot of money to people with real problems are the essential points that, I will
argue, will be true of all of AI someday—in fact, of computers in general—
and will change the role that AI research plays in the scientific study of
thought.

Open Problems

Although there has been much activity and progress in the twenty-five year
history of AI, some very central aspects of cognition have not yet been
achieved by computer programs. Our abilities to reason about others’ be¬
liefs; know the limits of our knowledge; visualize; be reminded of relevant
events; learn; reason by analogy; make plausible inferences; realize when
they are wrong, and know how to recover are not at all understood.
It is a fact that these and many other fundamental cognitive capabilities
may remain problematic for some time. But it is also a fact that computer
programs have successfully achieved a level of performance on a range of
intelligent behaviors unmatched by anything other than the human brain.
The failure of AI to achieve some seemingly simple cognitive capabilities in
computer programs becomes, according to the view of AI presented in this
paper, part of the set of phenomena to be explained by the new science.

ARTIFICIAL INTELLIGENCE AND THE STUDY OF MIND

Artificial intelligence research in problem-solving, language processing, and

so forth, has produced some impressive and useful computer systems; it has
also influenced, and been influenced by, research in many fields. What, then,
is the relation between AI and other disciplines that study the various as¬
pects of mind, for example, psychology, linguistics, philosophy, and
sociology?
AI certainly has a unique method—designing and testing computer pro¬
grams—and a unique goal—making those programs seem intelligent. It has
been argued from time to time that these attributes make AI independent of
other disciplines: “Artificial intelligence was an attempt to build intelligent
246 AVRON BARR

machines without any prejudice toward making the system simple, biolog¬
ical, or humanoid.” (Minsky, 1963, p. 409.) But we do not start from scratch
in building the first program to accomplish some intelligent behavior; the
ideas about how that program is to work must come from somewhere. Fur¬
thermore, most AI researchers are interested in understanding the human
mind and actively seek hints about its nature in their experiments with their
programs.
The interest within AI in results and open problems of other disciplines
has been fully reciprocated by interest in and application of AI research
activity among researchers in other fields. Many experimental and theoret¬
ical insights in psychology and linguistics, at least, have been sparked by AI
techniques and results. Furthermore, I argue later on, this flow is likely to
increase dramatically in the future; its source is the variety of new phenom¬
ena displayed by AI systems—the number, quality, utility, and level of
activity of which will soon greatly increase. But first let us examine what
kind of interactions have taken place between AI and other disciplines in the
last 25 years.

The Language of Computation

As I defined it at the outset, AI is a branch of computer science. Its prac¬

titioners are trained in the various subfields of computer science: formal
computing theory, algorithm design, hardware and operating-systems ar¬
chitecture, programming-languages, and programming. The study of each of
these subareas has produced a language of its own, indicating our under¬
standing of the important known phenomena of computing. The underlying
assumption of our research is that this language (which involves concepts
like process, procedure, interpreter, bottom-up and top-down processing,
object-oriented programming, trigger, etc.), and the experience with compu¬
tation that it embodies, will, in turn, assist us in understanding the phenom¬
ena of mind.
Before I go on to discuss the utility of these computational concepts, it
should be stated that, in fact, our understanding of computation is quite
limited. John von Neumann dreamed of an information theory of the nature
of thinking:

. . . that body of experience which has grown up around the planning, evaluat¬
ing, and coding of complicated logical and mathematical automata will be the
focus of much of this information theory. . . .

it would be very satisfactory if one could talk about a “theory” of such auto¬
mata. Regrettably, what at this moment exists—and to what I must appeal—
can as yet be described only as an imperfectly articulated and hardly for¬
malized “body of experience.” (von Neumann, 1958, p. 2.)
Al: COGNITION AS COMPUTATION 247

And ten years later, in their superb treatise on perceptronlike automata,

Minsky and Papert lament:

We know shamefully little about our computers and their computations ....
We know very little, for instance, about how much computation a job should
require ....

The immaturity shown by our inability to answer questions of this kind is

exhibited even in the language used to formulate the questions. Word pairs
such as “parallel” vs. “serial,” “local” vs. “global,” and “digital” vs.
“analog” are used as if they referred to well-defined technical concepts. Even
when this is true, the technical meaning varies from user to user and context to
context. But usually they are treated so loosely that the species of computing
machine defined by them belongs to mythology rather than science. (Minsky
and Papert, 1969, pp. 1-2.)

There is still no adequate theory of computation for understanding the na¬

ture and scope of symbolic processes, but there is rapidly accumulating
experience with computation of all sorts—useful new concepts continually
emerge.

The Computational Metaphor

The discipline most closely related to AI is cognitive psychology. Both deal

primarily with the same kinds of behavior—perception, memory, problem¬
solving—and they are siblings. Modern cognitive psychology emerged from
its behavior-oriented precursors in conjunction with the rise of AI. That
there might be a relation between the new field of AI and traditional interests
of psychologists was evident from the beginning:

Our fundamental concern . . . was to discover whether the cybernetic ideas

have any relevance for psychology. The men who have pioneered in this area
have been remarkably innocent about psychology .... There must be some
way to phrase the new ideas so that they can contribute to and profit from the
science of behavior that psychologists have created. (Miller, Galanter, and
Pribram, 1960, p. 3.)

What in fact happened was that the existence of computing served as an

inspiration to traditional psychologists to begin to theorize in terms of inter¬
nal, cognitive mechanisms. Using the concepts of computation as metaphors
for processes of the mind strongly influenced the form of modern theories of
cognitive psychology—for example, theories expressed in terms of
memories and retrieval processes.

Computers accept information, manipulate symbols, store items in “memory”

and retrieve them again, classify inputs, recognize patterns, and so on.
Whether they do these things just like people was less important than that they
248 AVRON BARR

do them at all. The coming of the computer provided a much-needed reassur¬

ance that cognitive processes were real. (Neisser, 1976, p. 5.)

The metaphorical use of the language of computation in describing mental

processes was found to be, at least for a time, quite a fertile ground for
sprouting psychological theories:

During a period of concept formation, we must be well aware of the metaphor¬

ical nature of our concepts. However, during a period in which the concepts
can accommodate most of our questions about a given subject matter, we can
afford to ignore their metaphorical origins and confuse our description of real¬
ity with that reality. (Arbib, 1972, p. 11.)

When pioneering work by Newell, Shaw, and Simon, and other research
groups showed that “programming up” their intuitions about how humans
solve puzzles, find theorems, and so forth, was adequate for getting impres¬
sive results, the link between the study of human problem-solving and AI
research was firmly established.
Consider, for example, computer programs that play chess. Current pro¬
grams are quite proficient—the best experimental systems play at the human
expert level, but not so well as human chess masters. The programs work by
searching through a space of possible moves, that is, considering alternative
moves and their consequences several steps ahead in the game, just as
human players do. These programs, even some of the earliest versions,
could search through thousands of moves in the time it takes human players
to consider only a dozen or so alternatives. The theory of optimal search,
developed as a mathematical formalism (paralleling, as a matter of fact,
much work on optimal decision theory in operations research), constitutes
some of the core ideas of AI.
The reason that computers cannot beat the best human players is that
looking ahead is not all there is to chess. Since there are too many possible
moves to search exhaustively, even on the fastest imaginable computers,
alternative moves (board positions) must be evaluated without knowing for
sure which move will lead to a winning game, and this is one of those
abilities that human chess experts cannot make explicit. Psychological stud¬
ies have shown that chess masters have learned to see thousands of mean¬
ingful configurations of pieces when they look at chess positions, which
presumably helps them decide on the best move, but no one has yet sug¬
gested how to design a computer program that can identify these configu¬
rations.
Due to the lack of theory or intuition about human perception and learn¬
ing, AI progress on computer chess was virtually stopped, but it is quite
possible that new insights into a very general problem were gained. The
computer programs had pointed out more clearly than ever what kinds of
things it would be useful for a cognitive system to learn to see. It takes many
Al: COGNITION AS COMPUTATION 249

years for a chess expert to develop expertise, ability to understand the game
in terms of these concepts and patterns that he or she cannot explain easily,
if at all. The general problem is of course to determine what it is about our
experience that we apply to future problem-solving: What kind of knowledge
do we glean from our experience? The work on chess indicated some of the
demands that would be placed on this knowledge.

Language Translation and Linguistics

Ideas about getting computers to deal in some useful way with the human
languages, called natural languages by computer scientists, were conceived
before any machines were ever built. The first line of attack was to try to use
large bilingual dictionaries stored in the computers to translate sentences
from one language to another. [Barr and Feigenbaum, 1981, vol. 1, pp. 233-
238.] The machine would just look up the translation(s) of each word in the
original sentence, figure out its meaning (perhaps expressed in some interlin¬
gua), translate the words, and produce a syntactically correct version in the
target language.
It did not work. It became apparent early on that processing language in
any useful way involved understanding, which, in turn, involved a great deal
of knowledge about the world—in fact, it could be argued that the more we
know, the more we understand each sentence we read. And the level of
world knowledge needed for any useful language-processing is much higher
than our original intuitions led us to expect.
There has been a serious debate about whether AI work in computational
linguistics has enlightened us at all about the nature of language. [See
Dresher and Hornstein, 1976; and replies by Winograd, 1977; Schank and
Wilensky, 1977.] The position taken by AI researchers is that if our goal in
linguistics is to include understanding sentences like Do you have the time?
and We’ll have dinner after the kids wash their hands, which involve the
total relationship between the speakers, then there is much more to it than
the syntactic arrangement of words with well-defined meanings; that al¬
though the study in linguistics of systematic regularities within and between
natural languages is an important key to the nature of language and the
workings of the mind, it is only a small part of the problem of building a
useful language processor and, therefore, only a small part of an adequate
understanding of language.

For both people and machines, each in their own way, there is a serious
problem in common of making sense out of what they hear, see, or are told
about the world. The conceptual apparatus necessary to perform even a partial
feat of understanding is formidable and fascinating. (Schank and Abelson,
1977, p. 2.)

Linguists have almost totally ignored the question of how human understand¬
ing works. ... It has nevertheless been consistently regarded as important that
250 AVRON BARR

computers deal well with natural language. . . . None of these high-sounding

things are possible, of course, unless the computer really “understands” the
input. And that is the theoretical significance of these practical questions—to
solve them requires no less than articulating the detailed nature of “under¬
standing.” If we understood how a human understands, then we might know
how to make a computer understand, and vice versa. (Ibid., p. 8.)

This idea that building AI systems requires the articulation of the detailed
nature of understanding, that is, that implementing a theory in a computer
program requires us to work out our fuzzy ideas and concepts, has been
suggested as a major contribution of AI research.

Whenever an AI researcher feels he understands the process he is theorizing

about in enough detail, he then begins to program it to find out where he was
incomplete or wrong. . . . The time between the completion of the theory and
the completion of the program that embodies the theory is usually extremely
long. (Ibid., p. 20.)

And Newell, in a thorough discussion of eight possible ways we might view

the relation of AI to psychology, suggests that building programs “forces
psychologists to become operational, that is, to avoid the fuzziness of using
mentalistic terms.” (Newell, 1970, p. 365.)
Certainly the original conception of the machine-translation effort, al¬
though it seemed intuitively sensible, fell far short of what would be required
to enable a machine to handle language, indicating a limited conception of
what language is. It is in the broadening of this conception that AI has
contributed most to the study of language. [Schank and Abelson, 1977, p. 9.]
Thus, AI can show, as illustrated in the examples of chess and language
understanding, that intuitive notions and assumptions about mental pro¬
cesses just do not work. Furthermore, analyzing the behavior of AI pro¬
grams implemented on the basis of existing, inadequate concepts can offer
hints on how the concepts of the theory affect its performance and on why
the theories do not work.

Scientific Languages and Theory Formation

Laurence Miller, in an article that reviews the dialogue between psycholo¬

gists and AI researchers about the contribution of AI to the understanding of
mind, concludes that

the critics of AI believe that it is easy to construct plausible psychological

theories; the difficult task is demonstrating that these theories are true. The
advocates of AI believe that it is difficult to construct adequate psychological
theories; but once such a theory has been constructed, it may be relatively
simple to demonstrate that it is true. (L. Miller, 1978, p. 113.)
Al: COGNITION AS COMPUTATION 251

And Schank and Abelson agree:

We are not oriented toward finding out which pieces of our theory are
quantifiable and testable in isolation. We feel that such questions can wait.
First we need to know if we have a viable theory .... (Schank and Abelson,
1977, p. 21.)

Just as AI must consider the same issues that are addressed in psychology
and linguistics, other aspects of knowledge, dealt with by other traditional
disciplines, must be considered. For example, current ideas in AI about
linking computing machines together into coherent systems or cooperative
problem-solvers forces us to consider the sociological aspects of knowing. A
fundamental problem in AI is communication among many individual units,
each of which knows some things relevant to some problems as well as
something about the other units. The form of communication between units,
the organizational structure of the complex, and the nature of the individ¬
uals’ knowledge of each other are all questions that must find some engineer¬
ing solution if the apparent power of distributed processing is to be realized.
These issues have been studied in other disciplines, albeit from very
different perspectives and with different goals and methods. We can view
the different control schemes proposed for interprocess communication, for
example, as attempts to design social systems of knowledgeable entities.
Our intuitions, once again, form the specifications for the first systems. Reid
G. Smith has proposed a contract net where individual entities negotiate
their role in attacking the problem via requests for assistance from other
processors, proposals for help in reply, and contracts indicating agreement
to delegate part of the problem to another processor; and W. Kornfeld and
Carl Hewitt have developed a model explicitly based on problem-solving in
the scientific community. [R. G. Smith, 1978; Kornfeld and Hewitt, 1981.]
Only after we have been able to build many systems based on such models
will we be able to identify key factors in the design of such systems.
There is another kind of study of the mind, conducted by scientists who
seek to understand the workings of the brain. The brain is a “mechanism’’
that has been associated with computing machines since their invention, and
that has puzzled computer scientists greatly:

We know the basic active organs of the nervous system (the nerve cells). There
is every reason to believe that a very large-capacity memory is associated with
this system. We do most emphatically not know what type of physical entities
are the basic components for the memory in question, (von Neumann, 1958,
p. 68.)

If research on AI generates a language for describing what a computational

system is doing, in terms of processes, memories, messages, and so forth,
then that language may very well be the one in which the function of the
252 AVRON BARR

neural mechanisms should be described. [Lenat, 1981; Torda, 1982.] And, as

Herbert Simon points out, this functionality may be shared by nature’s other
brand of computing device, DNA:

It might have been necessary a decade ago to argue for the commonality of the
information processes that are employed by such disparate systems as com¬
puters and human nervous systems. The evidence for that commonality is now
overwhelming, and the remaining questions about the boundaries of cognitive
science have more to do with whether there also exist nontrivial commonalities
with information processing in genetic systems than with whether men and
machines both think. (Simon, 1980, p. 45.)

One more example of the overlap of concerns between AI and related

disciplines: Enabling an individual to know something about what another
knows, without actually knowing it, involves defining the nature of what is
known elsewhere—who are the experts on what kinds of problems and what
might they know that could be useful. This is related directly to the kind of
categorization of knowledge that is the essence of library science. But in¬
stead of dealing with categories in which static books will be filed, AI must
consider the dynamic aspects of systems that know and learn.
The relation, then, between AI and disciplines like psychology, linguis¬
tics, sociology, brain science, and library science is a complex one. Cer¬
tainly, our current understanding of the phenomena dealt with by these
disciplines—cognition, perception, memory, language, social systems, and
categories of knowledge—has provided intuitions and models on which the
first AI programs were built. And, as has happened in psychology and lin¬
guistics, these first systems may, in turn, show us new aspects of the phe¬
nomena that we have not considered in studying their natural occurrence.
But most importantly, the development of AI systems, of useful computer
tools for knowledge-oriented tasks, will expose us to many new phenomena
and variations that will force us to increase our understanding.

THE PRACTICE OF ARTIFICIAL INTELLIGENCE

Artificial intelligence and computer science in general employ a unique

method among disciplines involved in advancing our understanding of cogni¬
tion—building computers and programs and observing and trying to explain
patterns in the behavior of these systems. Programs are the phenomena to be
studied:

Conceptual advances occur by (scientifically) uncontrolled experiments in our

own style of computing . . . the solution lies in more practice and more atten¬
tion to what emerges there as pragmatically successful. (Newell, 1981, p. 4.)
Al: COGNITION AS COMPUTATION 253

Observing our own practice—that is, seeing what the computer implicitly tells
us about the nature of intelligence as we struggle to synthesize intelligent
systems—is a fundamental source of scientific knowledge for us. (Ibid., p. 19.)

Thus, AI is one of the “sciences of the artificial,” as Herbert Simon has

defined them in an influential paper—half of the job is designing systems so
that their performance will be interesting. [Simon, 1969 and 1981.] There is a
valuable heuristic in generating these designs: The systems that we are
naturally inclined to want to build are those that will be useful in our envi¬
ronment. Our environment will shape them, as it shaped us. As Simon
described the development of time-sharing systems:

Most actual designs have turned out initially to exhibit serious deficiencies, and
most predictions of performance have been startlingly inaccurate.

Under these circumstances, the main route open to the development and im¬
provement of time-sharing systems is to build them and see how they behave.
(Ibid., p. 21.)

Genus Symbol Manipulators

Newell and Simon’s psychologically phrased idea of observing the behavior

of programs follows from their twenty-five-year research program in what
they have named “information processing psychology.” Newell and Simon
developed, in the early years of this enterprise, some of the first computer
programs that showed reasoning capabilities. This research on chess¬
playing, theorem-proving, and problem-solving programs was undertaken as
an explicit attempt to model corresponding human behaviors. But Newell
and Simon took the strong position that these programs were not to serve
simply as metaphors for human thought, but were themselves theories. In
fact, they argued that programs were the natural vehicle for expressing
theories in psychology.

An abstract concept of an information processing system has emerged with the

development of the digital computers. In fact, a whole array of different ab¬
stract concepts has developed, as scientists have sought to capture the essence
of the new technology in different ways. . . .

With a model of an information processing system, it becomes meaningful to

try to represent in some detail a particular man at work on a particular task.
Such a representation is not metaphor, but a precise symbolic model on the
basis of which pertinent specific aspects of the man’s problem solving behavior
can be calculated. (Newell and Simon, 1972, p. 5.)

Taking the view that artificial intelligence is theoretical psychology, simulation

(the running of a program purporting to represent some human behavior) is
254 AVRON BARR

simply the calculation of the consequences of a psychological theory. (Newell,

1973c, p. 47.)

... a framework comprehensive enough to encourage and permit thinking is

offered, so that not only answers, but questions, criteria of evidence, and
relevance all become affected. (Ibid., p. 59.)

Newell and Simon, in their view that computer programs are a vehicle for
expressing psychological theories rather than just serving as a metaphor for
mental processes, were already taking a strong position relative to even the
new breed of cognitive psychologists who were talking in terms of com¬
puterlike mental mechanisms. As Paul R. Cohen puts it, in his review of AI
work on models of cognition:

We should note that we have presented the strongest version of the informa¬
tion-processing approach, that advocated by Newell and Simon. Their position
is so strong that it defines information-processing psychology almost by exclu¬
sion: It is the field that uses methods alien to cognitive psychology to explore
questions alien to AI. This is an exaggeration, but it serves to illustrate why
there are thousands of cognitive psychologists, and hundreds of AI research¬
ers, and very few information-processing psychologists. (Cohen, 1982, p. 7.)

However, Newell and Simon did not stop there. A further development in
their thinking identified brains and computers as two species of physical
symbol system—the type of system which, they argue, must underlie any
intelligent behavior.

At the root of intelligence are symbols, with their denotative power and their
susceptibility to manipulation. And symbols can be manufactured of almost
anything that can be arranged and patterned and combined. Intelligence is
mind implemented by any patternable kind of matter. (Simon, 1980, p. 35.)

A physical symbol system has the necessary and sufficient means for general
intelligent action. (Newell and Simon, 1976a, p. 116.)

Information processing psychology is concerned essentially with whether a

successful theory of human behavior can be found within the domain of sym¬
bolic systems. (Newell, 1970, p. 372.)

The basic point of view inhabiting our work has been that programmed com¬
puter and human problem solver are both species belonging to the genus IPS.
(Newell and Simon, 1972, p. 869.)

It is this view of computers—as systems that share a common, underlying

structure with the human intelligence system—that promotes the behavioral
view of AI computer research. Although these machines are not limited by
the rules of development of their natural counterpart, they will be shaped in
their development by the same natural constraints responsible for the form
of intelligence in nature.
Al: COGNITION AS COMPUTATION 255

The Flight Metaphor

The question whether machines could think was certainly an issue in the
early days of AI research, although dismissed rather summarily by those
who shaped the emerging science:

To ask whether these computers can think is ambiguous. In the naive realistic
sense of the term, it is people who think, and not either brains or machines. If,
however, we permit ourselves the ellipsis of referring to the operations of the
brain as “thinking,” then, of course, our computers “think” .... (McCulloch,
1965, p. 368.)

Addressing fundamental issues like this one in their early writing, several
researchers suggested a parallel with the study of flight, considering cogni¬
tion as another natural phenomenon that could eventually be achieved by
machines.

Today, despite our ignorance, we can point to that biological milestone, the
thinking brain, in the same spirit as the scientists many hundreds of years ago
pointed to the bird as a demonstration in nature that mechanisms heavier than
air could fly. (Feigenbaum and Feldman, 1963, p. 8.)

It is instructive to pursue this analogy a bit farther. Flight, as a way of

dealing with the contingencies of the environment, takes many forms—from
soaring eagles to hovering hummingbirds. If we start to study flight by exam¬
ining its forms in nature, our initial understanding of what we are studying
might involve terms like feathers, wings, weight-to-wing-size ratios, and
probably wing-flapping, too. This is the language we begin to develop—
identifying regularities and making distinctions among the phenomena. But
when we start to build flying artifacts, our understanding changes im¬
mediately.

Consider how people came to understand how birds fly. Certainly we observed
birds. But mainly to recognize certain phenomena. Real understanding of bird
flight came from understanding flight; not birds. (Papert, 1972, pp. 1-2.)

Even if we fail a hundred times at building a machine that flies by flapping

its wings, we learn from every attempt. And eventually, we abandon some of
the assumptions implicit in our definition of the phenomena under study and
realize that flight does not require wing movement or even wings.

Intelligent behavior on the part of a machine no more implies complete func¬

tional equivalence between machine and brain than flying by an airplane im¬
plies complete functional equivalence between plane and bird. (Armer, 1963, p.
392.)
256 AVRON BARR

Every new design brings new data about what works, what does not, and
clues as to why. Every new contraption tries some different design alterna¬
tive in the space defined by our theory-language. And every attempt clarifies
our understanding of what it means to fly.
But there is more to the sciences of the artificial than defining the true
nature of natural phenomena. The exploration of the artifacts themselves,
the stiff-winged flying machines, because they are useful to society, will
naturally extend the exploration of various points of interface between the
technology and society. While nature’s exploration of the possibilities is
limited by its mutation mechanism, human inventors will vary every param¬
eter they can think of to produce effects that might be useful—exploring
constraints on the design of their machines from every angle. The space of
“flight” phenomena will be populated by examples that nature has not had a
chance to try.

Exploring the Space of Cognitive Phenomena

This argument, that the utility of intelligent machines will drive the explora¬
tion of their capabilities, suggests that the development of AI technology has
begun an exploration of cognitive phenomena that will involve aspects of
cognition that are not easy to study in nature. In fact, as with the study of
flight, AI will enable us to see natural intelligence as a limited capability, in
terms of design tradeoffs made in the evolution of biological cognition.

Computer science is an empirical discipline. . . . Each new machine that is built

is an experiment. . . . Each new program that is built is an experiment. It poses
a question to nature, and its behavior offers clues to an answer. . . .

We build computers and programs for many reasons. We build them to serve
society and as tools for carrying out the economic tasks of society. But as basic
scientists we build machines and programs as a way of discovering new phe¬
nomena and analyzing phenomena we already know about. . . . the phenomena
surrounding computers are deep and obscure, requiring much experimentation
to assess their nature. (Newell and Simon, 1976u, p. 114.)

For what will AI systems be useful? How will they be involved in the
economic tasks of society? It has certainly been argued that this point is one
that distinguishes biological systems from machines:

The human is a physical symbol system, yes, with a component of pure cogni¬
tion describable by mechanisms. . . . But the human is more: The human is an
animate organism, with a biological basis and an evolutionary and cultural
history. Moreover, the human is a social animal, interacting with others, with
the environment, and with itself. The core disciplines of cognitive science have
tended to ignore these aspects of behavior .... (Norman, 1980, pp. 3-4.)
Al: COGNITION AS COMPUTATION 257

The difference between natural and artificial devices is not simply that they are
constructed of different stuff; their basic functions differ. Humans survive ....
(Ibid., p. 10.)

Tools evolve and survive based on their utility to the people who use them.
Either the users find better tools, or their competitors find them. This pro¬
cess will certainly continue with the development of cognitive tools and will
dramatically change the way we think about AI.

We measure the intelligence of a system by its ability to achieve stated ends in

the face of variations, difficulties and complexities posed by the task environ¬
ment. This general investment of computer science in attaining intelligence . . .
becomes more obvious as we extend computers to more global complex and
knowledge-intensive tasks—as we attempt to make them our agents, capable
of handling on their own the full contingencies of the natural world. (Newell
and Simon, 1976a, pp. 114-115.)

In fact, this change has already begun in AI laboratories, but the place where
the changing perception of AI systems is most dramatic and accelerated is,
not surprisingly in our society, the marketplace.

Al, INC.

To date, three of the emerging AI technologies have attracted interest as

commercial possibilities: robots for manufacturing, natural-language front-
ends for information-retrieval systems, and expert systems. The reason that
a company like General Motors invests millions of dollars in robots for the
assembly line is not scientific curiosity or propaganda about “retooling”
their industry. General Motors believes these robots are essential to its
economic survival. AI technology will surely change many aspects of Ameri¬
can industry, but its application to real problems will just as surely change
emerging technology—change our perception of its nature and its implica¬
tions about knowledge. The remaining discussion focuses on this issue in the
context of expert systems.

Expert Systems

With work on the DENDRAL system in the mid-1960s, AI researchers

began pushing work on problem-solving systems beyond constrained do¬
mains like chess, robot planning, blocks-world manipulations, and puzzles:
They started to consider symbolically expressed problems that were known
to be difficult for the best human researchers to solve. [Lindsay, Buchanan,
Feigenbaum, and Lederberg, 1980.]
258 AVRON BARR

One needs to move toward task environments of greater complexity and open¬
ness—to everyday reasoning, to scientific discovery, and so on. The tasks we
tackled, though highly complex by prior psychological standards, still are sim¬
ple in many respects. (Newell and Simon, 1972, p. 872.)

Humans have difficulty keeping track of all the knowledge that might be
relevant to a problem, exploring all alternative solution-paths, and making
sure none of the valid solutions is overlooked in the process. Work on
DENDRAL showed that when the human expert could explain exactly what
he or she was doing in solving problems, the machine could achieve expert-
level performance.
Continued research at Stanford’s Heuristic Programming Project next
produced the MYCIN system, an experiment in modeling medical diagnostic
reasoning. [Shortliffe, 1976.] In production rules of the form if (condition)
then (action), Shortliffe encoded the kind of information about their rea¬
soning process that physicians were most able to give—advice about what to
do in certain situations. In other words, the if parts of the rules contain
clauses that attempt to differentiate a certain situation, and the then part
describes what to do if we find ourselves in that situation. This production-rule
knowledge representation worked surprisingly well; MYCIN was able to
perform its task in a specific area of infectious-disease diagnosis as well as
the best experts in the country.
Furthermore, the MYCIN structure was seen to be, at least to some
extent, independent of the domain of medicine. So long as experts could
describe their knowledge in terms of if-then rules, the reasoning mechanism
that MYCIN used to make inferences from a large set of rules would come
up with the right questions and eventually, a satisfactory analysis. MYCIN-
like systems have been successfully built in research laboratories for appli¬
cations as diverse as mineral exploration, diagnosis of computer-equipment
failure, and even for advising users about how to use complex systems.

Transfer of Expertise

There is an important shift in the view of expert systems described in the

preceding paragraph, which illustrates the changing perspective on AI that I
have suggested will take place as it becomes an applied science. Early work
on expert systems, building on AI research in problem-solving, focused on
representing and manipulating facts in order to get answers. But through
MYCIN, whose reasoning mechanism is actually quite shallow, it became
clear that the way that these systems interacted with people who had knowl¬
edge and those who needed it was an important, deep constraint on the
system’s architecture—knowledge representations and reasoning mecha¬
nisms.

A key idea in our current approach to building expert systems is that these
programs should not only be able to apply the corpus of expert knowledge to
Al: COGNITION AS COMPUTATION 259

specific problems, but they should also be able to interact with the users and
experts just as humans do when they learn, explain, and teach what they know.
. . . these transfer of expertise (TOE) capabilities were originally necessitated
by “human engineering” considerations—the people who build and use our
systems needed a variety of “assistance” and “explanation” facilities. How¬
ever, there is more to the idea of TOE than the implementation of needed user
features: These social interactions—learning from experts, explaining one’s
reasoning, and teaching what one knows—are essential dimensions of human
knowledge. They are as fundamental to the nature of intelligence as expert-
level problem-solving, and they have changed our ideas about representation
and about knowledge. (Barr, Bennett, and Clancey, 1979, p. 1.)

Randall Davis’s TEIRESIAS system, built within the MYCIN

framework, was the first to focus on transferral aspects of expert systems.
[Davis, 1976.] TEIRESIAS offered aids for experts who were entering
knowledge into the system and for the system’s users. For example, in order
for an expert to figure out why a system has come up with the wrong
diagnosis or is asking an inappropriate question, he or she has to understand
its behavior in his or her own terms: The system must explain its reasoning
in terms of concepts and procedures with which the expert is familiar. The
same type of explanation facility is necessary for the eventual user of an
expert system, who will want to be assured that the system’s answers are
well founded. Expert-systems technology had to be extended to facilitate
these kinds of interactions, and in the process, our conception of what
constituted an expert system has changed. No longer did the systems simply
solve problems; they now transferred expertise from people who had it to
people who could use it.

We are building systems that take part in the human activity of transfer of
expertise among experts, practitioners, and students in different kinds of do¬
mains. Our problems remain the same as they were before: We must find good
ways to represent knowledge and meta-knowledge, to carry on a dialogue, and
to solve problems in the domain. But the guiding principles of our approach
and the underlying constraints on our solutions have subtly shifted: Our sys¬
tems are no longer being designed solely to be expert problem solvers, using
vast amounts of encoded knowledge. There are aspects of “knowing” that
have so far remained unexplored in AI research: by participation in human
transfer of expertise, these systems will involve more of the fabric of behavior
that is the reason we ascribe knowledge and intelligence to people. (Barr,
Bennett, and Clancey, 1979, p. 5.)

The Technological Niche

It is the goal of those who are involved in the commercial development of

expert-systems technology to incorporate that technology into some device
that can be sold. But the environment in which expert systems operate is our
own cognitive environment; it is within this sphere of activity—people solv-
260 AVRON BARR

ing their problems—that the eventual expert-system products must be found

useful. They will be engineered to our minds!

With these systems, it will at last become economical to match human beings in
real time with really large machines. This means that we can work toward
programming what will be, in effect, “thinking aids.” In the years to come we
expect that these man-machine systems will share, and perhaps for a time be
dominant, in our advance toward the development of “artificial intelligence.”
(Minsky, 1963, p. 450.)

It is a long way from the expert systems developed in research laboratories

to any products that fit into peoples’ lives; in fact, it is even hard to envision
what such products will be. Egon Loebner of Hewlett-Packard Laboratories
tells of a conversation he had many years ago with Vladimir Zworykin, the
inventor of television technology. Loebner asked Zworykin what he had in
mind for his invention when he was developing the technology in the
1920s—what kind of product did he think his efforts would produce. The
inventor said that he had had a very clear idea of the eventual use of TV: He
envisioned medical students in the gallery of an operating room getting a
clear picture on their TV screens of the details of the operation being con¬
ducted below them.
We cannot, at the outset, understand the application of a new technology,
because it will find its way into realms of application that do not yet exist.
Egon Loebner has described this process in terms of the technological niche,
paralleling modern evolution theory. [Loebner, 1976; Loebner and Borden,
1969.] Like the species and their environment, inventions and their applica¬
tions are codefined—they constantly evolve together, with niches repre¬
senting periods of relative stability, into a new reality. “Moreover, the
niches themselves are . . . defined in considerable measure by the whole
constellation of organisms themselves. There can be no lice without hairy
heads for them to inhabit, nor animals without plants.” (Simon, 1980, p. 44.)
Thus, technological inventions change as they are applied to people’s needs,
and the activities that people undertake change with the availability of new
technologies. And as people in industry try to push the new technology
toward some profitable niche, they will also explore the nature of the under¬
lying phenomena. Of course, it is not just the scientists and engineers who
developed the new technology who are involved in this exploration: Half of
the job involves finding out what the new capabilities can do for people.
The commercial application of TV technology was foreseen by David
Sarnoff, according to the model he had used for the radio broadcasting
industry. It is important to note that the commercial product that resulted
from TV technology, the TV-set receiver, was only part of a gigantic system
that had to be developed for its support (actually imported from radio, with
modifications and extensions), involving broadcast technology, networks,
regulation of air waves, advertising, and so forth. Loebner refers to this need
Al: COGNITION AS COMPUTATION 261

for system-wide concern with product development as the Edisonian model

of technological innovation: Edison’s achievement of the invention of the
long-life, commercially feasible light bulb was conducted in parallel with his
successful development of the first dynamo for commercially producing
electric power and his design and implementation of the first electric-power
distribution network.

The Knowledge Industry

Among the scientific disciplines that study knowledge, the potential for com¬
mercial applications of artificial intelligence presents unique opportunities.
In order to identify and fill the niches in which intelligent machines will
survive, we must ask questions about knowledge from a rather different
perspective. We must identify the role that various aspects of intelligence
play, or could play, in the affairs of people, in such a way that we can
identify correctable shortcomings in the way things are done.
There is no question that the current best design of an intelligent system,
the human brain, has its limitations. Computers have already been used to
aid people in dealing with such shortcomings as memory failure and confu¬
sions, overloading in busy situations, their tendency to boredom, and their
need for sleep. These extended capabilities—total recall, rapid processing,
and uninterrupted attention—are cognitive capabilities that we have been
willing to concede to the new species in genus symbol manipulators. They
have helped us do the things we did before and have made some entirely new
capabilities possible; for example, airline reservation systems, 24-hour
banking, and Pac-Man (although the truly challenging computer games are
yet to come!). Intelligence is also going to be present in this new species, as
envisioned twenty years ago by Marvin Minsky: “I believe . . . that we are
on the threshold of an era that will be strongly influenced, and quite possibly
dominated, by intelligent problem-solving machines.” (Minsky, 1963, p.
406.) Finding a way to apply this new intellectual capability, for effectively
applying relevant experience to new situations, is the task ahead for AI, Inc.
It may be a while in coming, and it may involve a rethinking of the way we go
about some cognitive activities. But it is extremely important that the devel¬
opment of intelligent machines be pursued, for the human mind is not only
limited in its storage and processing capacity, it also has known bugs: It can
easily be misled, stubborn, and even blind to the truth, especially when
pushed to the limits.
And, as is nature’s way, everything gets pushed to the limits, including
humans. We must find a way of organizing ourselves more effectively, of
bringing together the energies of larger groups of people toward a common
goal. Intelligent systems, built from computer and communications technol¬
ogy, will someday know more than any individual human about what is
going on in complex enterprises involving millions of people, like a multina¬
tional corporation or a city. And intelligent systems will be able to explain
262 AVRON BARR

each person’s part of the task. We will build more productive factories this
way and maybe someday, a more peaceful world. We must keep in mind,
following our analogy with flight, that the capabilities of intelligence as it
exists in nature are not necessarily its natural limits: “There are other facets
to this analogy with flight; it, too, is a continuum, and some once thought
that the speed of sound represented a boundary beyond which flight was
impossible.” (Armer, 1963, p. 398.)
 ARTIFICIAL INTELLIGENCE
Subcognition as Computation

Douglas R. Hofstadter

The philosopher John Searle has recently made quite a stir in the cognitive-
science and philosophy-of-mind circles with his celebrated “Chinese room”
thought experiment, whose purpose is to reveal as illusory the aims of
artificial intelligence (AI), and particularly to discredit what he labels
“strong AI”—the belief that a programmed computer can, in principle, be
conscious. [Searle, 1981.] Various synonymous phrases could be substituted
for “be conscious” here, such as: “think,” “have a soul” (in a humanistic
rather than a religious sense), “have an inner life,” “have semantics” (as
distinguished from “mere syntax”), “have content” (as distinguished from
“mere form”), “have intentionality,” “be something it is like something to
be” (a somewhat ponderous yet appealing phrase due to philosopher
Thomas Nagel), “have personhood,” and others. Each of these phrases has
its own peculiar set of connotations and imagery attached to it as well as its
own history and proponents. For our purposes, however, we shall consider
them all as equivalent and lump them all together, so that the claim of strong
AI now becomes very strong indeed.
At the same time, various AI workers have been developing their own
philosophies of what AI is and have developed some useful terms and slo¬
gans to describe their endeavor. Some of them are: information processing,
cognition as computation, physical symbol system, symbol manipulation,
expert system, and knowledge engineering. There is some confusion as to
what words like symbol and cognition actually mean, just as there is some
confusion as to what words like semantics and syntax mean.
It is the purpose of this paper to try to delve into the meanings of such
elusive terms and, at the same time, shed some light on the views of John
Searle, on the one hand, and Allen Newell and Herbert Simon, on the other
hand—visible AI pioneers who are responsible for several of the terms in the
previous paragraph. The thoughts expressed herein were originally triggered

I would like to thank Daniel C. Dennett and Marsha J. Meredith for their careful reading of, and
valuable comments on, earlier drafts of this paper.

263
264 DOUGLAS R. HOFSTADTER

by the preceding paper called “Artificial Intelligence: Cognition as Compu¬

tation” by Avron Barr. However, they can be read independently of that
paper.
The questions are obviously not trivial and certainly not resolvable in a
single paper. Most of the ideas in this paper, in fact, were stated earlier and
more fully in my book Godel, Escher, Bach: an Eternal Golden Braid.
[Hofstadter, 1979.] However, it seems worthwhile to extract a certain
stream of ideas from that book and enrich it with some more recent musings
and examples, even if the underlying philosophy remains entirely the same.
In order to do justice to these complex ideas, many topics must be interwo¬
ven, and they include the nature of symbols, meaning, thinking, perception,
cognition, and so on. That explains why this paper is not three pages long.

COGNITION VERSUS PERCEPTION: THE 100-MILLISECOND

DIVIDING LINE

In Barr’s paper, AI is characterized repeatedly by the phrase “information¬

processing model of cognition.” Although when I first heard that phrase
years ago, I tended to accept it as defining the nature of AI, something has
gradually come to bother me about it, and I would like to try to articulate
that here.
Now what’s in a word? What’s to object to here? I won't attempt to say
what’s wrong with the phrase so much as try to show what I disagree with in
the ideas of those who have promoted it; then, perhaps, the phrase’s conno¬
tations will float up to the surface, so that other people can see why I am
uneasy with it.
I think the disagreement can be put in its sharpest relief in the following
way. In 1980, Simon delivered a lecture that I attended (the Procter Award
Lecture for the Sigma Xi annual meeting in San Diego), and in it he declared
(and I believe I am quoting him nearly verbatim): “Everything of interest in
cognition happens above the 100-millisecond level—the time it takes you to
recognize your mother.” [Simon, 1981.] Well, our disagreement is simple;
namely, I take exactly the opposite viewpoint: Everything interesting in
cognition happens below the 100-millisecond level—the time it takes you to
recognize your mother. To me, the major queston of AI is this: “What in the
world is going on to enable you to convert from 100,000,000 retinal dots to
one single word mother in one-tenth of a second?” Perception is where it’s
at!

THE PROBLEM OF LETTERFORMS: A TEST CASE FOR

ARTIFICIAL INTELLIGENCE

The problem of intelligence, as I see it, is to understand the fluid nature of

mental categories; to understand the invariant cores of percepts such as your
Al: SUBCOGNITION AS COMPUTATION 265

mother’s face; to understand the strangely flexible yet strong boundaries of

concepts such as chair or the letter a. Years ago, long before computers,
Wittgenstein had already recognized the centrality of such questions in his
celebrated discussion of the “nonpindownability” of the meaning of the
word game. [Wittgenstein, 1922.] To emphasize this and make the point as
starkly as I can, I hereby make the following claim: The central problem of
AI is the question: “What is the letter al" Donald Knuth, on hearing me
make this claim once, appended, “And what is the letter il"—an amend¬
ment that I gladly accept. In fact, perhaps the best version would be this:
The central problem of AI is: “What are A and II" By making these claims,
I am suggesting that for any program to handle letterforms with the flexibility
that human beings do, it would have to possess full-scale general intel¬
ligence.
Many people in AI might protest, pointing out that there already exist
programs that have achieved expert-level performance in specialized do¬
mains without needing general intelligence. Why should letterforms be any
different? My answer would be that specialized domains tend to obscure
rather than clarify the distinction between strengths and weaknesses of a
program. A familiar domain such as letterforms provides much more of an
acid test.
It is strange that AI has said so little about this classic problem. To be
sure, some work has been done. There are a few groups with interest in
letters, but there has been no all-out effort to deal with this quintessential
problem of pattern recognition. Since letterform understanding is currently
the ultimate target of my own research project in AI, I would like to take a
moment and explain why I see it as contrasting so highly with domains at the
other end of the expertise spectrum.
Each letter of the alphabet comes in literally thousands of different official
versions (typefaces), not to mention millions, billions, trillions, of unofficial
versions (those handwritten ones that you and I and everyone else produce
all the time). There thus arises the obvious question, “How are all a's like
each other?” The goal of an AI project would be, of course, to give an exact
answer in computational terms. However, even taking advantage of the
vagueness of ordinary language, we are hard put to find a satisfactory intui¬
tive answer, because we simply come up with phrases such as “they all have
the same shape.” Clearly, the whole problem is that they don’t have the
same shape. And it does not help to change “shape” to “form” or to tack on
phrases such as “basically,” “essentially,” or “at a conceptual level.”
There is also the less obvious question, “How are all the various letters in
a single typeface related to each other?” This is a grand analogy problem if
ever there was an analogy problem. We are asking for a b that is to the
abstract notion of b-ness as a given a is to the abstract notion of a-ness. We
have to take the qualities of a given a and, so to speak, “hold them loosely in
the hand,” as we see how they “slip” into variants of themselves as we try
to carry them over to another letter. Here is the very hingepoint of thought,
the place where one thing slips into alternative, subjunctive, variations on
266 DOUGLAS R. HOFSTADTER

itself. Here, that “thing” is a very abstract concept—namely, the way that
this particular shape manifests the abstract quality of being an a. The prob¬
lem of a is, thus, intimately connected with the problems of b through z and
with that of stylistic consistency.
The existence of optical character readers might lead us to believe at first
that the letter-recognition problem has been solved. If we consider the prob¬
lem a little more carefully, however, we see that the surface has barely been
scratched. In truth, the way that most optical character-recognition pro¬
grams work is by a fancy kind of template matching, in which statistics are
done to determine which character, out of a fixed repertoire of, say, 100
stored characters, is the “best match.” This is about like assuming that the
way I recognize my mother is by comparing the scene in front of me with
stored memories of the appearances of tigers, cigarettes, hula hoops, gam¬
bling casinos, and can openers (and of course all other things in the world
simultaneously) and somehow instantly coming up with the “best match.”

THE HUMAN MIND AND ITS ABILITY TO RECOGNIZE AND

REPRODUCE FORMS

The problem of recognizing letters of the alphabet is no less deep than that of
recognizing your mother, even if it might seem so, given that the number of
Platonic prototype items is on the small side (26, if we ignore all characters
but the lowercase alphabet). We can even narrow it down further—to just a
handful. As a matter of fact, Godfried Toussaint, editor of the pattern-
recognition papers for the IEEE Transactions, has said to me that he would
like to put up a prize for the first program that could say correctly, of 20
characters that people easily can identify, which are a's and which are b's.
To carry out such a task, a program cannot just recognize that a shape is an
a; it has to see how that shape embodies n-ness. And then, as a test of
whether the program really knows its letters, it would have to carry “that
style” over to the other letters of the alphabet. This is the goal of my
research: To find out how to make letters slip in “similar ways to each
other,” so as to constitute a consistent artistic style in a typeface—or simply
a consistent way of writing the alphabet.
By contrast, most AI work on vision pertains to such things as aerial
reconnaissance or robot-guidance programs. This would suggest that the
basic problem of vision is to figure out how to recognize textures and how to
mediate between two and three dimensions. But what about the fact that
although we are all marvelous face-recognizers, practically none of us can
draw a face at all well—even of someone we love? Most of us are flops at
drawing even such simple things as pencils and hands and books. I person¬
ally have learned to recognize hundreds of Chinese characters (shapes that
involve neither three dimensions nor textures) and yet, on trying to repro¬
duce them from memory, find myself often drawing confused mixtures of
Al: SUBCOGNITION AS COMPUTATION 267

characters, leaving out basic components, or, worst of all, being unable to
recall anything but the vaguest “feel” of the character and not being able to
draw a single line.
Closer to home, most of us have read literally millions of, say, k's with
serifs, yet practically none of us can draw a k with serifs in the standard
places. (This holds, of course, for any letter of the alphabet.) I suspect that
many people—perhaps most—are not even consciously aware of the fact
that there are two different types of lowercase a and of lowercase g, just as
many people seem to have a very hard time drawing a distinction between
lowercase and uppercase letters, and a few have a hard time telling letters
drawn forwards from letters drawn backwards.
How can such a fantastic “recognition machine” as our brain be so ter¬
rible at rendition? Clearly, there must be something very complex going on,
enabling us to accept things as members’ categories and to perceive how
they are members of those categories, yet not enabling us to reproduce those
things from memory. This is a deep mystery.
In his book Pattern Recognition, Mikhail Bongard concludes with a series
of 100 puzzles for a visual pattern-recognizer, whether human, machine, or
alien, and, to my mind, it is no accident that he caps off his set with letter-
forms. [Bongard, 1970.] In other words, he works his way up to letterforms
as being at the pinnacle of visual recognition ability. There exists no pattern-
recognition program in the world today that can come anywhere close to
doing those Bongard problems. [Ibid.] And yet Barr cites Simon as writing
the following statement:

The evidence for that commonality [between the information processes that
are employed by such disparate systems as computers and human nervous
systems] is now overwhelming, and the remaining questions about the bound¬
aries of cognitive science have more to do with whether there also exist non¬
trivial commonalities with information processing in genetic systems than
with whether men and machines both think. (Simon, 1980, p. 45.)

I find it difficult to understand how Simon can believe this in an era when
computers still cannot do basic kinds of subcognitive acts (acts that we feel
are unconscious, acts that underlie cognition).
In another lecture in 1979 (the opening lecture of the first meeting of the
Cognitive Science Society, also in San Diego), I recall Simon proclaiming
that despite much doubting by people not in the know, there is no longer any
question as to whether computers can think. If he had meant that there
should no longer be any question about whether machines may eventually
become able to think or whether we humans are machines (in some abstract
sense of the term), then I would be in accord with his statement. But after
hearing and reading such statements over and over again, I don’t think that s
what he meant at all. I get the impression that Simon genuinely believes that
today’s machines are intelligent and that they really do think (or perform
268 DOUGLAS R. HOFSTADTER

“acts of cognition”—to use a bit of jargon that adds nothing to the meaning
but makes it sound more scientific). I will come back to that shortly, since it
is in essence the central bone of contention in this article but first, a few
more remarks on AI domains.

TOY DOMAINS, TECHNICAL DOMAINS, PURE SCIENCE,

AND ENGINEERING

There is a tendency in AI today toward flashy, splashy domains—that is,

toward developing programs that can do such things as medical diagnosis,
geological consultation (for oil prospecting), designing experiments in
molecular biology, molecular spectroscopy, configuring Vax installations,
designing VLSI (very-large-scale integration) circuits, and on and on. Yet,
there is no program that has common sense; no program that learns things
that it has not been explicitly taught how to learn; no program that can
recover gracefully from its own errors. The “artificial expertise” programs
that do exist are rigid, brittle, and inflexible. Like chess programs, they may
serve a useful intellectual or even practical purpose, but despite much fan¬
fare, they are not shedding much light on human intelligence. Mostly, they
are being developed simply because various agencies or industries fund
them.
This does not follow the traditional pattern of basic science. That pattern
is to try to isolate a phenomenon, to reduce it to its simplest possible mani¬
festation. For Newton, this meant the falling apple and the moon; for Ein¬
stein, the thought experiment of the trains and lightning flashes and, later,
the falling elevator; for Mendel, it meant the peas; and so on. You don’t
tackle the messiest problems before you’ve tackled the simpler ones; you
don’t try to run before you can walk. Or, to use a metaphor based on
physics, you don’t try to tackle a world with friction before you’ve got a
solid understanding of the frictionless world.
Why do AI people eschew toy domains? Once, about ten years back, the
MIT “blocks world” was a very fashionable domain. Roberts and Guzman
and Waltz wrote programs that pulled 3-D blocks out of 2-D television-
screen dot matrices. Patrick Winston, building on their work, wrote a pro¬
gram that could recognize instantiations of certain concepts compounded
from elementary blocks in that domain (arch, table, house, and so on). Terry
Winograd wrote a program that could “converse” with a person about activ¬
ities, plans, past events, and some structures in that circumscribed domain.
Gerald Sussman wrote a program that could write and debug simple pro¬
grams to carry out tasks in that domain, thus effecting a simple kind of
learning. [Winston, 1975; Winograd, 1972o; Sussman, 1975.] Why, then, did
interest in this domain suddenly wane?
Surely no one could claim that the domain was exhausted. Every one of
those programs exhibited glaring weaknesses and limitations and specializa-
Al: SUBCOGNITION AS COMPUTATION 269

tions. The domain was phenomenally far from being understood by a single,
unified program. Here, then, was a nearly ideal domain for exploring what
cognition truly is—and it was suddenly dropped. Researchers at Massachu¬
setts Institute of Technology were at one time doing truly basic research on
intelligence and then quit. Much basic research has been supplanted by large
teams marketing what they vaunt as “knowledge engineering.” Firmly
grounded engineering is fine, but it seems to me that this type of engineering
is not built on the solid foundations of a science, but on a number of recipes
that have worked with some success in limited domains.
In my opinion, the proper choice of domain is the critical decision that an
AI researcher makes when beginning a project. If you choose to get involved
in medical diagnosis at the expert level, then you are going to get mired
down in a host of technical problems that have nothing to do with how the
mind works. The same goes for the other earlier-cited ponderous domains
that current work in expert systems involves. By contrast, if you are in
control of your own domain, and can tailor it and prune it so that you keep
the essence of the problem while getting rid of extraneous features, then you
stand a chance of discovering something fundamental.
Early programs on the nature of analogy [Evans, 1968], sequence ex¬
trapolation [Simon and Kotovsky, 1963, among others], and so on, were
moving in the right direction. But then, somehow, it became a common
notion that these problems had been solved. Simply because Thomas Evans
had made a program that could do some very restricted types of visual-
analogy problem “as well as a high school student,” many people thought
the book was closed. However, we need only look at Bongard’s 100 to see
how hopelessly far we are from dealing with analogies. [Bongard, 1970.] We
need only look at any collection of typefaces (look at any magazine’s adver¬
tisements for a vast variety) to see how enormously far we are from under¬
standing letterforms. As I claimed earlier, letterforms are probably the quint¬
essential problem of pattern recognition. It is both baffling and disturbing to
me to see so many people working on imitating cognitive functions at the
highest level of sophistication when their programs cannot carry out cogni¬
tive functions at much lower levels of sophistication.

Al AND THE TRUE NATURE OF INTELLIGENCE

There are some notable exceptions. The Schank group at Yale, whose origi¬
nal goal was to develop a program that could understand natural language,
has been forced to retreat, and to devote most of its attention to the organi¬
zation of memory, which is certainly at the crux of cognition (because it is
part of subcognition, incidentally)—and the group has gracefully accom¬
modated this shift of focus. [Schank, 1982.] I will not be at all surprised,
however, if eventually the group is forced into yet further retreats—in fact,
all the way back to Bongard problems or the like. Why? Simply because
270 DOUGLAS R. HOFSTADTER

their work (on such things as how to discover what “adage” accurately
captures the “essence” of a story or episode) already has led them into the
deep waters of abstraction, perception, and classification. These are the
issues that Bongard problems illustrate so perfectly. Bongard problems are
the idealized (“frictionless”) versions of these critical questions.
It is interesting that Bongard problems are in actuality nothing other than
a well-worked-out set of typical IQ-test problems, the kind that Lewis Ter-
man and Alfred Binet first invented 60 or more years ago. [Terman, 1916;
Binet, 1916.] Over the years, many other less talented people have invented
similar visual puzzles that had the unfortunate property of being filled with
ambiguity and multiple answers. This (among other things) has given IQ
tests a bad name. Whether or not IQ is a valid concept, however, there can
be little question that the original insight of Terman and Binet—that care¬
fully constructed simple visual-analogy problems probe close to the core
mechanisms of intelligence—is correct. Perhaps the political climate created
a kind of knee-jerk reflex in many cognitive scientists, causing them to shy
away from anything that smacked of IQ tests, since issues of cultural bias
and racism began raising their ugly heads. But we need not be so Pavlovian
as to jump whenever a visual-analogy problem is placed in front of us. In
any case, it will be good when AI people are finally driven back to looking
at the insights of people working in the 1920s and 1930s, such as Wittgenstein
and his “games,” Kohler and Koffka and Wertheimer and their “gestalts,”
and Terman and Binet and their IQ-test problems. [Ibid.; Wittgenstein, 1922;
Wertheimer, 1925; Kohler, 1929; Koffka, 1935.]
I was saying that some AI groups seem to be less afraid of toy domains,
or, more accurately put, they seem to be less afraid of stripping down their
domain in successive steps, to isolate the core issues of intelligence that it
involves. Aside from the Schank group, N. Sridharan at Rutgers University
has been doing some very interesting work on prototype deformation,
which, although it springs from work in legal reasoning in a quite messy real-
world domain, has been abstracted into a form in which it is perhaps more
like a toy domain (or, perhaps less pejorative-sounding, an idealized domain)
than at first would appear. [Sridharan, 1980.] The Lindsay-Norman-
Rumelhart group at San Diego has been doing work for years on understand¬
ing errors, such as grammatical slips, typing errors, errors in everyday phys¬
ical actions (such as winding your watch when you mean to switch television
channels), for the insights it may offer into the underlying (subcognitive)
mechanisms. [Norman, 1981.]
Then there are those people who are working on various programs for
perception, whether visual or auditory. One of the most interesting was
Hearsay-II, a speech-understanding program developed at Carnegie-Mellon
University, Simon’s home. It is, therefore, very surprising to me that Simon,
who surely was very aware of the wonderfully intricate and quite beautiful
architecture of Hearsay-II, could then make a comment indicating that per¬
ception and, in general, subcognitive (under 100 milliseconds) processes,
have no interest. [Reddy, 1976.]
AI: SUBCOGNITION AS COMPUTATION 271

There are surely many other less publicized groups that are also working
on more humble domains and on more pure problems of mind, but from
looking at the proceedings of AI conferences, we might get the impression
that, indeed, computers must really be able to think these days, since, after
all, they are doing anything and everything cognitive—from opthalmology to
biology to chemistry to mathematics—even discovering scientific laws from
looking at tables of numerical data, to mention one project (“Bacon”) that
Simon has been involved in.

EXPERT SYSTEMS VERSUS HUMAN FLUIDITY

The problem is that AI programs are carrying out all these cognitive activi¬
ties in the absence of any subcognitive activity. There is no substrate that
corresponds to what goes on in the brain. There is no fluid recognition and
recall and reminding. These programs have no common sense, little sense of
similarity or repetition or pattern. They can perceive some patterns as long
as they have been anticipated—and, particularly, as long as the place where
they will occur has been anticipated—but they cannot see patterns where
nobody told them explicitly to look. They do not learn at a high level of
abstraction.
This style is in complete contrast to how people are. People perceive
patterns anywhere and everywhere, without knowing in advance where to
look. People learn automatically in all aspects of life. These are just facets of
common sense. Common sense is not an area of expertise, but a general
that is, domain-independent—capacity that has to do with fluidity in repre¬
sentation of concepts, an ability to sift what is important from what is not, an
ability to find unanticipated analogical similarities between totally different
concepts (reminding, as Schank calls it). We have a long way to go before
our programs exhibit this cognitive style.
Recognition of one’s mother’s face is still nearly as much of a mystery as
it was 30 years ago. And what about such things as recognizing family
resemblances between people, recognizing a French face, recognizing kind¬
ness or earnestness or slyness or harshness in a face? Even recognizing
age—even sex!—these are fantastically difficult problems! As Donald
Knuth has pointed out, we have written programs that can do wonderfully
well at what people have to work very hard at doing consciously (e.g., doing
integrals, playing chess, medical diagnosis, etc.)—but we have yet to write a
program that remotely approaches our ability to do what we do without
thinking or training—things like understanding a conversation partner with
an accent at a loud cocktail party with music blaring in the background,
while at the same time overhearing wisps of conversations in the far corner
of the room. [Knuth, 1974a.] Or perhaps finding our way through a forest on
an overgrown trail. Or perhaps just doing some anagrams absentmindedly
while washing the dishes.
Asking for a program that can discover new scientific laws without having
 272 DOUGLAS R. HOFSTADTER

a program that can, say, do anagrams, is like wanting to go to the moon

without having the ability to find your way around town. I do not make the
comparison idly. The level of performance that Simon and his colleague
Langley wish to achieve in “Bacon” is on the order of the greatest scien¬
tists. It seems they feel that they are but a step away from the mechanization
of genius. After his Procter Lecture, Simon was asked by a member of the
audience, “How many scientific lifetimes does a five-hour run of ‘Bacon’
represent?” He replied, “Probably not more than one.”

ANAGRAMS AND EPIPHENOMENA

Well, I feel we’re much further away from human-level performance than
Simon does. I, for one, would like to see a program that does anagrams the
way a person does. Why anagrams? Because they are a toy domain where
some very significant subcognitive processes play the central role.
What I mean is this. When we look at a “jumble” such as telkin in the
newspaper, we immediately begin shifting around letters into tentative
groups, making such stabs as knitle, klinte, linket, keltin, tinkle—and then
we notice that, indeed, tinkle is a word. The part of this process that I am
interested in is the part that precedes the recognition of tinkle as a word. It’s
that part that involves experimentation, based only on the style or “feel” of
English words—using intuitions about letter affinities, plausible clusters and
their stabilities, syllable qualities, and so on. When we first read a jumble in
the newspaper, we play around, rearranging, regrouping, reshuffling, in
complex ways that we have no control over. In fact, it feels as if we throw
the letters up into the air separately, and when they come down, they have
somehow magically “glommed” together in some English-like word! It’s a
marvelous feeling—and it is anything but cognitive, anything but conscious.
(Yet, interestingly, we take credit for being good at anagrams, if we are
good!)
It turns out that most literate people can handle “jumbles” (my term for
single-word anagrams) of five or six letters, sometimes seven or eight letters;
with practice, maybe even ten or twelve. But beyond that, it gets very hard
to keep the letters in our head. It is especially hard if there are repeated
letters, since we tend to get confused about which letters there are multiple
copies of. (In one case, I rearranged the letters dinnal into nadlid—
incorrectly. You can try raregarden if you dare.) Now in one sense, the fact
that the problem gets harder and harder with more and more letters is hardly
surprising. It is obviously related to the famous “7 plus or minus 2” figure
that psychologist George A. Miller first reported in connection with short¬
term memory capacity. [Miller, 1956.] But there are different ways of inter¬
preting such a connection.
One way to think that this might come about is to assume that concepts
for the individual letters get ‘activated” and then interact. When too many
Al: SUBCOGNITION AS COMPUTATION 273

get activated simultaneously, then we get swamped with combinations and

we drop some letters and make too many of others, and so on. This view
would say that we simply encounter an explosion of connections, and our
system gets overloaded. The view does not postulate any explicit storage
location in memory—a fixed set of registers or data structures—in which
letters get placed and then shoved around. In this model, short-term memory
and its associated magic number is an epiphenomenon (or innocently emer¬
gent phenomenon, as Daniel Dennett calls it [Dennett, 1978a]), by which I
mean it is a consequence that emerges out of the design of the system, a
product of many interacting factors; something that was not necessarily
known, predictable, or even anticipated to emerge at all. This is the view
that I advocate.
A contrasting view might be to build a model of cognition in which we
have an explicit structure called short-term memory that contains about
seven (or five, or nine) “slots” into which certain data structures can be
fitted, and when it is full, well, then, it is full and we have to wait until an
empty slot opens up. This is one approach that has been followed by Newell
and associates in work on production systems. The problem with this ap¬
proach is that it takes something that clearly is a very complex consequence
of underlying mechanisms and simply plugs it in as an explicit structure,
bypassing the question of what those underlying mechanisms might be. It is
difficult for me to believe that any model of cognition based on such a
“bypass” could be an accurate model.
When an operating system begins thrashing at around 35 users, do you tell
the systems programmer, “Hey, go raise the thrashing number in Tenex
from 35 to 60, okay?”? The number 35 is not stored in some magic location
in Tenex, so that it can be modified; that number comes out of a host of
strategic decisions made by the designers of Tenex and the computer’s
hardware, and so on. There is no “thrashing-threshold dial” to crank on an
operating system, unfortunately. Why should there be a short-term-
memory-size dial on an intelligence? Why should 7 be a magic number built
into the system explicitly from the start? If the size of short-term memory
really were explicitly stored in our genes, then surely it would take only a
simple mutation to reset the dial at 8 or 9 or 50, so that intelligence would
evolve at ever-increasing rates. I doubt that AI people think that this is even
remotely close to the truth; and yet they sometimes act as if it made sense to
assume it is a close approximation to the truth.
It is standard practice for AI people to bypass epiphenomena (collective
phenomena, if you prefer) by simply installing structures that mimic the
superficial features of those epiphenomena. (Such mimics are the
“shadows” of genuine cognitive acts, as John Searle calls them in his pro¬
vocative paper “Minds, Brains, and Programs.” [Searle, 1981.]) The expec¬
tation—or at least the hope—is for tremendous performance to issue forth;
yet, the systems lack the complex underpinning necessary.
The anagrams problem is one that exemplifies mechanisms of thought that
274 DOUGLAS R. HOFSTADTER

AI people have not explored. How do those letters swirl among one another,
fluidly and tentatively making and breaking alliances? “Glomming” to¬
gether, then coming apart, almost like little biological objects in a cell. AI
people have not paid much attention to such problems as anagrams. Perhaps
they would say that the problem is already solved. After all, a virtuoso
programmer has made a program print out all possible words that anagram¬
matize into other words in English. Or perhaps they would point out that in
principle, you can do an “alphabetize” followed by a “hash” and thereby
retrieve from any given set of letters all the words they anagrammatize into.
Well, this is all fine and dandy, but it is really beside the point. It is merely a
show of brute force and has nothing to contribute to our understanding of
how we actually do anagrams ourselves, just as most chess programs have
absolutely nothing to say about how chess masters play (as Adriaan de
Groot, and later, William Chase and Herbert Simon have pointed out), [de
Groot, 1965; Chase and Simon, 1973.]
Is anagrams simply a trivial, silly, “toy” domain? Or is it serious? I
maintain that it is a far purer, far more interesting domain than many of the
complex real-world domains of the expert systems, precisely because it is so
playful, so unconscious, so enjoyable for people. It is obviously more re¬
lated to creativity and spontaneity than it is to logical derivations, but that
does not make it—or the mode of thinking that it represents—any less
worthy of attention. In fact, because it epitomizes the unconscious mode of
thought, I think it more worthy of attention.
In short, it seems to me that something fundamental is missing in the
orthodox AI “information-processing” model of cognition, and that is some
sort of substrate from which intelligence emerges as an epiphenomenon.
Most AI people do not want to tackle that kind of underpinning work. Could
it be that they really believe that machines already can think, already have
concepts, already can do analogies? It seems that a large camp of AI people
really do believe these things.

NOT COGNITION, BUT SUBCOGNITION, IS COMPUTATIONAL

Such beliefs arise, in my opinion, from a confusion of levels, exemplified by

part of the title of Barr’s paper, “Cognition as Computation.” Am I really
computing when I think? Admittedly, my neurons may be performing sums
in an analog way, but does this pseudo-arithmetical hardware mean that the
epiphenomena themselves are also doing arithmetic or should be—or even
can be—described in conventional computer science terminology? Does the
fact that taxis stop at red lights mean that traffic jams stop at red lights? We
should not confuse the properties of objects with the properties of statistical
ensembles of those objects. In this analogy, traffic jams play the role of
thoughts, and taxis play the role of neurons or neuron-firings. It is not meant
to be a serious analogy, only one that emphasizes that what you see at the
Al: SUBCOGNITION AS COMPUTATION 275

top level need not have anything to do with the underlying swarm of activi¬
ties bringing it into existence. In particular, something can be computational
at one level but not at another level.
Yet, many AI people, despite considerable sophistication in thinking
about a given system at different levels, still seem to miss this. Most AI work
goes into efforts to build rational thought (cognition) out of smaller rational
thoughts (elementary steps of deduction, for instance, or elementary mo¬
tions in a tree). It comes down to thinking that what we see at the top level of
our minds—our ability to think—comes out of rational “information¬
processing” activity, with no deeper levels below that.
Many interesting ideas, in fact, have been inspired by this hope. I find
much of the work in AI to be fascinating and provocative, yet somehow I
feel dissatisfied with the overall trend. For instance, there are some people
who believe that the ultimate solution to AI lies in getting better and better
theorem-proving mechanisms in some predicate calculus. They have devel¬
oped extremely efficient and novel ways of thinking about logic. Some peo¬
ple—Simon and Newell, particularly—have argued that the ultimate solu¬
tion lies in getting more and more efficient ways of searching a vast space of
possibilities. (They refer to selective heuristic search as the key mechanism
of intelligence.) Again, many interesting discoveries have come out of this.
Then there are others who think that the key to thought involves making
some complex language in which pattern-matching or backtracking or inher¬
itance or planning or reflective logic is easily carried out. Now admittedly,
such systems, when developed, are good for solving a large class of prob¬
lems, exemplified by such AI chestnuts as the missionary-and-cannibals
problem, “cryptarithmetic” problems, retrograde chess problems, and
many other specialized sorts of basically logical analysis. However, these
kinds of techniques of building small logical components up to make large
logical structures have not proven good for such things as recognizing your
mother or drawing the alphabet in a novel and pleasing way.
One group of AI people who seem to have a different attitude consists of
those who are working on problems of perception and recognition. There,
the idea of coordinating many parallel processes is important, as is the idea
that pieces of evidence can add up in a self-reinforcing way, so as to bring
about the locking-in of a hypothesis that no one of the pieces of evidence
could on its own justify. It is not easy to describe the flavor of this kind ot
program architecture without going into multiple technical details. How¬
ever, it is very different in flavor from ones operating in a world where
everything comes clean and precategorized—where everything is specified
in advance: “There are three missionaries and three cannibals and one boat
and one river and . . . ,” which is immediately turned into a predicate-
calculus statement or a frame representation, ready to be manipulated by an
“inference engine.” The missing link seems to be the one between percep¬
tion and cognition, which I would rephrase as the link between subcognition
and cognition, that gap between the sub-100-millisecond world and the
super-100-millisecond world.
 276 DOUGLAS R. HOFSTADTER

Earlier, I mentioned the brain and referred to the “neural substrate” of

cognition. Although 1 am not pressing for a neurophysiological approach to
AI, I am unlike many AI people in that I believe that any AI model even¬
tually has to converge to brainlike hardware or at least to an architecture
that at some level of abstraction is isomorphic to brain architecture (also at
some level of abstraction). This may sound empty, since that level could be
anywhere, but I believe that the level at which the isomorphism must apply
will turn out to be considerably lower than (I think) most AI people believe.
This disagreement is intimately connected to the question of whether cogni¬
tion should or should not be described as computation.

PASSIVE SYMBOLS AND FORMAL RULES

One way to explore this disagreement is to look at some of the ways that
Simon and Newell express themselves about symbols.

At the root of intelligence are symbols, with their denotative power and their
susceptibility to manipulation. And symbols can be manufactured of almost
anything that can be arranged and patterned and combined. Intelligence is
mind implemented by any patternable kind of matter. (Simon, 1980, p. 35.)

From this quotation and from others, we can see that to Simon and Newell, a
symbol seems to be any token, any character inside a computer that has an
ASCII code (a standard but arbitrarily assigned sequence of seven bits). To
me, by contrast, symbol connotes something with representational power.
To them (if I am not mistaken), it would be fine to call a bit (inside a
computer) or a neuron-firing a symbol; however, I cannot feel comfortable
with that usage of the term.
To me, the crux of the word symbol is its connection with the verb to
symbolize, which means to denote, to represent, to stand for, and so on.
Now in the preceding quotation, Simon refers to the denotative power of
symbols—yet, in another part of his paper, Barr quotes Newell and Simon
as saying that thought is “the manipulation of formal tokens.” [Newell and
Simon, 1976a.] It is not clear to me which side of the fence they really are
on.
It takes an immense amount of richness for something to represent some¬
thing else. The letter I does not in and of itself stand for the person I am or
for the concept of selfhood. That quality comes to it from the way that the
word behaves in the totality of the English language. It comes from a mas¬
sively complex set of usages and patterns and regularities, ones that are
regular enough for babies to be able to detect, so that they, too, eventually
come to say “I” to talk about themselves.
Formal tokens such as I or hamburger are in themselves empty; they do
not denote. Nor can they be made to denote in the full rich intuitive sense of
Al: SUBCOGNITION AS COMPUTATION 111

the term by having them obey some rules. You can’t simply push around
some Pnames of LISP atoms according to complex rules and hope to come
out with genuine thought or understanding. (This, by the way, is probably a
charitable way to interpret Searle’s point in his previously mentioned pa¬
per—namely, as a rebellion against claims that programs that can manipu¬
late tokens such as John, ate, a, hamburger actually have understanding.)
Manipulation of empty tokens is not enough to create understanding—
although it is enough to imbue them with meaning in a limited sense of the
term, as I stress repeatedly in my book Godel, Escher, Bach: an Eternal
Golden Braid. [Hofstadter, 1979, particularly chaps. 2-6.]

ACTIVE SYMBOLS AND THE ANT COLONY METAPHOR

So what is enough? What am I advocating? What do I mean by symbol? I

gave an exposition of my concept of active symbols. [Ibid., chaps. 11-12.]
However, the notion was first presented in that book in the Dialogue “Pre¬
lude, Ant Fugue,” which revolved about a hypothetical conscious ant col¬
ony. [Hofstadter, 1979, pp. 311-336.] The purpose of the discussion was not
to speculate about whether ant colonies are conscious or not, but to set up
an extended metaphor for brain activity—a framework in which to discuss
the relation between holistic, or collective, phenomena, and the microscopic
events that make them up.
One of the ideas that inspired the Dialogue has been stated by Edward O.
Wilson in his book The Insect Societies this way: “Mass communication is
defined as the transfer, among groups, of information that a single individual
could not pass to another.” (E. O. Wilson, 1971, p. 226.) We have to imagine
teams of ants cooperating on tasks and that information passes from team to
team that no ant is aware of (if ants indeed are “aware” of information at
all—but that is another question). We can carry this up a few levels and
imagine hyperhyperteams carrying and passing information that no hyper¬
team, not to mention team or solitary ant, ever dreamt of.
I feel it is critical to focus on collective phenomena, particularly on the
idea that some information or knowledge or ideas can exist at the level of
collective activities while being totally absent at the lowest level. In fact, we
can even go so far as to say that no information exists at that lowest level. It
is hardly an amazing revelation, when transported back to the brain: namely,
that no ideas are flowing in those neurotransmitters that spark back and forth
between neurons. Yet, such a simple notion undermines the idea that
thought and “symbol manipulation” are the same thing, if by symbol we
mean a formal token such as a bit or a letter or a LISP Pname.
What is the difference? Why couldn’t symbol manipulation—in the sense
that I believe Simon and Newell and many writers on AI mean it—
accomplish the same thing? The crux of the matter is that these people see
symbols as lifeless, dead, passive objects—things to be manipulated by
278 DOUGLAS R. HOFSTADTER

some overlying program. I see symbols—representational structures in the

brain (or perhaps someday in a computer)—as active, like the imaginary
hyperhyperteams in the ant colony. That is the level at which denotation
takes place, not at the level of the single ant. The single ant has no right to be
called “symbolic,” because its actions stand for nothing. (Of course, in a
real ant colony, we have no reason to believe that teams at any level
genuinely stand for objects outside the colony (or inside it, for that matter)—
but the ant colony metaphor is only a thinly disguised way of making discus¬
sion of the brain more vivid.) [Wheeler, 1911; Marais, 1937; Meyer, 1966.]

WHO SAYS ACTIVE SYMBOLS ARE COMPUTATIONAL ENTITIES?

It is the vast collections of ants (read neural firings, if you prefer) that add up
to something genuinely symbolic. And who can say whether there exist
rules—formal, computational rules—at the level of the teams themselves
(read concepts, ideas, thoughts) that are of full predictive power in describ¬
ing how they will flow? I am speaking of rules that allow us to ignore what is
going on “down below” yet still yield perfect or at least very accurate
predictions of the teams’ behavior.
To be sure, there are phenomenological observations that can be for¬
malized to sound like rules that will describe, very vaguely, how those
highest-level teams act. But what guarantee is there that we can skim off the
full fluidity of the top-level activity of a brain, and encapsulate it—without
any lower substrate—in the form of some computational rules?
To ask an analogous question, what guarantee is there that there are rules
at the cloud level (more properly speaking, the level of cold fronts, isobars,
trade winds, etc.) that will allow us to say accurately how the atmosphere is
going to behave on a large scale? Perhaps there are no such rules; perhaps
weather prediction is an intrinsically intractable problem. Perhaps the be¬
havior of clouds is not expressible in terms that are computational at then-
own level, even if the behavior of the microscopic substrate—the
molecules—is computational.
The premise of AI is that thoughts themselves are computational entities
at their own level. At least, this is the premise of the information-processing
school of AI, and I have very serious doubts about it.
The difference between my active symbols (teams) and passive informa¬
tion-processing symbols (ants, tokens) is that the active symbols flow and
act on their own. In other words, there is no higher-level agent (read pro¬
gram) that reaches down and shoves them around. Active symbols must
incorporate within their own structures the wherewithal to trigger and cause
actions. They cannot just be passive storehouses, bins, or receptacles of
data. Yet to Newell and Simon, it seems, even so tiny a thing as a bit is a
symbol. This is brought out repeatedly in their writings on physical symbol
systems.
Al: SUBCOGNITION AS COMPUTATION 279

A good term for the little units that a computer manipulates (as well as for
neuron firings) is tokens. All computers are good at token manipulation;
however, only some—the appropriately programmed ones—could support
active symbols. (I prefer not to say that they would carry out symbol manip¬
ulation, since that gets back to that image of a central program shoving
around some passive representational structures.) The point is, in such a
hypothetical program (and none exists as yet), the symbols themselves are
acting!
A simple analogy from ordinary programming might help to convey the
level distinction that I am trying to make here. When a computer is running a
LISP program, does it do function calling? To say yes would be unconven¬
tional. The conventional intuition is that functions call other functions and
the computer is simply the hardware that supports function-calling activity.
In somewhat the same sense, although with much more parallelism, symbols
activate or trigger or awaken other symbols in a brain.
The brain itself does not manipulate symbols; the brain is the medium in
which the symbols are floating and in which they trigger each other. There is
no central manipulator, no central program. There is simply a vast collection
of “teams”—patterns of neural firings that, like teams of ants, trigger other
patterns of neural firings. The symbols are not “down there ’ at the level of
the individual firings; they are “up here” where we do our verbalization. We
feel those symbols churning within ourselves in somewhat the same way we
feel our stomach churning. We do not do symbol manipulation by some sort
of act of will, leave alone some set of logical rules of deduction. We cannot
decide what we will next think of nor how our thoughts will progress.
Not only are we not symbol manipulators; in fact, quite to the contrary,
we are manipulated by our symbols! As Scott Kim has put it, rather than
speak of “free will,” perhaps it is more appropriate to speak of “free
won’t.” This way of looking at things turns everything on its head, placing
cognition—that rational-seeming level of our minds—where it belongs,
namely, as a consequence of much deeper processes of myriads of interact¬
ing subcognitive structures. The rational has had entirely too much made of
it in AI research; it is time for some of the irrational and subcognitive to be
recognized (no pun intended) for its pivotal role.

THE SUBSTRATE OF ACTIVE SYMBOLS DOES NOT SYMBOLIZE

“Cognition as computation” sounds right to me only if I interpret it quite

liberally, namely, as meaning “cognition is an activity that can be supported
by computational hardware. ” But if I interpret it more strictly as cognition
is an activity that can be achieved by a program that shunts around meaning¬
carrying objects called symbols in a complicated way,’ then I don t buy it.
In my view, meaning-carrying objects won’t submit to being shunted about
(it’s demeaning); meaning-carrying objects carry meaning only by viitue of
280 DOUGLAS R. HOFSTADTER

being active, autonomous agents themselves. There can’t be an overseer

program, a pusher-around.
To paraphrase a question asked by neurophysiologist Roger Sperry,
“Who shoves whom around inside the computer?” (He asked it of the
cranium.) [Sperry, 1965, p. 79.] If some program shoves data structures
around, then you can bet it’s not carrying out cognition; or more precisely, if
the data structures are supposed to be meaning-carrying, representational
things, then it’s not cognition. Of course, at some level of description, pro¬
grams certainly will be shoving formal tokens around, but it’s only agglomer¬
ations of such tokens en masse that, above some unclear threshold of collec¬
tivity and cooperativity, achieve the status of genuine representation. At
that stage, the computer is not shoving them around any more than our brain
is shoving thoughts around! The thoughts themselves are causing flow. (This
is, I believe, in agreement with Sperry’s own way of looking at matters.
[See, for example, Sperry, 1965.]) Parallelism and collectivity are of the
essence, and in that sense, my response to the title of Barr’s paper is no,
cognition is not computation.
At this point, some people might think that I sound like Searle, suggesting
that there are elusive causal powers of the brain that cannot be captured
computationally. [Searle, 1981.] I hasten to say that this is not my point of
view at all! In my opinion, AI—even Searle’s “strong AI”—is still possible,
but thought will simply not turn out to be the formal dream of people in¬
spired by predicate calculus or other formalisms. Thought is not a formal
activity whose rules exist at that level.
Many linguists have maintained that language is a human activity whose
nature could be entirely explained at the linguistic level—in terms of com¬
plex grammars, without recourse or reference to anything such as thoughts
or concepts. Now many AI people are making a similar mistake: They think
that rational thought simply is composed of elementary steps, each of which
has some interpretation as an atom of rational thought, so to speak. That’s
just not what is going on, however, when neurons fire. On its own, a neuron
firing has no meaning, no symbolic quality whatsoever. I believe that those
elementary events at the bit level—even at the LISP function level (if AI is
ever achieved in LISP, something I seriously doubt)—will have the same
quality of having no interpretation. It is a level shift as drastic as that
between molecules and gases that takes place when thought emerges from
billions of in-themselves-meaningless neural firings.
A simple metaphor, hardly demonstrating my point but simply giving its
flavor, is provided by Winograd’s program SHRDLU, which, using the full
power of a DEC-10 computer, could deal with whole numbers up to ten in a
conversation about the “blocks world.” [Winograd, 1972a.] It knew noth¬
ing ds cognitive level—of larger numbers. Alan Turing invents a similar
example, a rather sly one, where he has a human ask a computer to do a
sum, and the computer pauses thirty seconds and then answers incorrectly.
[Turing, 1950.] Now this need not be a ruse on the computer’s part. It might
Al: SUBCOGNITION AS COMPUTATION 281

genuinely have tried to add the two numbers at the symbol level and made a
mistake, just as you or I might have, despite having neurons that can add
fast.
The point is simply that the lower-level arithmetical processes (the adds,
the shifts, the multiplies, etc.) out of which the higher level of any AI pro¬
gram is composed are completely shielded from its view. To be sure,
Winograd could have artificially allowed his program to write little pieces of
LISP code that would execute and return answers to questions in English
such as “What is 720 factorial?” but that would be similar to our trying to
take advantage of the fact that we have billions of small analog adders in our
brain some time when we are trying to check a long grocery bill. We simply
don’t have access to those adders! We can’t reach them.

SYMBOL TRIGGERING PATTERNS ARE THE ROOTS OF MEANING

What’s more, we oughtn’t to be able to reach them. The world is not

sufficiently mathematical for that to be useful in survival. What good would
it do a spear thrower to be able to calculate parabolic orbits when in reality
there is wind and drag, and the spear is not a point mass—and so on? It’s
quite the contrary: A spear thrower does best by being able to imagine a
cluster of approximations of what may happen and anticipating some plau¬
sible consequences of them.
As Jacques Monod in Chance and Necessity and Richard Dawkins in The
Selfish Gene both point out, the real power of brains is that they allow their
owners to simulate a variety of plausible futures. [Monod, 1971; Dawkins,
1976.] This is to be distinguished from the exact prediction of eclipses by
iterating differential equations step by step far into the future, with very high
precision. The brain is a device that has evolved in a less exact world than
the pristine one of orbiting planets, and there are always far more chances
for the best-laid plans to “gang agley.” Therefore, mathematical simulation
has to be replaced by abstraction, which involves discarding the irrelevant
and making shrewd guesses based on analogy with past experience. Thus,
the symbols in a brain, rather than playing out a scenario precisely
isomorphic with what actually will transpire, play out a few scenarios that
are probable or plausible or even some scenarios from the past that may
have no obvious relevance other than as metaphors. (This brings us back to
the “adages” of the Yale group. [Schank, 1982.])
Once we abandon perfect mathematical isomorphism as our criterion for
symbolizing and suggest that the value of symbol-triggering patterns comes
largely from their suggestive value and their metaphorical richness, this
severely complicates the question of what it means when we say that a
symbol in the brain symbolizes anything. This is closely related to perhaps
one of the subtlest issues, in my opinion, that AI should be able to shed light
on, and that is the question “What is meaning?” This is actually the crucial
282 DOUGLAS R. HOFSTADTER

issue that Searle is concerned with in his earlier-mentioned attack on AI;

although he camouflages it, and sometimes loses track of it by all sorts of
evasive maneuvers, it turns out in the end that what he is truly concerned
with is the idea that computers have no semantics—and he of course means
computers do not now have, and never will have, semantics. [Searle, 1982.]
If he were talking only about the present, I would agree. However, he is
making a point in principle, and I believe he is wrong there.
Where do the meanings of the so-called active symbols, those giant clouds
of neural activity in the brain, come from? To what do they owe their de-
notational power? Some people have maintained that it is because the brain
is physically attached to sensors and effectors that connect it to the outside
world, enabling those clouds to mirror the actual state of the world (or at
least some parts of it) faithfully and to affect the world outside as well
through the use of the body. I think that those things are part of denotational
power, but not its crux. When we daydream or imagine situations, when we
dream or plan, we are not manipulating the concrete physical world, nor are
we sensing it. In imagining fictional or hypothetical or even totally impossi¬
ble situations, we are still making use of, and contributing to, the meaning¬
fulness of our symbolic neural machinery. However, the symbols do not
symbolize specific, real, physical objects. The fundamental active symbols
of the brain represent semantic categories—classes, in AI terminology.
Categories do not point to specific physical objects. However, they can be
used as masters off of which copies—instances—can be rubbed, and then
those copies are activated in various conjunctions. These activations then
automatically trigger other instance-symbols into activations of various sorts
(teams of ants triggering the creation of other teams of ants, sometimes
themselves fizzling out). The overall activity will be semantic—
meaningful—if it is isomorphic, not to some actual event in the real world,
but to some event that is compatible with all the known constraints on the
situation.
Those constraints are not at the molecular or any such fine-grained level;
they are at the rather coarse-grained level of ordinary perception. They are
to some extent verbalizable constraints. If I utter the Schankian cliche,
“John went to a restaurant and ate a hamburger,” there is genuine represen¬
tational power in the patterns of activated symbols that your brain sets up,
not because some guy named John actually went out and ate a hamburger
(although, most likely, this is a situation that has at some time occurred in
the world), but because the symbols, with their own “lives” (autonomous
ways of triggering other symbols) will if left alone cause the playing-out of an
imaginary yet realistic scenario. [Note added in press: I have it on good
authority that one John Findling of Floyds Knobs, Indiana, did enter a
Burger Queen restaurant and did eat one (1) hamburger. This fact, though
helpful, would not, through its absence, have seriously marred the argu¬
ments of the present paper.]
Thus, the key thing that establishes meaningfulness is whether or not the
Al: SUBCOGNITION AS COMPUTATION 283

semantic categories are “hooked up” in the proper ways, so as to allow

realistic scenarios to play themselves out on this inner stage. That is, the
triggering patterns of active symbols must mirror the general trends of how
the world works as perceived on a macroscopic level rather than mirror the
actual events that transpire.

BEYOND INTUITIVE PHYSICS: THE CENTRALITY OF SLIPPABILITY

Sometimes this capacity is referred to as intuitive physics. Intuitive physics

is certainly an important ingredient of the triggering patterns needed for an
organism’s comfortable survival. John McCarthy gives the example of per¬
sons able to avoid moving their coffee cups in a certain way, because they
can anticipate how they might spill and coffee might get all over their
clothes. Note that what is computed is a set of alternative rough descriptions
for what might happen rather than one exact trajectory. This is the nature of
intuitive physics.
However, as I stated earlier, there is much more required for symbols to
have meaning than simply that their triggering patterns yield an intuitive
physics. For instance, if you see someone in a big heavy leg cast and they
tell you that their kneecap was acting up, you might think to yourself,
“That’s quite a nuisance, but it’s nothing compared to my friend who has
cancer.” Now this connection is obviously caused by triggering patterns
having to do with symbols representing health problems. But what does this
have to do with the laws of motion governing objects or fluids? Precious
little. Sideways connections like this, having nothing to do with causality,
are equally essential in allowing us to place situations in perspective—to
compare what actually is with what, to our way of seeing things, might have
been or might even come to be. This ability, no less than intuitive physics, is
a central aspect of what meaning is.
This way in which any situation that is perceived seems to be surrounded
by a cluster, a halo, of alternative versions of itself, of variations suggested
by slipping any of a vast number of features that characterize the situation,
seems to me to be at the dead center of thinking. Not much AI work seems to
be going on at present (Schank’s and Sridharan’s groups excepted, per¬
haps—and I ought to include myself as another maverick investigating these
avenues) to mirror this kind of “slippability.” This is an issue that I covered
in some detail in Godel, Escher, Bach under various headings such as slip¬
pability,” “subjunctive instant replays,” “‘almost’ situations, “concep¬
tual skeletons and conceptual mapping,” “alternity” (a term due to George
Steiner), and so on. [Hofstadter, 1979; Steiner, 1975; Hofstadter, 1981;
1982a.]
If we return to the metaphor of the ant colony, we can envision these
“symbols with halos” as hyperhyperteams of ants, many of whose members
are making what appear to be strange forays in random directions, like
 284 DOUGLAS R. HOFSTADTER

flickering tongues of flame spreading out in many directions at once. These

tentative probes, which allow the possibility of all sorts of strange lateral
connections as from kneecap to cancer, have absolutely no detrimental ef¬
fect on the total activity of the hyperhypertcam. In fact, quite to the con¬
trary: The hyperhyperteam depends on its members to go wherever their
noses lead them. The thing that saves the team—what keeps it coherent—is
simply the regular patterns that are sure to emerge out of a random substrate
when there are enough constituents; statistics, in short.
Occasionally, some group of wandering scouts will cause a threshold of
activity to be reached in an unexpected place, and then a whole new area of
activity springs up—a new high-level team is activated (or, to return to the
brain terminology, a new symbol is awakened). Thus, in a brain as in an ant
colony, high-level activity spontaneously flows around, driven by the myr¬
iad lower-level components’ autonomous actions.

THE COAL OF Al SHOULD BE TO BRIDGE THE GAP BETWEEN

COGNITION AND SUBCOGNITION

Let me, for a final time, make clear how this is completely in contradistinc¬
tion to standard computer programs. In a normal program, you can account
for every single operation at the bit level, by looking “upwards” toward the
top-level program. You can trace a high-level function call downwards: It
calls subroutines that call other subroutines that call this particular machine-
language routine that uses these words and in which this particular bit lies.
So there is a high-level, global reason why this particular bit is being manipu¬
lated.
By contrast, in an ant colony, a particular ant’s foray does not achieve
some global purpose. It has no interpretation in terms of the overall colony’s
goals; only when many such actions are considered at once does their statis¬
tical quality then emerge as purposeful, or interpretable. Ant actions are not
the translation into machine language of some “colony-level” program. No
one ant is essential; even large numbers of ants are dispensable. All that
matters are the statistics: Thanks to them the information moves around at a
level far above that of the ants; ditto for neural firings in brains. Not ditto for
most current AI programs’ architecture.
AI researchers started out thinking that they could reproduce all of cogni¬
tion through a 100 per cent top-down approach: functions calling subfunc¬
tions calling subsubfunctions, and so forth, until it all bottomed out in some
primitives. Thus, intelligence was thought to be hierarchically decompos¬
able, with cognition at the top driving subcognition at the bottom. There
were some successes and some difficulties—difficulties particularly in the
realm of perception. 1 hen along came such things as production systems and
pattern-directed inference. Here, some bottom-up processing was allowed
Al: SUBCOGNITION AS COMPUTATION 285

to occur within essentially still a top-down context. Gradually, the trend has
been shifting, but there still is a large element of top-down quality in AI.
It is my belief that until AI has been stood on its head and is 100 per cent
bottom-up, it won’t achieve the same level or type of intelligence as humans
have. To be sure, when that kind of architecture exists, there will still be
high-level, global, cognitive events—but they will be epiphenomenal, like
those in a brain. They will not in themselves be computational. Rather, they
will be constituted out of, and driven by, many many smaller computational
events rather than the reverse. In other words, subcognition at the bottom
will drive cognition at the top. And, perhaps most importantly, the activities
that take place at that cognitive top level will neither have been written nor
anticipated by any programmer. [Hofstadter, 1982/?; 1982d.]
Let me then close with a return to the comment of Simon’s: “Nothing
below 100 milliseconds is of interest in the study of cognition.’’ [Simon,
1981.] I cannot imagine a remark about AI with which I could more vehe¬
mently disagree. Simon seems to be most concerned with having programs
that can imitate chains of serial actions that come from verbal protocols of
various experimental subjects. Perhaps, in some domains, even in some
relatively complex and technical ones, people have come up with programs
that can do this. But what about the simpler, noncognitive acts that in reality
form the substrate for those cognitive acts? Whose program carries those out?
At present, no one’s. Why is this?
It is because AI people have, in general, tended to cling to a notion that in
some sense, thoughts obey formal rules at the thought level, just as George
Boole believed that “the laws of thought” amounted to formal rules for
manipulating propositions. [Boole, 1854.] I believe that this Boolean dream
is at the root of the slogan “cognition as computation”—and I believe it will
turn out to be revealed for what it is: an elegant chimera.
 ENDNOTES TO THE PAPERS
ON ARTIFICIAL INTELLIGENCE

Allen Newell

The four papers in this volume devoted directly to artificial intelligence (AI)
are all quite independent of each other, having been generated by a process
that did not quite live up to the ideal envisioned by the editors, namely, a
main paper plus commentaries. Thus, there is no organic reason for a re¬
sponse. Nevertheless, in the interests of good form, I will make a few com¬
ments. Since structurally they cannot be footnotes, let them be endnotes.
They are just what strikes me as worth noting about the four pieces.

THE NEWELL PAPER

Relative to my own contribution, I find the other papers useful as a reminder

of additional intellectual issues that should have been considered. Boden’s
paper contains an emphasis on Humanizing versus Dehumanizing. Although
this has an echo in the issue Replacing versus Helping Humans, Boden’s
issue is clearly more fundamental and sweeping. Barr quotes Laurence
Miller on the differing attitudes of AI scientists and psychologists about the
difficulty of Constructing versus Testing Psychological Theories. I agree
with the importance of the intellectual issue, and it probably should have
been included in my list. It clusters with another issue in separating psychol¬
ogy from AI, namely, Memory versus Processing, but it is conceptually
quite distinct. However, Miller’s emphasis on the issue as the major expla¬
nation for the differing evaluations of the contribution of AI would not be
supported by the substantially more complex situation that my own paper
documents. [L. Miller, 1978.] Barr’s quotation from Minsky and Papert
complaining about the nontechnical status of words such as parallel and

This research was sponsored in part by the Defense Advanced Research Projects Agency
(DOD), ARPA Order No. 3597, monitored by the Air Force Avionics Laboratory under Con¬
tract F33615-78-C-1551. The views and conclusions contained in the paper are those of the
author and should not be interpreted as representing the official policies, either expressed or
implied, of the Defense Advanced Research Projects Agency or the United States government.

287
288 ALLEN NEWELL

analog, not only fits in with the overall picture I drew of the nature of
intellectual issues, but reinforces the existence of several particular issues.
Such quotations are data points for an objective determination of what intel¬
lectual issues existed for AI over time. Finally, Hofstadter’s paper makes
use of an issue of Unitary Objects versus Statistical Aggregates (probably
not quite the right moniker), which coordinates with other issues I did in¬
clude, such as Serial versus Parallel, but is not the same. Hofstadter’s paper
also exemplifies discourse at the level of intellectual issues, but that will
come up in my notes on his paper.

THE BODEN PAPER

Margaret Boden has provided another essay that continues to establish her
as one of the most knowledgeable and sympathetic critical observers of AI.
The themes she plays out here are ones she has been concerned with for
some time, and their development can be followed in her recently published
collection. [Boden, 1981a.] Especially worthy of attention is her concern
over the issue of whether AI is humanizing or dehumanizing. She argues
clearly for the possibility that AI is humanizing, against the cultural stereo¬
type that equates mechanization with dehumanization. As already noted,
this is an intellectual issue, as that term is used in my paper. Her discussion
reveals an important aspect of behavior with respect to intellectual issues,
namely, reasoning with generalized concepts. Thus, computers are mechan¬
ical, mechanical means nonhuman, therefore computers are dehumanizing,
which makes it difficult for considerations of a different flavor to get a
hearing, such as those that Boden introduces—active intelligence and the
potentialities of computer representation for exploring the nature of subjec¬
tivity. Of course, little progress would be made if one instance of slogan
reasoning simply replaced another, for example, computers have internal
representations, internal representations permit subjectivity, the subjective
is part of the humanistic tradition, therefore computers are humanizing. We
need to find ways to convert discussion at the level of intellectual issues into
scientific studies of the effects (and potential effects) of our artifacts on our
culture.
In this regard, the comments of Boden about the potential relevance of AI
to social psychology need to be underlined. In my own view, the best bet to
unravel the skein of substantive issues about whether complex artifacts,
such as the computer, are humanizing or dehumanizing lies in understanding
in detail social perception and interpretation and the processing that attends
them. Actually, there exists an active, substantial subpart of social psychol¬
ogy called social cognition, whose expressed intent is to bring over into
social psychology modern cognitive theories and results. [Clark and Fiske,
1982; Nisbett and Ross, 1980.] Cognitive psychology is not AI, and this
group is not directly dedicated to bringing over the results of AI into social
psychology. However, AI is one of the sources that created modern cogni-
ENDNOTES TO THE PAPERS ON Al 289

tive theory. I for one would expect the effect of AI on social psychology to
occur largely through such an indirect path.
I seem to be arriving at a view of intellectual issues as having substantial
ideological content—that one of their dysfunctional aspects is that they
permit discourse to stay focused on highly general concepts with ill-defined
semantics. Boden’s comment on another issue reinforces this conclusion,
namely, the psychological relevance of attempts to match experimental re¬
sults against theories expressed in programmed form, which she labels as
debatable. She identifies the basis for the debate to be the uncertain bound¬
ary between what aspects of the program are intendedly psychologically
relevant and what aspects are not. Boden has also made this point in her
early writings about the field. [Boden, 1972.] Indeed, concern with this
problem of the boundary was rampant in the 1960s. [Reitman, 1965, chap. 1.]
My point is that this issue has both ideological and scientific components.
It participates in an intellectual issue about the possible forms for theories of
human behavior—programs versus mathematics versus no-formalism. As an
indicator of its ideological character, the issue of surplus meaning of all
scientific theories has long been noted both negatively (unjustified incorpo¬
ration) and positively (novel implications that prove out). With respect to the
latter, theoretical physicists positively rhapsodize on letting the elegance of
the mathematical formulation lead the way. Yet, the aspect of this perennial
problem that is always brought to bear on AI programs as theories is the
negative side, as if this were a special feature of these AI theories. A little
later in her essay, Boden notes this tendency to apply to AI with a special
intensity concerns that afflict all science. It would seem that this phenome¬
non is not peculiar to AI, but is an aspect of intellectual issues generally.
There are also scientific issues involved here, and genuine advance is
possible in resolving them. In fact, this occurred in the mid-1970s. [Newell,
1973a; Anderson, 1976; Pylyshyn, 1978.] Ideally, an information-processing
theory posits a cognitive architecture for the subject; this is psychologically
significant, just like the posits of any theory. An architecture provides sym¬
bolic structures for specifying behavior—called programs, plans, behavior
schemas, methods, or whatever. Such structures are psychologically
significant even in their details, because the claim that a given architecture is
operative entails the twin claims that (1) the person’s behavior is specified by
such structures and (2) any such structures that can be created by the pro¬
cesses of the architecture can govern behavior. Such a technical advance
provides a principled way of stating exactly the theoretical significance of
the parts of a program proposed as a theory. However, as we saw before, it
will hardly settle the total issue, which is fueled by a larger set of concerns.

THE BARR PAPER

Barr attempts what I would call an internal mainstream description of AI—

one consonant, say (though naturally enough), with the Handbook of
290 ALLEN NEWELL

Artificial Intelligence. [Barr and Feigenbaum, 1981.] The picture comes

through fairly well. However, I think he has some problems that stem from
AI not being isolated enough for such an internalist description to be stable.
Barr keeps having to describe AI in relation to other fields, and these de¬
scriptions often come through as incomplete.
One place where Barr offers a unique perspective is in respect to the
growing emphasis in AI on applications. I believe he is right when he argues
the importance of the current development of AI applications to all of AI,
although I may locate their importance at a different place than Barr. He
seems to place the emphasis on particular system characteristics (e.g., his
list of five in the section on The Status of Artificial Intelligence); I see more
diffuse but perhaps pervasive effects.
Let me sketch one potential effect of particular interest from the perspec¬
tive of this volume. This is an enhanced development of an internal history
of AI that is less influenced intellectually by other fields than has been the
case heretofore. It is not that AI has failed so far to develop a strong identity.
Quite the contrary, the possession of the computer as a unique powerful
methodological tool and source of theoretical concepts has produced a field
brash enough to make its neighboring fields uncomfortable and a bit sniffy on
occasion. But coincident with this have been strong couplings with these
other fields—with the rest of computer science, psychology, logic, and lin¬
guistics. Each of these relations shows a different pattern of symbiotic,
competitive, and parasitic elements. But they all characterize a field that has
not been left to itself intellectually. For instance, in the 1960s, a substantial
segment of the attention of AI researchers was diverted for several years to
the development of time-sharing systems, and an analogous shift is now
occurring with respect to personal computers and VLSI (very-large-scale
integration) design. Without the strong coupling of AI and computer science,
this sort of switch could not happen. Its effect is to keep AI’s history inter¬
mingled with that of mainstream computer science. This intermingling is
pervasive, and the evidence for it can be found throughout the Handbook as
well as in Barr’s paper.
However, AI application systems are of AI’s own making. They provide a
locus for advances made within AI to fold back into new advances within AI,
without the involvement of other fields. The interest here is not whether
such a development would be a good thing or not, or whose norms should be
used to measure it, but how it would affect the character of AI. Because this
internal history would grow in an application milieu, the view of AI as an
engineering discipline would grow as well. Its bonds with computer science
would strengthen; its interest in the primeval problem of the nature of mind
would wane. This would probably strengthen cognitive science as a disci¬
pline rather than an interdiscipline, since some people in AI with strong
concerns about mind and human intelligence would find the application-
orientation too uncongenial and would shift their energies to cognitive sci¬
ence.
ENDNOTES TO THE PAPERS ON Al 291

Alternative futures are possible, of course, even given the continued

growth of a robust AI applications field. One is that the interdisciplinary
mishmash, so admirably exemplified by this volume, will simply continue to
expand and intermix. All attempts of manageable disciplines to congeal will
fail, destroyed by the continued emergence of highly significant research
results from remote regions. We might be tempted to call the whole thing a
superdiscipline; however, it will be a most untidy affair. Most scientists will
feel inadequate most of the time, because of their inability to maintain any
reasonably comprehensive view or acquire and maintain all the tools they
need. Such a state of affairs would hardly be unprecedented. For instance, it
occurs whenever fields are invaded from without by radically new, highly
technical tools. This happened in the mathematization of economics and
psychology, to mention just two that I happen to have witnessed. However,
these were transients. A permanent state of conceptual disarray and instabil¬
ity would be a novel experience for science.
The happiest alternative, of course, would be the development of a unified
theory of information-processing that would be applicable to everyone’s
intellectual concerns—in fact and not just in theory (sic). Although many
hope for such a thing (including myself), the sprawling character of the
intellectual terrain will surely make such a development tortuous and long in
coming, even after the essential elements of the successful synthesis have
surfaced.

THE HOFSTADTER PAPER

It is now necessary to turn to the paper by Douglas Hofstadter. It is not

evident to me why Hofstadter chose this forum to launch a substantive
argument, as opposed to commenting on the field’s structure. In fact, the
somewhat polemical and diffuse character of the paper makes it a little
difficult to ascertain all his intentions.
One of the most elegant and gracious institutional mechanisms of science
is the principle of public silence. All scientists are obligated to present their
work in public forum before their peers, but there is no corresponding obli¬
gation to respond. If you use a fellow scientist’s work, then you must ac¬
knowledge him or her appropriately. But if his or her work is irrelevant to
your own progress, then it may simply be ignored—met with silence. Fur¬
ther, the criterion of relevance is not simply that the work speak to the same
topic but that it contribute to the actual body of science involved. Thus is
prevented endless and socially destructive rounds of evaluation and disputa¬
tion. What is valuable is winnowed from the chaff; what fails to attract
continued attention falls by the wayside and is forgotten with a minimum of
pain and fuss. A scientist may feel ignored, but energy is devoted to making
progress rather than to criticism, and civility is much increased.
The present forum has a structure that does not permit the public silence
 292
ALLEN NEWELL

that is my instinctive reaction as a scientist to Hofstadter’s paper. In the

jargon ot AI, a slot is provided in which comments on each paper are ex¬
pected. This expectation means that remaining silent receives a special inter¬
pretation, though an ambiguous one. Thus, I make a few comments about
the paper to meet my obligations to the reader and the editors.
Hofstadter s paper shows the difficulty with attempting to operate at the
level of intellectual issues. The issue-complex he discusses is clearly
identified in my “Intellectual Issues” paper by the cluster made up of Sym¬
bolic versus Continuous Systems, Problem-Solving versus Recognition,
Psychology versus Neuroscience, Performance versus Learning, and Serial
versus Parallel. Elements of all of these issues can be found in Hofstadter’s
paper, with an emphasis on recognition and parallelism and the addition, as
previously noted, of a concern with what might be called a statistical
mechanics of the mind.
What is striking in Hofstadter’s paper—and what seems to me typical of
discussions centered on intellectual issues—is the abundance of strong opin¬
ion and argumentation from general conceptual considerations and the ab¬
sence of concrete scientific data or theory to build on. There is an abundance
of attacks on the general opinions of others, with a corresponding promotion
of the general opinions of self. There is strong concern with what other
people believe and assert, and weak concern with what is true of the
scientific phenomenon under discussion, independent of who said it. In
short, there is disputation.
A scientific discussion of the relative roles of mechanisms with time con¬
stants less than rather than more than 100 milliseconds would focus on what
functions are accomplished by mechanisms at the different durations, how
to characterize them theoretically, and what is the evidence for them. The
scientific discussion would focus on formulating a question about the rela¬
tive roles that would add something to our scientific characterization of the
mind and its operation. It would not focus in any serious way on what other
scientists said, especially in their summary or epigrammatic statements.
What counts is what the theory and the data say and why they say it, not
what scientists say.
Theories and data are surrounded at all times by a cloud of commentary_
that is how we all struggle to communicate and understand the substance of
the science we are generating. Intellectual issues seem to be one of the ways
of generating this cloud of commentary that accompanies scientific work.
Hofstadter, as any other scientist, is free to use all the commentary he wants
in his paper, which certainly includes using other scientist’s commentary as
agreement or foil. This will help to orient us to his work and let us under-
stand it. The problem with the present paper is that, in a poor paraphrase of
Gertrude Stein, there is no work there that works. Less epigrammatically,
Hotstadter s paper is a (tiny) example of the scientific process gone awry, in
which (on the evidence of the text) a paper was written to accomplish sci¬
ence entirely by means of commentary.
ENDNOTES TO THE PAPERS ON Al 293

I cannot leave Hofstadter’s paper without warning the reader about its
level of scholarship. I will only select one instance, which is of some impor¬
tance both to me and the argument in his paper; numerous additional exam¬
ples, both large and small, occur in the text. Hofstadter devotes a section,
Passive Symbols and Formal Rules, to examining the position of Simon and
myself on the nature of symbols and contrasting this position in stark terms
with his own view (the whole contrast being cast in a peculiarly personal
idiom). Since his paper is available in this same volume, it is not necessary to
quote extensively. However, a quote from the opening paragraph of that
section will help to convey the content of the section:

We can see that to Simon and Newell, a symbol seems to be any token, any
character inside a computer that has an ASCII code (a standard but arbitrarily
assigned sequence of seven bits). To me, by contrast, symbol connotes some¬
thing with representational power. To them (if I am not mistaken), it would be
fine to call a bit (inside a computer) or a neuron-firing a symbol; however, I
cannot feel comfortable with that usage of the term. (Hofstadter in his paper in
this volume.)

Hofstadter is indeed mistaken, absolutely and unequivocally. My concern

here is that the reader not take seriously any of Hofstadter’s characteriza¬
tions of our position. Simon and I have written extensively about the nature
of symbols, and many of these writings are well known. [Newell and Simon,
1972; Newell and Simon, 1976a; Newell, 19806.] Here are quotations from
the latter, entitled “Physical Symbol Systems,” in which I devoted the
entire paper to going on record as clearly as I knew how as to our (AI
researchers’) understanding of the nature of symbols, circa 1980:

The most fundamental concept for a symbol is that which gives symbols their
symbolic character, i.e., which lets them stand for some entity. We call this
concept designation, though we might have used any of several other terms,
e.g., reference, denotation, naming, standing for, aboutness, or even symboli¬
zation or meaning, (p. 156.)

Representation is simply another term to refer to a structure that designates

. . . . (p. 176.)

The usual formulations of universal machines also tend to use the term symbol
for the alphabet of distinctive patterns that can occur in the memory medium
(e.g., the 0 and 1 tape symbols for our Turing machine). As defined, these
entities are not symbols in the sense of our symbol system. They satisfy only
part of the requirements for a symbol, namely being the tokens in expressions.
It is of course possible to give them full symbolic character by programming an
accessing mechanism that gets from them to some data structure, (p. 158.)

These quotations are simply the parts of the paper that address directly
and unequivocally the characterization that Hofstadter’s quotation presents.
294 ALLEN NEWELL

The rest of my paper on symbol systems deals at length with the positive
characterization of how symbols represent in general, and how this can
happen by means of finite mechanisms. The treatment is entirely consistent
with Simon’s and my other writings on symbols. The point here is not
whether this theory is known to be right, only that the description of it in
Hofstadter’s paper bears little relation to its actual form.

A FINAL WORD

The only final remark I would make about the collection of papers is that the
historical pictures they all draw of the field seem roughly consonant (though
their evaluations are not necessarily so). To first order, they seem to be
talking about the same history. This is true also of the paper by Pylyshyn
about cognitive science. [Pylyshyn in this volume.] However, I have not yet
seen the other papers in the volume. It will be interesting to see the extent to
which quite different views exist of the total historical framework and to
discover the sources of such differences.
SECTION 4
 LINGUISTICS AND ITS
RELATIONS TO
OTHER DISCIPLINES

Thomas G. Bever

Language is central to our conception of what it means to be human. Lin¬

guistics, the science of language, has been an influential discipline for many
years. As speakers, we have strong intuitions about what a satisfactory
linguistic theory must include. Linguistic theories have presented precise
versions of different scientific approaches to the description of all behavior
and knowledge. The dominant philosophy of behavioral science between
1920 and 1960 was an extreme form of operationalism, behaviorism. Behav¬
iorism required that all abstract terms of a theory be resolvable into observ¬
able stimuli or responses. Thirty years of linguistic investigation clarified the
limits of such a framework. In the late 1950s, a leading researcher on animal
behavior, B. F. Skinner, published a definitive behaviorist treatment of the
psychology of language. [Skinner, 1957.] His book was reviewed by a lin¬
guist, Noam Chomsky, who summarized the weaknesses of the behaviorist
explanation of grammatical structures. [Chomsky, 1959.] The main point
was that behaviorist restrictions on theoretical terms make it impossible to
describe structural facts about language. Chomsky argued that abstractness
of theoretical devices is not only normal in science, it allows for greater
descriptive adequacy in the case of language. This review, and the concomi¬
tant development of nonbehaviorist grammars for describing sentences, sig¬
naled a revolution in linguistics that has had broad implications for other
behavioral and social sciences. The overall effect of discoveries in linguistics
has been to raise the possibility that the evolution and contemporary nature
of symbolic behavior involves highly specific mechanisms of learning and
representation.

LINGUISTICS—THE SCIENCE OF LANGUAGE

Language is a symbolic structure that humans use as a means of communica¬

tive and personal representation of the world. Two central aspects of this
structure carry meaning: reference and propositional relations.
297
298 THOMAS G. BEVER

Reference

Words refer to objects in the world or mind. Prima facie, such reference is an
obvious and straightforward kind of relation between a symbol and the
world; dog refers to four-footed canines, bachelor to unmarried human
males, and so on. Consideration of the symbol/referent relation has been a
central concern of the philosophy of language, since it bears directly on how
humans represent and know the world around them. The relation between
reference and meaning is not so simple as it seems. We cannot define the
symbolic meanings of words as the sum of their referents, since that would
incorrectly imply that a trivial change in the referent (e.g., in the number of
dogs in the world) would change the meaning of a word. Furthermore,
certain words have no external reference, for example, unicorn, and certain
words seem to be defined only in terms of their effects on other words, for
example, not, did. [Frege, 1879; Katz, 1971.]
Such facts demonstrate that knowing a language involves something other
than simply knowing a large number of symbol/referent associations. This is
also attested by the fact that many animals can master many such associa¬
tions: A skilled shepherd dog can distinguish a large number of different
commands involved in moving a herd of sheep. Yet, we do not conclude that
such a dog has mastered a language.

Propositional Relations

Languages involve the relation of individual words in meaningful proposi¬

tions. For example, in the sentence dogs bark, there is a relation between
dogs and bark that transcends the separate references that each word might
have. This relation is that between an agent of an action and the action itself.
Similarly, in dogs bite cats, the third word is the object of the action; in dogs
often bite cats, the second word specifies an aspect of the action. In general,
words are set in such relations to each other within meaningful propositions
that apply to entire events or states in the world. That is, utterances have
meanings that are independent of the reference of the individual words.

The Generative Model of Linguistic Structure

Sentences are the smallest units of language that present well-formed propo¬
sitions that can relate meanings and references. For this reason, current
linguistic investigations focus on the sentence as the major unit of investiga¬
tion. Within a particular language, sentences have distinct structural proper¬
ties. For example, in English, the dog chased the cat means something
different from the cat chased the dog but is very similar in meaning to the cat
was chased by the dog.
The goal of linguistic research is, in large part, to discover the model that
describes sentences and their interrelations. In this essay, I present an out-
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 299

line of the model as it is widely accepted, though it must be emphasized that

its status and details are always subject to change in interpretation. Three
features of linguistic descriptions have remained invariant over many de¬
cades. First, languages are characterized as having a number of levels of
representation. Each level is delimited and defined by a particular set of
units and relations. The levels are related by mapping rules that express
well-formed configurational relations between particular structures at each
level. [For a general review of the recent history of linguistics, see New-
meyer, 1980.]
A simple example from the English sound system clarifies these notions.
Sound sequences are analyzed at two primary levels—an internal array of
schematic units and an external string of actual acoustic/articulatory units.
Consider the word bitter. When the word is pronounced slowly, we can
isolate the individual sounds that correspond to the internal array, consisting
of lbl,lil,ltl,lerl. This differentiates the word at both levels from other words,
such as bidder {/blJit,/dlJerl). If the words are pronounced quickly, they
begin to converge onto the same acoustic sequence. In particular, the middle
consonants Itl and /d/ can be pronounced identically (as a flap of the tongue
against the roof of the mouth). Yet, even when rapidly spoken, the two
words can still be correctly differentiated. Close analysis shows that one
acoustic difference between the two rapidly spoken words is in the length of
the vowel that precedes the middle consonant—when the consonantal sound
corresponds to Idl the preceding vowel is longer than when the consonantal
sound corresponds to Itl. That is, we hear a difference between two conso¬
nants, which is acoustically signaled by the length of the preceding vowel.
This simple fact reveals a great deal about what we know when we know
English sound regularities; this knowledge is schematized below. Words
have an inner level of representation of sound sequences and a set of rules
that map them onto an outer level representing the actual acoustic se¬
quences:

Inner sound sequences for lexical items: for example, B I D R; B I T R.

Rules that map inner sequences onto acoustic/articulatory sequences, for
example: (1) Vowels are lengthened before voiced consonants; and (2) in
rapid speech, voiced and unvoiced consonants between vowels are spo¬
ken without voicing.

Rule 1 explains the difference in vowel length between words like hit and
hid, or back and bag. It also explains why the first vowel in bidder is longer
than that of bitter. Rule 2 explains why the second consonant in those two
words sounds alike when they are pronounced rapidly. The ordering of rules
1 and 2 explains why the first vowel in rapidly spoken bidder is longer than
that of bitter, even though the two middle consonants are identical after rule
2 applies to them. The logical ordering of the rules guarantees that the vowel
300 THOMAS G. BEVER

preceding Idl will be lengthened, before the Idl loses its voicing. [Chomsky
and Halle, 1968.]

Levels of Representation

This example displays a typical mapping relation between two levels of

representation in language. Each level captures certain kinds of facts about
our knowledge. The inner sound level represents the categorical knowledge
of words, expressed in terms of sound schemata; the outer level represents
how the words are actually pronounced, expressed in sequences of acoustic/
articulatory units. Mapping rules represent relations that the language allows
between these two levels.
The interrelation of distinct levels of representation is also part of the
description of sentences. The units at each level are the constituent phrases,
groupings of adjacent words and symbols into syntactically relevant units.
For example, in the sentence the dog often chased the cat, the constituent
phrases are the dog, often chased, the cat. Such phrases are postulated as
components of the outer level of sentence representation, in part because
they represent the natural units of phrasing in normal pronunciation. (Try
pronouncing the sentence with pauses between the phrases and contrast that
with pausing after every odd word.) They also represent the kinds of units
out of which the surface sequence is composed.
Phrase constituents are more abstract than lexical items but can serve the
same function as different kinds of lexical items. In the example sentence,
the dog and the cat are classified as nounphrases, since the main content
word of each is a noun; often chased is a verbphrase for the corresponding
reason. This kind of designation allows the description of structural similar¬
ity among sentences that differ in other superficial properties. The preceding
example has the same structure of the major surface constituents as both of
the following: dogs, chase, cats', the domestic canine, with considerable
frequency chases, the domestic feline. Each of these two sentences consists
of a nounphrase, a verbphrase, and a nounphrase. Variation in length
reflects the fact that constituents can contain constituents within them¬
selves. Certain constituents can contain constituents of the same type: This
makes those constituents recursive. For example, one type of nounphrase
can itself be an entire proposition, underscored in Harry believes that Bill is
a fool. Such recursion can repeat, for example, allowing Harry believes that
Sam knows that Bill is a fool. The creative component of sentence structure
is represented in the way in which phrase constituents can continually be
elaborated: Sentences of arbitrary length and complexity can be represented
as results of such self-embedding constituents.
Constituents are most salient in the surface structure of sentences, the
level that presents their actual order and arrangement. Sentences are also
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 301

described at inner levels of representation, where the arrangement of con¬

stituents reflects more abstract properties. For example, in the dog often
chased the cat, the first phrase is defined by its structure and position as the
grammatical subject and the last, as the grammatical object.
The differentiation of syntactic levels allows the description of various
surface arrangements of constituents that correspond to the same inner ar¬
rangement. For example, the cat chased the dog often has the same abstract
relations between constituents as the preceding case, as does often the dog
chased the cat. In each case, often is arranged as part of the verbphrase in
the inner structural level, while appearing in different positions in the outer
structure. As in the sound system, the inner structure is mapped onto the
outer structure by a set of rules. The particular orders and arrangements that
are allowed in a language are the result of several sets of such rules. For
example, transformations are rules that arrange an inner structure into a
particular surface order. Such rules can rearrange the position of a verb
modifier like often to appear in a position not adjacent to the verb. The
passive transformation arranges the grammatical subject to precede the verb
and the grammatical object to follow it, while introducing other words.
When first proposed, these mapping transformations were one of the main
objects of linguistic investigation. The goal was to determine the set of
transformations for each language and the universal laws that govern how
they operate to map the inner syntactic level onto the outer syntactic level.
It is important to note that the inner forms are all abstract—that is, they do
not correspond to some actual sentence. This is tantamount to the claim that
when speakers know a language, they know an abstract property for which
they are given no specific example.
As it stands, the model assigns an inner and outer structure to a sentence
but does not provide the meaning. The model is an account of structurally
well-formed sentences regardless of what they mean (if anything). The
model makes it possible to specify exactly what syntactic information con¬
tributes to the meaning of a sentence. An initial proposal was that the mean¬
ing of a sentence is determined entirely by its inner structure. This captured
the intuition that surface variations in constituent order (as between the
active and the passive) did not vary propositional relations between con¬
stituents.

The Development of the Model

Subsequent developments in the model have turned on two questions: What

are the levels that contribute to meaning? What is the relative descriptive
importance of transformations and constraints on how they apply ? The rela¬
tion of the model to meaning has been interpreted in two distinct ways by
different theories. In one case, generative semantics, it was proposed that
302 THOMAS G. BEVER

the inner structure of a sentence is the meaning without any further interpre¬
tation. This set up an appealing model in which the inner structure of a
sentence is its meaning, and transformations map it directly onto actual
sequences. [Lakoff and Ross, 1976.] There were a number of difficulties with
this model; most notable was the fact that, as the notion of meaning was
broadened, the corresponding range of possible inner structures had to be
enriched. If two sentences have the same meaning, then on this model it
must be represented by their having an identical inner structure. Simple
examples demonstrate how unwieldy this proposal is. Consider the sen¬
tences “John is an unmarried eligible human male” and “John is a
bachelor.” They clearly have the same meaning, yet construction of a trans¬
formation to relate them involves great complexity.
A further difficulty with the unification of inner structure and meaning
was that certain aspects of meaning appeared to be contributed by surface
structure. For example, different interpretations of one boy ate each jelly
bean and each jelly bean was eaten by one boy are reflected directly in their
surface order. Such examples motivated an expansion of what information
contributes to meaning to include surface structure as well as inner struc¬
ture.
This change occurred at roughly the same time that it was noticed that
outer and inner constituent arrangements are limited to the same con¬
figurations in any given language. [Emonds, 1970.] That is, even though
the possible transformations that apply to inner structures could conceivably
produce many varieties of surface arrangements, most of this power is never
actually used within a language. This motivated the generalization that every
surface structure must be potentially generated by some arrangement of
inner structure constituents. Transformations became correspondingly
simplified to the point at which there is only one transformation, which maps
surface structures from underlying structures, namely, the rule move-X
(where X is either an entire clause or a nounphrase). Conditions on possible
transformations have been enriched, so that the possible surface
configurations are restricted in the right way.
With such developments, it is harder to identify a unique syntactic level
as the primary abstract structure. Viewed intuitively, the first version of the
syntactic model generated surface sequences by applying various transfor¬
mations to the inner syntactic structure; in general, the resulting sequences
were always possible sentences. In its current manifestation, the model
describes many surface sequences that do not correspond to sentences:
They are filtered out by conditions of various kinds, ranging from limitations
on particular surface properties to conditions on possible relations between
surface and inner structures. In this way, the focus of linguistic investigation
has shifted from discovering the transformations to discovering the condi¬
tions of well-formed structures at each level. An important hypothesis (not
yet conclusively proven) is that the conditions are the same for every level of
representation. [Chomsky, 1981.]
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 303

The Interpretation of Such Models and Facts They Explain

The essential goal of syntactic description is to generate descriptions for the

well-formed sentence in a language. The syntactic model provides for every
sentence a description consisting of several levels of representation and
rules interrelating the levels. A sentence is marked as well formed only if it
corresponds to a well-formed description in the model. The postulation of
various levels of representation also allows for the structural differentiation
of types of sentences. In particular, it allows us to make claims about unac¬
ceptability of different kinds. For example, sentences may be unacceptable
because they are meaningless (good is a sugar), unusable {the oyster the
oyster the oyster split split split), or ungrammatical {Harry and Bill didn’t
know that each other was there). Such differences show that there is a
distinction between acceptability of a word sequence and its grammatical
status. Sequences can be acceptable and grammatical, and unacceptable and
ungrammatical. But both of the other combinations exist as well—a se¬
quence can be ungrammatical but acceptable, and the converse. This
clarifies an important empirical interpretation of the grammar: It describes
what sentences are grammatical, not which sequences are acceptable.

LINGUISTICS AND COGNITIVE PSYCHOLOGY

Cognitive psychology is the study of the acquisition and memory of knowl¬

edge and beliefs. There are three main current themes in this field. The first
explores the span of consciousness, the second is concerned with the form
of mental representations, and the third focuses on learning as a special kind
of problem-solving. In each case, linguistics has made specific contributions
by providing a precise and rich formal description of a complex behavior.

Language and the Psychological Present

For several decades, a major concern has been the transformation of infor¬
mation during the first phases of acquiring knowledge. This study involves
the borderlines between processes of attention, perception, and recognition.
When we perceive an object in the world, we use a series of retransforma¬
tions of the stimulus input, starting with the most peripheral sensory repre¬
sentation and moving to the most abstract. We can interpret each stage of
representation as a level of consciousness; the span of attention increases in
their time range and complexity as it shifts from dependence on one kind of
representation to another. Experimental techniques intrude on the ordinary
flow of information, showing which points are under conscious control and
which are not. This provides a picture of consciousness that is somewhat
fragmented. Stable representations of information hop from stage to stage
via brief unconscious periods of transformation.
304 THOMAS G. BEVER

In most cases, the study of this serial process involves information that is
itself of little interest, as the perception and retention of letter groups. Such
stimuli have little internal structure that might yield information on the kind
of transformations that can occur at each stage. A related problem is that the
stimuli do not comprise a natural subset of the world that could call on
psychologically indigenous processes. Language is such a behavior, and
linguistics offers a rich fund of information about possible kinds of represen¬
tations.
The study of how listeners comprehend speech corresponds to the studies
of how information is transformed during immediate encoding. The plausible
view of a retransformation is that listeners first transform the acoustic struc¬
ture into a sequence of separate sounds, which are grouped into words and
then into phrases and sentences and, finally, into meanings.
Careful research on comprehension processes has shown that the succes¬
sive-stage model of encoding is incorrect; rather, the levels of language
representation interact directly during comprehension. We can use our high-
level expectations about meaning to guide our low-level expectations about
sounds. Of course, the points at which such information can be related
between levels depend on the presence of complete representations at the
separate levels. This gives special status to points in a sentence that corre¬
spond to complete units at more than one level of representation; for ex¬
ample, the end of a clause is also typically the end of a meaning proposition
and the end of a constituent. Accordingly, numerous studies show that at
such points listeners integrate the meaning of a sentence with other aspects
of it and its contextual role.
Most important is the fact that sentence comprehension clarifies certain
aspects of consciousness. First, we seem to understand sentences directly as
we hear them, even though our understanding itself modifies the sounds that
we perceive. Various experiments demonstrate that an enormous amount of
computational activity occurs during sentence comprehension, but we are
conscious only of the unified output—the pairing of a sound sequence with a
meaning. This simple fact tells us a great deal about the relation between
separable unconscious processes and their unification within consciousness.
In the case of sentences, the unity of conscious experience can even be
observed in contrast with the duration of the sentence itself. Consider the
sentence you are reading right now. When did it begin as a conscious experi¬
ence as a sentence? After the first word? The second? In fact, it appears to
have begun at the beginning and to have ended exactly at the end. In this
sense, sentences are impenetrable units of conscious experience. Since they
can be quite long, and still be perceived as sentences, one consequence of
language is that it expands the range of our momentary conscious experi¬
ence. Rather than being limited to six items, a sentence expands our momen¬
tary span to 20 words, or 100 sounds! [G. A. Miller, 1956.]
It does so, of course, by utilizing the unconscious mechanisms for rapid
calculation of structural organization. This allows us to use language as a
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 305

case study in the role of structures in expanding the domains of conscious

activity.

The Relation between Knowledge and Behavior

A second problem of cognition has been the relation of knowledge to behav¬

ior. For example, humans know many things to be true about logic and
integers but make characteristic kinds of mistakes in everyday reasoning and
mathematics. The systematic nature of such mistakes reveals specific mech¬
anisms for implementing knowledge. This is reflected in three experimental
domains of cognitive psychology: memory, perception, and behavior-
production. In each case, the development of generative linguistic theory
has stimulated new research and a general controversy over how directly the
grammar of a language is involved in ongoing behavior.
This controversy first expressed itself in the study of human memory for
sentences. The accepted view of memory in general is that it uses canonical
schemata, such as a typical face or a typical bird, as basic representational
schemata. Individual experiences are retained in terms of such schemata and
variations on them. Linguistic research stimulated considerable investiga¬
tion of sentence memory by providing a number of levels of representation
at which such canonical schemata can be stated. Shortly after the emergence
of generative grammar, it was natural to investigate the possibility that it is
the inner structures of sentences that we retain when we remember them. If
true, this would offer a precise theory of the canonical memory structures,
and a theory of how they are mapped onto external forms. It was exciting to
think that formal linguistic analysis could provide results that would be
immediately integrated with psychological investigations of a traditional
problem such as memory. This interaction between linguistics and psychol¬
ogy became the cornerstone of a new interdisciplinary field, psycholinguis¬
tics. [G. A. Miller, 1962.]
The initial results were encouraging: It appeared to be the case that sen¬
tences that are more complex in their grammatical description are harder to
remember. For example, a passive construction involves more transforma¬
tions than an active construction and is, indeed, harder to recall. Such facts,
however, turned out to be due to other factors, such as the relative infre¬
quency of the construction and the fact that it presents the major phrases out
of the typical subject-verb-object order. When meanings and order are care¬
fully contolled, transformational complexity does not predict memory
difficulty. [Fodor and Garrett, 1967.]
There are two ways to resolve such a discrepancy between a formal
theory of behavior and direct behavioral evidence. We can argue that the
link between the formal theory and the behavior is extremely abstract and
that specific behavioral facts might not directly reveal the formal theory.
Alternatively, we can argue that behavioral disconfirmations actually
disconfirm the linguistic theory itself: If the behavioral predictions are
306 THOMAS G. BEVER

wrong, then the formal theory is wrong. Most scholars took the first course,
assuming that the failure of behavioral predictions based on transformational
analyses was due to the lack of a theory of how the formal theory is utilized
in behavior. This interpretation presumed a distinction between competence
(the formal theory) and performance (behavior). Such a distinction allowed a
retrenchment of the field of psycholinguistics: Linguistic analyses were in¬
voked as describing the structure of the language but not necessarily the
structure intrinsic to language behavior. The role of linguistics was to pro¬
vide structural analyses of sentences. The role of psychology was to explore
which of the analyses actually play a part in speech behavior. [Bever, 1970;
Fodor et al., 1974.]
The distinction between competence and performance seemed clear in the
case of memory, since there are many ways we can choose to memorize and
many different reasons to do so. Speech comprehension and production,
however, offer fewer choices—we generally have one purpose in com¬
prehension, to understand ideas; and generally one purpose in talking, to
express ideas. We might expect that a grammar would express itself more
directly in such behaviors, because there is less opportunity to impose
behavior-specific strategies. However, both behaviors appear to use heurist¬
ics for interrelating sound and meaning: Heuristics are responsive to certain
structural properties of sentences but to other factors as well.
Consider, for example, the usual arrangement of actor, action, and object.
In English, this is the normal order. Various perceptual studies suggest that
listeners impose such an interpretation on sentences as a preliminary (and
unconscious) hypothesis. Clearly, such a strategy is a powerful simplifying
device but has nothing directly to do with a formal grammatical property:
Most clauses do conform to this pattern, but not all.
Despite the clear evidence that certain aspects of talking and listening
depend on nongrammatical patterns and processes, some linguists have
adopted the position that the failure of a grammar directly to predict all
behavioral facts invalidates that grammar. One proposal, in particular, is
that transformations fail to correspond directly to behavioral data because
language is not transformational. According to this view, individual lexical
items carry sufficient structural information to map directly onto meanings.
[Bresnan, 1978.] It is beyond the scope of this paper to present the linguistic
arguments that might bear on such a question. However, it clarifies what
remains an open issue on the relation between formal grammar and speech
behavior: Which properties of behavior directly reflect grammatical knowl¬
edge and which reflect properties of perception and behavior production.

General and Specific Learning Mechanisms

The final area of interaction between linguistics and cognitive psychology

involves how the formal grammar is learned. Within current cognitive psy¬
chology, the issue involves the specificity of mechanisms that learn and
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 307

process any skill. In one extreme view, all learning is carried out by a
general-purpose processing mechanism that applies equally to the learning
of language and other abilities. In the alternative view, learning is carried out
by a set of innate mental “modules,” each highly specialized for its task; the
mind would be a federation of such distinct modules.
These two views can agree that the adult mind is highly differentiated into
component capacities; the views oppose each other with regard to the nature
of learning. The computational model was the dominant theory in the psy¬
chology of learning for many years: All learning was held to be the associa¬
tion of stimuli and responses—intelligence consisted of the ability to form
large numbers of such associations. This view was an important rationale for
studying learning in animals, since human mechanisms of skill acquisition
were assumed to be the same, only more capacious. A similar computational
view is now held by many researchers in computer science. The computer is
viewed as a minor brain that can represent the essential features of the
child’s mind. Although the processes may be richer than simple association
of stimuli and responses, the basic claim about learning is the same as that of
the traditional stimulus/response learning theorists: Learning results from
the application of general mechanisms to specific experiences. In particular,
language would be learned as the result of general problem-solving
capacities that children (or computers) apply to the language they experi¬
ence. [Minsky, 1968; Simon, 1969 and 1981.]
Modular theorists argue that speech data the child hears are always too
impoverished to account for what is learned. Therefore, children must have
a particular capacity, innately tailored to learn language; otherwise they
could not extrapolate from limited data to the correct complete knowledge of
their language. The general claim of such theorists is that only artificial and
uninteresting skills are truly learned by the application of general intelli¬
gence. [Chomsky, 1980a.]
This has brought back into psychology an old controversy over whether
the mind is a collection of separate faculties or whether it depends on the
deployment of nonspecific intelligence. All parties to the controversy agree
that some highly specific processes are innate and that there is some capac¬
ity for nonspecific learning. The empirical question involves the nature and
relative importance of the two kinds of processes.

LINGUISTICS AND THE BIOLOGY OF COMMUNICATION

The contrast between general and specific bases for language is central to
biological considerations: Is language the result of a linguistically isolated
evolutionary change, or did it become possible as the result of some general
mutation? Evolutionary speculation about a behavior pattern is limited by
lack of direct records. What we can do is reason from contemporary mor¬
phological and behavioral facts. Three biological properties of language are
308 THOMAS G. BEVER

particularly relevant: the morphology of the vocal tract and attempts to

teach animals a language that is visual; the possibility of a maturationally
determined critical period for language acquisition; and the neurological
specialization of the left hemisphere of the brain for language.

Language Without Mouths

One aspect of language that differentiates it from other animal activity is the
richness of its acoustic structure. This depends on an articulatory apparatus
that allows for a moving occlusion (produced by the tongue) that separates
two relatively independent chambers in front of and behind it. An ape’s
mouth, for example, simply cannot differentiate the variety of acoustic reso¬
nances that are characteristic of language: The jaw is too recessed to allow
the tongue to separate two large vocal cavities. Paleontological evidence
suggests that relatively recent hominids still had a recessed jaw, which
would not allow for phonological variety. This would mark the evolution of
language capacity as very recent, perhaps less than 100,000 years.
[Lieberman, 1968.]
It is startling to think of the evolution of language as so recent and so
dependent on a peripheral physiological change. Furthermore, the existence
of sign languages in the deaf suggests that vocal expression is not intrinsi¬
cally necessary for language to emerge. One line of investigation concerns
whether sign language is itself a distinct language or whether it is a visual
subset of spoken languages. The research to date suggests that it is an
independent language with rules and properties of its own that are not de¬
pendent on phonetic-based language. Most interesting is the fact that sign
language also appears to respect the more abstract universals of language,
such as the separation of levels of analysis. [Klima and Bellugi, 1979.]
Of course, so far as we know, sign language emerged after spoken lan¬
guage, and may be historically parasitic on it in certain ways. One direct test
of the phonetic basis for linguistic evolution would be to train an animal to
use signs: The most obvious animal is an ape because of its intelligence and
closeness to human beings. If the training were successful, then we would
have some basis for claiming that the particular articulatory developments in
humans might, indeed, have been solely responsible for linguistic capacity.
Teaching apes to talk has, of course, broader implications for our under¬
standing of information and its transmission. A number of recent studies
have attempted to teach chimpanzees sign language or the use of computer-
controlled abstract symbols. [Gardner and Gardner, 1971; Terrace et al.,
1979; Premack, 1976.] The achievements of the chimps are considerable, but
there is no convincing evidence that they learn language structure. Rather,
like shepherd dogs, they can be taught specific sequences and specific rela¬
tional symbols; the capacity to generalize from such training to sentences in
general appears to be lacking. At the moment, such studies have the ambi¬
guity of experimental failures: All we can be absolutely sure of is that exist-
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 309

ing techniques have failed to teach apes to talk. We can also be reasonably
sure that if some method is found that is more successful, it will involve a
learning environment that is highly contrived. It will be radically unlike the
environment that stimulates language learning in children.

The Critical-Period Hypothesis

Children acquire the language of the community in which they grow up. This
is obvious evidence in favor of the hypothesis that language is learned. The
acquisition of behavior patterns by animals shows similar learning, for ex¬
ample, a gosling acquires a permanent attachment to its own parent. This
apparent learning turns out to be highly constrained: During a specific phase
of growth, the gosling becomes parentally fixated on any object that has
approximately the right size and the right speed of movement. This is a
typical example of a “critical period,” during which specific experiences
must occur. After the period has passed, the ability of the animal to learn is
closed off. [K. Lorenz, 1965.]
One argument for the biological specificity of language would be that it
also has a critical period. The strongest test of this would be to expose
certain children to language at different ages to see if their learning ability
changes. This ethically unacceptable experiment has been simulated in a few
instances in which children brought up in isolation or by animals are discov¬
ered in adolescence. A recent case study is Genie, a girl isolated until age 11.
Her linguistic ability is still minimal after a number of years of intensive
training. This does not prove that there is a critical period for language, since
many of her other social and cognitive skills are abnormal as well. [Fromkin
et al., 1973.]
Nature occasionally performs a different kind of experiment, in which the
brain is injured during childhood. This gives an opportunity to observe the
speed with which language function recovers. If there were a critical period,
we would expect injury after a certain age to involve categorically greater
loss of language. Clinical surveys have found evidence consistent with this;
injury after age twenty is often followed by long-lasting aphasia, while injury
before age ten is not. [Lenneberg, 1967.] The interpretation of such facts,
however, is not unequivocally in support of a critical period for language¬
learning. First, the critical age would seem to extend for ten years, which is
rather unconvincingly critical. Second, recovery of a function by an injured
brain is not the same as original acquisition of that function. It has long been
known that a seriously injured brain can operate by compensatory functions,
not necessarily replacement of damaged structures with identical ones. [K.
Goldstein, 1948.] Accordingly, the loss of plasticity of neural recovery for
speech may reflect a general loss of plasticity. In particular, it has not been
shown that the loss of language-recovery plasticity at age twenty is any
greater than the loss of plasticity for other capacities of similar complexity.
A supplementary approach has been to explore the evidence for a critical
310 THOMAS G. BEVER

period in the ability to learn second languages. If such a critical period

existed, it would be consistent with the view that language-learning is depen¬
dent on specific maturational mechanisms that automatically deteriorate af¬
ter a certain age. Despite the frequent intuition that learning a second lan¬
guage in adulthood is difficult, research on the topic is not unequivocal.
What differences there are might be attributed to a loss of motivation, or of
general memory capacity, as opposed to a specific loss of linguistic capacity.

Cerebral Asymmetries

The left hemisphere of human brains is relatively specialized for language.

This is simply demonstrated by the fact that injury to that hemisphere results
in aphasia, while injury to the right hemisphere rarely does so. Furthermore,
children with only a left hemisphere learn language better than children with
only a right hemisphere (these children having had the other hemisphere
surgically removed). Such a lateral coincidence with language is consistent
with the view that the left hemisphere is biologically predisposed for lan¬
guage. Clinical facts, indeed, suggest that such a predisposition must exist
but leave open the question of how specific to language the predisposition is.
For example, if the left hemisphere is predisposed for all processing that
involves interrelations of parts and wholes, then language might emerge in
that hemisphere, since sentence-processing always involves interrelating
words and phrases with an entire meaning.
A growing body of research is consistent with this view: The left hemi¬
sphere appears to be the more frequent seat of relational activities of a
variety of kinds. Accordingly, its specialization for language does not show
that language is uniquely formed by the hemisphere itself. [Levy, 1972;
Bever, 1980a.]

LINGUISTICS AND PHILOSOPHY

Language is the most salient vehicle for human knowledge and judgment.
Accordingly, it is often treated as central to issues involving the philosophy
of mind. Indeed, certain periods in philosophy have been dominated by the
view that in solving linguistic problems, we can solve fundamental problems
concerning epistemology and truth. Modern linguistics bears on three main
issues: the biological basis of knowledge, reductionism as the standard form
of scientific explanation, and the existence of Platonic a priori structures.

Empiricist and Nativist Theories of Mind

The source of human knowledge is one of the oldest problems of philosophy.

The question is whether we represent the world as it is or whether we
impose our own structure on reality. Empiricism—taken as a theory of
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 311

learning rather than as a method for advancing knowledge—is the extreme

view that categories of experience are derived from experience of the world;
nativism is the alternative view that forms of experience are innately deter¬
mined and actual experience merely provides specific content for the innate
forms. Language provides certain critical examples for this debate.
One issue concerns analytic truths, that is, sentences that are true by
definition. For example, both sentences that follow are true and have always
been so in our experience, but the first could change, while the second could
not without a change in the meaning of the words: (1) Human bachelors are
between the height of 1 foot and 9 feet, and (2) Human bachelors are unmar¬
ried males. The truth of the second sentence is categorical, rooted in the
very meaning of the words. There is no amount of contingent experience
that could account for the child’s discovery of the categorical nature of such
a truth—yet children do discover such distinctions between necessary and
contingent truths. This would seem definitively to disprove the empiricist’s
position, since the distinction must be innate if it is not learned. Empiricists,
however, have countered in a number of ways. The most obvious way is to
deny that the analytic/synthetic distinction really exists and have it reduced
to the distinction between truths that are held with different strengths of
belief: Some are held very strongly, some less so. [Quine, 1953.]
We might reply that the justification of the categorical distinction is not at
issue, since, in any case, we must account for the untutored belief that it
exists. This, however, pushes the problem out of the realm of explaining
knowledge into the behavioral domain of explaining the acquisition of false
and true beliefs. Since false beliefs exist in any case, the claim that the
categorical distinction is false serves as a philosophical coup for empiricists
that moves the problem into psychology.
The acquisition of linguistic structure itself is not so easily dealt with by
empiricists. Children appear to acquire intricate knowledge about distinct
kinds of linguistic facts without receiving explicit training and often without
very many speech data. This shows that the child must contribute a great
deal of structure to what he or she learns. This observation has been turned
into a nativist syllogism by Chomsky and others:

1. To be proven: language is innate.

2. Language has property P.
3. P cannot be learned.
4. Therefore P is innate.
5. Therefore language is innate.

There are numerous properties P that might be invoked. The fact that differ¬
ent kinds of knowledge contribute to the usability of a sentence itself might
be undiscoverable from actual data given. How does the child actually learn
that certain uses of a sentence are acceptable for pragmatic as opposed to
312 THOMAS G. BEVER

structural reasons? It seems unlikely that all children receive explicit train¬
ing in the different ways that sentences can be used. No explicit information
tells a child that sentences can be used ironically to have the opposite of
their ostensible meaning, yet the child acquires this distinction.
Linguists tend to cite more technical properties of linguistic structure as
unlearnable by standard learning theories. For example, the restriction that
transformational rules must apply to the highest level constituent of a given
type is never explicitly presented to the child. Clearly, the child must apply
some pattern-recognition strategies to the data as given and impose on them
the presumption that such a condition exists. But this imposition itself can
draw on only hypotheses internal to the child: Hence, the concept of such a
condition on rules must be an innate hypothesis.
The accumulation of such arguments adds up to the claim that the inter¬
esting formal properties of language are all innate. To paraphrase Chomsky,
there is a “language-learning organ” that can extrapolate a full grammar
from extremely scanty data. Language is not learned but rather grows in the
mind, much as the liver or kidney grows in the body. [Chomsky, 1981.]
This nativist claim, of course, is itself open to empirical investigation:
What children have learned when they use language appropriately is an
object of psychological inquiry. Chomsky, however, puts the nativist claim
in a context that confronts empiricism in its broader sense: He extrapolates
from the relatively narrow claims about language development to broader
claims about all human knowledge. He suggests that there is a general
theory-forming capacity that is itself an innate mental organ.
The clarity of this attack on empiricism has created a rare situation. A
philosophical issue involving the nature of knowledge might actually be
resolved in part by appeal to empirical data. In this case, critical data are the
patterns of language acquisition: If children can be shown to use gradually
some general pattern-learning skills to learn language, then the specific
nativist claim is weakened; conversely, insofar as children can be shown to
extrapolate creatively beyond speech data in ways that express idiosyncratic
formal linguistic universals, the nativist claim is strengthened. At the mo¬
ment, the weight of evidence from language acquisition is in favor of the
nativist position. This, however, may be related to the fact that no strong
theory of pattern acquisition has been proposed that might explain language
acquisition as a special case. No one has yet shown that such a theory is
impossible in principle. [Slobin, 1971 and 1979.]

Reductionist and Functionalist Explanations

The second issue concerns what counts as an explanation in psychological

theories. The classical position is reductionism, the view that a true explana¬
tion of a behavior is couched in terms of independently motivated laws
applying to a level that is mapped onto the behavior. In the case of language,
an obvious goal in this tradition would be to explain linguistic knowledge in
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 313

terms of neurophysiological mechanisms that cause it. The problem is that

the phenomenon itself is both highly structured and behavioral, not physical.
This allows us to speculate that the same behavioral pattern can be related to
a variety of different kinds of physical events. The ready availability of
computer programs as models stimulates the view that causal reductionism
is not a necessary position: Rather than viewing linguistic structure as
caused by neurological structures, we can opt for a weaker reductionism, for
example, the position that functional and physiological mechanisms corre¬
spond without a unique causal relation between them. On this view, a cor¬
rect explanation of language involves the specification of rules, not how
those rules are physiologically instantiated. As in other cases, language is
not unique in this, but it is the most elaborate indigenous behavior we know
of and thereby highlights the problems for the reductionist most poignantly.
[Fodor, 1980.]

Psychological and Platonic Interpretations of Language

The third philosophical problem concerns the ontological status of linguistic

structure. Is it truly a unique expression of human psychology, or is it an
abstract structure without cause? We can observe such a contrast between
color vision (clearly physiological in origin) and formal logic. The latter is a
branch of mathematics and clearly not simply a codification of human reas¬
oning. The truths of logic are true independent of any human cause, though
which truths we happen to know may of course be due to human perspec¬
tives and limitations. Is language like color vision, or is it like logic?
At first, it might seem obvious that language is like color vision in that it
has unique and universal properties that seem to have no general psychologi¬
cal motivation; it is learned with little training and from impoverished data.
Such facts as these, however, bear directly on the claim that there is a
special psychological capacity to learn language of a particular form; these
facts do not speak unambiguously to the question what is the actual cause of
the form of language itself.
Certain features of language seem to have properties analogous to logic
and mathematics. For example, the concept of true-by-definition is itself
internal to language. Whether naive people are able or not to differentiate
analytic from synthetic truths, most linguists agree that the distinction is
intrinsic to language itself. There is no clear extrinsic cause for the existence
of such definitional concepts within a language; they simply are true of the
language itself, independent of who knows it or why.

APPLICATIONS OF LINGUISTICS

Language is central to many human activities. Accordingly, there are numer¬

ous ways in which attempts have been made to apply linguistics to practical
314 THOMAS G. BEVER

and theoretical problems. Today, computers are the most salient devices for
practical implementation of linguistic models. In the classroom, such prob¬
lems as language-learning and reading involve a mixture of theory and prac¬
tice. Finally, linguistics has been taken as a model for studying other human
symbolic activities.

Computer Science

One of the frequently proposed tests for the adequacy of a computer model
of human thought is whether we could tell that it were not human simply by
interacting with it through language. If the computer model responds in a
humanly plausible manner, success has been achieved. [Turing, 1936.] This
operational test is surprisingly easy to achieve if the communication context
is restricted. For example, a well-known computer program simulates the
conversational activity of an analyst. This program is effective just because
the analyst it simulates provides little new information in a therapy session.
Rather, the program reformulates statements typed in by the user and turns
them into corresponding questions. This leads us to conclude that the pro¬
gram is not a successful replication of a human being, even though it might
be hard to detect the difference in the restricted context.
We must treat every language capacity in a computer relative to its appli¬
cation. Three major kinds of applications have been attempted during the
past few decades: automatic translation, automatic comprehension, and
special-purpose implementations.
Automatic translation would seem to be an obvious candidate for com¬
puter processes. After all, much of translation involves word-by-word map¬
ping from one language to another. Computers can certainly implement
bilingual dictionaries with great speed. This does not, however, guarantee
good translation; in fact, it often produces completely incomprehensible
sequences in the target language. This is partly because words in sentences
represent meaning by virtue of their interaction together; partly because
there are numerous multiword idioms in every language; and because cor¬
rect interpretation of a sentence often involves an understanding of its rela¬
tion to other sentences in the same discourse and to the nonlinguistic context
in which it is used. Human translators bring their knowledge of the world to
bear as they translate, something that a computer cannot do, no matter how
sophisticated its linguistic program.
Machine translation dominated attempts to apply computers to language
for several decades, but there has been no general solution. Limited suc¬
cesses can transfer sentences in scientific documents from one language to
another, but human readers are expected to compensate for the occasional
mistakes that appear in even this highly restricted context. There is no
reason, however, why moderately good translation should be impossible in
principle. As computers become more powerful and have more storage,
“brute force" solutions become possible, that is, solutions that involve a
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 315

great deal of memorized whole-sentence transfer from language to language.

Such devices are unlikely to replace translators for artistic and diplomati¬
cally sensitive translation but may serve many useful functions. As the pro¬
grams depend increasingly on such brute solutions, they will depart from
implementing linguistic theory itself. This is not to say that humans do not
use memorized sentences and phrases in translation; rather, it reflects the
fact that language activity is not the same as linguistic structure. The role of
linguistics is to provide a set of descriptive constructs that may be used by
the programs but that may also not be used.
One result of attempts to create automatic translation programs was the
realization that translation necessarily involves some comprehension. That
is, we often can translate a sentence only after we have understood it in the
original language. This focuses attention on the problem of comprehension
itself: Can we construct a computer program that understands sentences in
English? The notion of what counts as understanding must itself be
specified. If understanding means construction of a human representation,
then it is unlikely that a program can be a general solution. However, we
may define understanding itself in terms of some limited objectives. For
example, the program can be attached to some physical device that carries
out English commands. Understanding is then defined in terms of correct
behavior by the device. Such devices and programs have been developed
that meet with the same kind of limited success as translation programs.
[Winograd, 1972b.] Given a limited physical world and a limited vocabulary,
limited success is always likely. Again, most existing programs make only
haphazard use of linguistic theory itself. Indeed, this is currently a subject of
controversy in the field of computational linguistics, the discipline devoted
to automatic use of language. Some argue that incorporating linguistic
theory is most likely to lead to an ultimately successful comprehension
system. Others argue that linguistic theory is irrelevant to comprehension
activity just as physical theory is irrelevant to playing billiards.
Comprehension programs generally avoid the problems of acoustic analy¬
sis of speech by starting with typewritten text. There have been, however,
repeated attempts to construct a voice-typewriter, a machine that would
produce a phonetic (or conventionally spelled) output from normal speech.
Such programs have failed so far, though limited success is likely via brute-
force methods.
The creation of “visible speech” has long been a concern of those inter¬
ested in aiding the deaf. This led to the search for acoustic invariants in the
speech signal that could be used as building blocks for recognition. The
simple example previously presented of the English sound system illustrates
one of the primary difficulties in this approach: Phonetic units do not consis¬
tently correspond to the same representational units of sound, either in
content or in serial location. In the words bitter and bidder, the distinction
between Itl and Idl is not acoustically reflected in a medial consonant but in
the relative length of the preceding vowel. The amount of computation re-
 316 THOMAS G. BEVER

quired to untangle just this simple problem is prodigious. Such language-

specific processes combine with universal aspects of speech production to
make the general problem even more difficult. The physical structure of
sounds is heavily determined by the sounds that surround them, just be¬
cause they are produced by continuously moving organs that adjust current
activity as a function of what came before and what is to come afterward.
This serial blending of sounds creates a nearly intractable problem for auto¬
matic recognition that is based on linguistic theory.
The problem may ultimately be solved in other ways; for example, a large
memory for an individual speaker’s phonetic style will make possible the
automatic recognition of speech by that speaker. Practical success, how¬
ever, may or may not involve linguistic theory itself. We can treat the
problem as a special case of signal processing and entirely ignore the lin¬
guistic attributions of phonetic segmentation and structure. To date, such
programs have not been successful, but there is no reason in principle why
they should not be.
The underlying issue will remain the same—even accurate dictation in¬
volves knowledge of the language and the world. Imagine the difficulty of
taking dictation in a language we do not know about problems we do not
understand. Even if the phonetics were identical with English, many odd
and undiscoverable mistakes would be made.
Linguistic theory may be of some direct application to special-purpose
devices that involve language. One area is in programming languages them¬
selves. It is of course possible to restrict the comprehension problem to the
application of programs. Ideally, we could construct programs by writing
normal English sentences that specify the steps. The possibility of this is
limited by the fact that programming operations themselves often involve
technical operations and relations between operations that are not a part of
normal experience or language. Thus, programming in English would neces¬
sarily involve highly specialized formulations. A more general role of lin¬
guistic theory is to provide constraints on the kinds of programming lan¬
guages we construct that will make these languages easier to learn and use.
The reasoning is that insofar as programming languages have universal prop¬
erties of natural language, they will be easier to manipulate. In actual prac¬
tice, most programming languages reflect only very elementary properties of
natural language—basically a semantic-based context-free structure, with
minimal rules of ordering and embedding.
The most fruitful areas of application involve special-purpose devices that
carry out limited language functions. For example, speech-production de¬
vices are of great practical importance, especially for the handicapped. At
the moment, several commercially available machines allow us to type in
normally spelled English and hear comprehensible pronunciations of it.
These devices incorporate models of the vocal tract as well as some
language-specific sound rules like the ones previously presented. Ulti¬
mately, general mastery of normal intonation and rhythmic patterns seems
LINGUISTICS: RELATIONS TO OTHER DISCIPLINES 317

achievable. Of course, here, too, the perfect solution would involve under¬
standing the utterance; but much of practical value can be achieved by
computer-generated speech, just because human listeners compensate for its
imperfections.

Linguistics in the Classroom

Two out of the traditional pedagogic three Rs involve language. Accord¬

ingly, it would seem reasonable that a correct understanding of what lan¬
guage is would be important for education. One rationale is that a correct
understanding of language would help students use it in reading and writing.
Linguistic theory has often been applied by the creators of reading texts.
Before 1960, the theory involved primary focus on sound-to-phoneme corre¬
spondences and the distribution of words and phrases. Reading texts were
constructed with the rationale that students had to build up sound-phoneme
correspondence units and master typical phrase patterns. With the advent of
transformational grammar, certain texts focused on training the reading of
whole sentences and their relation to other sentences. Indeed, at least one
set of texts was developed to teach junior-high-school students linguistic
theory itself.
Research on the effectiveness of such efforts is hard to carry out. To date,
there is no clear evidence that understanding linguistic theory leads to more
effective reading or writing. There is no doubt that knowledge of language
behaviors can be used by text writers to improve the way that language skills
are taught. However, as has been a repeated theme in this essay, the relation
between linguistic theory and behavior is itself a matter of controversy. In
particular, reading and writing may be skills that depend primarily on
behavior-specific mechanisms quite independent of language. The role of
linguistics would again be to provide consistent theoretical analyses of sen¬
tences in a language. Which of those analyses is relevant to which behaviors
is still an empirical question.
1
 THOUGHT, LANGUAGE,
AND COMMUNICATION

George A. Miller

The average American has always been more interested in communication

than in language. (More interested in communication than in thought, some
would say.) People can be persuaded that speech is an important form of
communication and that language is the competence underlying speech, but
it is difficult for most people to hold that perspective. George Herbert
Mead's pragmatic approach feels more comfortable. According to Mead, the
exigencies of living in groups created a need to communicate; language
evolved in order to meet that need; then language made it possible to think in
abstractions, to develop a concept of the self, and to behave purposefully.
But language is what it is, not because the human brain is what it is, but
because social interactions are what they are.
This relativistic, empiricist vision of human nature and human culture as
shaped by environmental pressures has always been attractive to American
psychologists—and was once attractive to American linguists. Reduced to
absurdity, of course, it becomes the strict behaviorism of B. F. Skinner, who
would explain speech as a chain of conditioned reflexes established by en¬
vironmentally controlled reinforcements and elicited by the occurrence in
the environment of the appropriate discriminative stimuli. But there is a
broader, pragmatic conception of human nature, one in which the human
brain itself has been shaped by the social environment in which it evolved. A
careful historian would not dismiss as a behaviorist everyone who was in¬
fluenced by this pragmatic vision.
An alternative view is that the nature of human language has nothing to do
with any need for social communication.1 Any complex organism—an ape,
say—must possess highly developed information-processing capacities in
order to survive in an unpredictable environment. Plans must be formulated,
alternatives evaluated, data collected and stored, responses made, and con¬
sequences noted. Precisely how the nervous system coordinates the flow of

*1 am indebted to Dan Sperber for suggesting this way of stating the alternative view.
319
320 GEORGE A. MILLER

information required to keep this system functioning is not yet well under¬
stood, but some theorists have spoken, not entirely metaphorically, of the
languages of the brain. The basic communicative process is self-regulatory,
not social.
According to this view, the unique event in human evolution occurred
when our ancestors somehow became able to use the languages of the brain
for social communication. Some mutation made it possible to externalize the
internal information-processing—some kinds of internal processing, at
least—in vocal noises. Thus, human patterns of thought did not await the
evolution of language but preceded it. Social cooperation based on the frag¬
mentary clues to thought normally expressed in human speech would sel¬
dom succeed if the internal information-processing were not closely similar
from one individual to the next; such uniformity is not characteristic of the
environments in which people live, but it could be achieved genetically. In
short, language is what it is, not because social interactions are what they
are, but because the brain is what it is.
People frequently have trouble understanding spoken messages that they
do not expect, even when the acoustic signal is loud and clear. The reliance
on expectations is sometimes called top-down processing of speech, by
contrast to bottom-up processing that begins with the acoustic signal, seg¬
ments it, then filters and classifies the segments. Engineers who have tried to
build speech-recognition systems have had their greatest success with bot¬
tom-up approaches, from which it is sometimes hastily concluded that the
importance of top-down processing has been vastly overrated. A more pru¬
dent conclusion would be that engineers have not yet been able to simulate
top-down processing very well, a conclusion that would follow from decent
respect for the complexity of neural information-processing and our ignor¬
ance of how it is done. Everything a person knows about another person’s
situation, past or present, can be used to limit what that other person could
be expected to say and so make perception of what they do say easier. That
kind of coordination is possible because different people think the same
way, but no machine has yet been programmed to think as people do.
Part of the internal information-processing that all people do by virtue of
being human is described by the universal properties and principles of levels
of syntactic representation. The good news is that it has been possible to
characterize even a fragment of this internal processing. The bad news is
that that is but a small part of the system that psychologists eventually hope
to understand.
 CURRENT VIEWS OF
LANGUAGE AND GRAMMAR

Ray C. Dougherty

A basic problem in linguistics is to define the processes by which a child

acquires a language on the basis of exposure to adult speakers of the lan¬
guage. In 1957, Noam Chomsky published Syntactic Structures and defined
a language as a set of sentences, and a grammar as a device for defining the
sentences of a language. [Chomsky, 1957.] In most linguistic representa¬
tions, a grammar is a combination of algebraic devices that defines all of the
sentence types (active, passive, question, etc.) of a language and enumerates
all of the individual sentences. The linguist’s formal grammar is considered
to be a representation of a person’s knowledge of his or her language.
Viewed in this perspective, the questions asked in linguistics are: How
can a child, exposed to sets of sentences, acquire a grammar? What is the
algebraic form of the grammar? What type of data must be presented to a
child for the child to acquire a grammar? It is traditional to say that a child
“learns” a language on the basis of exposure to sets of sentences. Current
linguistics has developed a biological perspective which permits us to con¬
sider a grammar as an information-engine that is resident in an information¬
processing host. An information-engine, of which a grammar is one example,
is a highly structured body of knowledge that has the capacity to form a copy
of itself in an information-processing host identical with the one in which it
finds itself. One might think of a grammar as a computer program and a host
as a type of symbol-processing computer. Grammar reproduction is the
process by which a computer program (a grammar) in one host can form a
copy of itself in another, identically structured host.

FOUR POSITIONS

One can obtain an understanding of the basic problems, if not the solutions,
by considering recent works in which four distinct positions have been dis¬
cussed. One position, the empiricist view, characterized by one author cited
later as a mere recording of observations without a structuring activity on
321
322 RAY C. DOUGHERTY

the part of the subject, governs little current research. Each of the three
remaining positions has its champions.
Chomsky, who has formulated most of the issues defining the biological
perspective, presents an “innatist,” or, as Piaget calls it, “preformationist”
view, according to which all of the basic linguistic structures are precoded
into the genetic constitution of a human being and triggered by experience.
Jean Piaget presents a “constructionist” view, which claims that the
basic properties of a grammar are neither implicit in the data to which a child
is exposed nor innate in the mental organization of a child. Constructivism
assumes that grammatical structures emerge as constructions developed by
general sensorimotor processes when applied to language data. A child
learning a language essentially creates a grammar anew by general processes
of sensory perception and motor coordination that are not language-specific.
Seymour Papert has recently developed a computational approach, called
Project LOGO, which is closer to Piaget’s constructivism than to
Chomsky’s innatism. Papert has attempted to develop a computer-based
model of learning that uses mechanisms developed in computer science to
characterize neurological structures. Papert developed the computer lan¬
guage LOGO in order to model the information processes that function to
characterize knowledge and the mechanisms by which knowledge is repro¬
duced in a new brain. Papert assumes that one could better understand the
processes of grammar and language if one had more understanding of the
nature of machines that learn and machines that learn how to learn. He uses
LOGO to design such machines and model the neurological processes in¬
volved.

THE INNATIST VIEW

In the following passage, Chomsky outlines the biological perspective and

his innatist views. He indicates his skepticism concerning constructions of
sensorimotor intelligence.

It is a curiosity of our intellectual history that cognitive structures developed

by the mind are generally regarded and studied very differently from physical
structures developed by the body. There is no reason why a neutral scientist,
unencumbered by traditional doctrine, should adopt this view. Rather, he
would, or should approach cognitive structures such as human language more
or less as he would investigate an organ such as the eye or heart, seeking to
determine: (1) its character in a particular individual; (2) its general properties,
invariant across the species apart from gross defect; (3) its place in a system of
such structures; (4) the course of its development in the individual; (5) the
genetically determined basis for this development; (6) the factors that gave rise
to this mental organ in the course of evolution. The expectation that construc¬
tions of sensimotor intelligence determine the character of a mental organ such
as language seems to me hardly more plausible than a proposal that the funda-
CURRENT VIEWS OF LANGUAGE AND GRAMMAR 323

mental properties of the eye or the visual cortex or the heart develop on this
basis. Furthermore, when we turn to specific properties of this mental organ,
we find little justification for any such belief, so far as I can see. (Chomsky,
1980fl, p. 37.)

Chomsky believes that the structures of language and the complex pro¬
cesses of grammar development could not arise from organism/environment
interaction alone. Grammar is species-specific and genetically determined.
Chomsky elaborates:

I have touched on only a few examples. In each case, when we investigate the
particular properties of human cognition, we find principles that are highly
specific and narrowly articulated, structures of a marvelous intricacy and deli¬
cacy. As in the case of physical organs, there seems to be no possibility of
accounting for the character and origin of basic mental structures in terms of
organism-environment interaction. Mental and physical organs alike are deter¬
mined, it seems, by species-specific, genetically determined properties, though
in both cases interaction with the environment is required to trigger growth and
will influence and shape the structures that develop. Our ignorance—
temporary, let us hope—of the physical basis for mental structures compels us
to keep to abstract characterization, in this case, but there is no reason to
suppose that the physical structures involved are fundamentally different in
character and development from other physical organs that are better under¬
stood, though a long tradition has tacitly assumed otherwise. (Ibid., pp. 51-
52.)

THE CONSTRUCTIVIST VIEW

At the same conference where Chomsky offered these views, Piaget defined
constructivism as an alternative to both empiricism and innatism (or prefor-
mationism). Said Piaget:

Fifty years of experience have taught us that knowledge does not result from a
mere recording of observations without a structuring activity on the part of the
subject. Nor do any a priori or innate cognitive structures exist in man; the
functioning of intelligence alone is hereditary and creates structures only
through an organization of successive actions performed on objects. Conse¬
quently, an epistemology conforming to the data of psychogenesis could be
neither empiricist nor preformationist, but could consist only of a constructiv¬
ism, with a continual elaboration of new operations and structures. The central
problem, then, is to understand how such operations come about, and why,
even though they result from nonpredetermined constructions, they eventually
become logically necessary. (Piaget, 1980, p. 23.)

Piaget defines what he sees as the basic problem of research into the
nature of language and grammar: To choose between constructivism and
324 RAY C. DOUGHERTY

preformationism. He also considers empiricism and preformationism

equally vacuous.

The problem is therefore to choose between two hypotheses: authentic con¬

structions with stepwise disclosures to new possibilities, or successive actuali¬
zation of a set of possibilities existing from the beginning. (Ibid., p. 25.)

In a word, the theories of preformation of knowledge appear, for me, as devoid

of concrete truth as empiricist interpretations, for the origin of logico-
mathematical structures in their infinity cannot be localized either in objects or
in the subject. Therefore, only constructivism is acceptable, but its weighty
task is to explain both the mechanisms of the formation of new concepts and
the characteristics these concepts acquire in the process of becoming logically
necessary. (Ibid., p. 26.)

THE COMPUTATIONAL VIEW

Papert’s project LOGO can be understood as having developed out of, or at

least having been inspired by, many of Piaget’s ideas about the genetics of
grammar reproduction. Piaget’s book, Genetic Epistemology, the published
version of the Woodbridge Lectures he delivered at Columbia University in
1968, discusses self-regulation of the growing knowledge in a new brain:

These few examples may clarify why I consider the main problem of genetic
epistemology to be the explanation of the construction of novelties in the
development of knowledge. From the empiricist point of view, a “discovery”
is new for the person who makes it, but what is discovered was already in
existence in external reality and there is therefore no construction of new
realities. The nativist or apriorist maintains that the forms of knowledge are
predetermined inside the subject and thus again, strictly speaking, there can be
no novelty. By contrast, for the genetic epistemologist, knowledge results from
continuous construction, since in each act of understanding, some degree of
invention is involved; in development, the passage from one stage to the next is
always characterized by the formation of new structures which did not exist
before, either in the external world or in the subject’s mind. The central prob¬
lem of genetic epistemology concerns the mechanism of this construction of
novelties which creates the need for the explanatory factors which we call
reflexive abstraction and self-regulation. However, these factors have fur¬
nished only global explanations. A great deal of work remains to be done in
order to clarify this fundamental process of intellectual creation, which is
found at all the levels of cognition, from those of earliest childhood to those
culminating in the most remarkable of scientific inventions. (Piaget, 1970, pp.
77-78.)

The mechanisms of self-regulation discussed by Piaget provided only

global explanations not specific ones. Papert’s project LOGO is an attempt
to make many of the ideas of Piaget’s genetic epistemology concrete. Papert,
CURRENT VIEWS OF LANGUAGE AND GRAMMAR 325

in discussing one of his examples, indicates how Piaget influenced his work:
“When I read Piaget this incident served me as a model for his notion of
assimilation, except I was immediately struck by the fact that his discussion
does not do full justice to his own idea.” (Papert, 1980, p. vii.) Papert
mentions that Piaget’s ideas gave him “a new framework” for examining his
observations. (Ibid., p. viii.) Papert’s project LOGO is a version of construc¬
tivism and opposed to both empiricism and preformationism. Papert states:

Noam Chomsky believes that we have a language acquisition device ... I do

not . . . [share] his view o he brain as made up of specialized neurological
organs matched to specific intellectual functions. I think that the fundamental
question for the future of education is not whether the brain is “a general
purpose computer” or a collection of specialized devices, but whether our
intellectual functions are reducible in a one-to-one fashion to neurologically
given structures.

It seems to be beyond doubt that the brain has numerous inborn “gadgets.”
But surely these “gadgets” are much more primitive than is suggested by
names like Language Acquisition Device and Math Acquisition Device. I see
learning language or learning mathematics as harnessing to this purpose numer¬
ous “gadgets” whose original purpose bears no resemblance to the complex
intellectual functions they come to serve. (Papert, 1980, p. 220.)

EXPERIMENTAL ANALYSIS

Much of the excitement of research into the structure of language and the
genetic processes governing grammar reproduction derives from the fact
that the results of research bear directly on traditional questions about the
structure of knowledge and the mechanisms by which it is communicated
from the older to the younger generations. In linguistics, traditional philo¬
sophic questions can be posed in sufficient detail that they can be subjected
to experimental analysis.
 RULES AND PRINCIPLES
IN PHONOLOGY AND SYNTAX

Samuel Jay Keyser

The following piece attempts to do two things. First, it attempts to show why
linguists suppose that mastery of a language is tantamount to mastery of a
system of rules rather than, say, memorization of a very large number of
words, phrases, sentences, and so on. Second, it draws attention to a major
difference that has emerged in recent years between the rule systems of
phonology and the rule systems of syntax and speculates on the significance
of this difference.1

WHY RULES?

A fundamental tenet of linguistic theory is that what people know when they
know how to speak is a system of rules. The argument goes as follows:

1. People have, in principle, the ability to produce an infinite number of

sentences; thus, consider the following sentences:

(a) John said that Bill is here.

(b) John said that Frank said that Bill is here.
(c) John said that Mary said that Frank said that Bill is here.
(d) John said that Ned said that Mary said that Frank said that Bill is
here.
2. Beginning with (a), note that each subsequent sentence is longer than
the preceding one by the addition of the phrase that (someone’s
name) said after the first occurrence of the verb form said. However,
this process, repeated three times, can be repeated any number of
times. Each time it is repeated, a new sentence is produced. Since
there is no upper limit on the number of times the phrase that (some¬
one’s name) said can be introduced into a sentence like (a), there is
no upper limit on the number of English sentences that can be pro¬
duced in this fashion.

1 The author is indebted to Morris Halle and James Higginbotham for a discussion of some of the
issues raised in this piece.
327
328 SAMUEL JAY KEYSER

3. The argument in (1) shows that there can be no upper limit on the
number of sentences in English. However, we know that the human
brain, the repository of linguistic ability, is finite. Even if we were to
assign one sentence to each atom of gray matter in the brain, we
would run out of atoms long before we ran out of sentences.
4. If knowledge is represented mentally and if we have the ability to
produce an infinite number of sentences, then how can we represent
this infinite knowledge in a finite space?
5. The answer is to assume that the knowledge that enables us to pro¬
duce an infinite number of sentences is not a representation of each
individual sentence, but rather a set of rules, finite in number, whose
output is, in principle, infinite.
6. Therefore, what we know when we know how to speak is, among
other things, a set of rules.2

This argument is based on syntax. However, the ability to speak a lan¬

guage goes beyond syntax to include phonology as well. People know how to
pronounce words as well as how to arrange them into sentences. Bever
notes, in this regard, that “Words have an inner level of representation of
sound sequences and a set of rules that map them onto an outer level repre¬
senting the actual acoustic sequences.” (Bever in his paper in this volume.)
It is not immediately obvious, however, that knowledge of how to pro¬
nounce words requires knowledge of rules. Let us consider, then, whether
phonological knowledge must also be represented by rules.

WHY PHONOLOGICAL RULES?

We have seen in syntax that rules are required since speakers know, in
principle, an infinite number of sentences. However, in phonology we might
argue that since the number of morphemes and words is finite, there is
nothing incoherent about supposing that a speaker simply memorizes the
phonetic makeup of each word. How might this work?
Consider the well-known fact that regular English plurals are pronounced
in three separate ways: either as the syllable [iz] as in kisses or as the
consonant [s] as in hits or as the consonant [z] as in dogs. We might say that
when a child learns the English plural, the child simply memorizes these
facts. In other words, just as a child must remember that the plural of mouse
is mice, the child must also remember that the plural of spouse is spouses
and not, say, spice. Such a model is theoretically possible, since the number
of words in English is finite.

2This argument makes use of that property of natural language that Bever calls “the creative
component of sentence structure.”
RULES AND PRINCIPLES IN PHONOLOGY AND SYNTAX 329

The Novel Form Argument

One way to argue for the view that phonology is rule-governed is to show
that novel forms do not present a problem for speakers. That is to say,
speakers are able to treat novel forms as if they were not novel at all.
Suppose, then, we were to ask someone to provide us with the plural form of
a word that ended in a consonant sound that was not a normal English
consonantal sound. Thus, consider the sound [xL which terminates the
name Bach', that is [bax].3 Even though this sound is not an English sound, a
native speaker of English would always pronounce the plural of this word
with an [s] and not with an [iz] or a [z]. This response can easily be verified
by asking native speakers of English whether they have ever heard the word
before and, if not, what they suppose its plural to be. Moreover, this re¬
sponse can only be understood if we assume that part of a native speaker’s
knowledge of the English plural is knowledge of a rule and not knowledge of
pairs of singular and plural forms. Thus, if there were no rule, it would be an
incredible coincidence that all English speakers treat the plural of Bach as [s]
and not, for example, [baxiz]. Notice that both forms are easily pronounce¬
able in English, and, were there no rule, we would have to assume that at
least one speaker would opt for [baxiz]. This, in fact, never happens.4
We saw that in syntax the argument in favor of rules was based on the
need for an infinite output. We have just seen how in phonology the argu¬
ment in favor of rules is based on novel forms. However, both of these
arguments are, in fact, based on the same property, namely, novelty. Thus,
for speakers who have never heard the name Bach before, the name consti¬
tutes a novel word whose plural will, nonetheless, be obvious to them.
Similarly, a sentence patterned after those in the preceding example that is,
say, three times as long as (d), or even ten times as long, will be a novel
sentence to anyone with the patience to put such a sentence together. None¬
theless, such a sentence, despite its novelty, will easily be recognized as
being a perfectly grammatical one in English by anyone who knows the
language.

Words Are Not Finite

It is not hard to show, however, that the premise on which the ruleless
hypothesis is based, namely, that the number of words in English is finite, is
itself incorrect. Consider, for example, the words lion and hunter. English

3This is the voiceless, velar sound that occurs in German words like ach.
4We will not pause here to provide a formal statement of the English plural rule. However,
several accounts can be found in contemporary literature. For one that also includes an account
of its historical evolution, the reader is referred to Keyser and O’Neil [1980].
330 SAMUEL JAY KEYSER

has a process called compound formation whereby these words can be put
together to form a single new word, lion hunter meaning a hunter who hunts
lions.5 Notice that if we can add a word to the word hunter once, there is no
reason why we cannot perform the same operation twice. Thus, suppose we
now add the new word lion hunter to the word hunter to produce another
new word, namely, lion hunter hunter. This is not only an acceptable word
in English, it is also perfectly clear what it must mean. A lion hunter hunter
means a hunter who hunts lion hunters.
We have now constructed an argument that is formally identical with that
in the first set of sentences. We must have phonological rules because our
knowledge of words, like our knowledge of sentences, is infinite. This is the
novel-form argument for rule-based linguistic systems. Indeed, all novel
forms can be construed as evidence in favor of the view that rule systems
underlie our linguistic competence.6

PHONOLOGICAL AND SYNTACTIC RULE SYSTEMS COMPARED

Even though syntax and phonology share the property of creativity (see
footnote 2), there is an interesting difference between the two components.
In his subsection on “General and Specific Learning Mechanisms,” Bever
makes reference to modular versus general purpose models of learning. The
modularity hypothesis assumes that “learning is carried out by a set of
innate mental ‘modules,’ each highly specialized for its task . . . .” In this
view, the mind is a “federation of such distinct modules.”
The modular model of mental representation describes the current state of
affairs in linguistic theory as well. This change toward modularity in lin¬
guistic theory has come about over the past few years. Noam Chomsky, in
his recent monograph, puts the matter in the following terms: “There has
been a gradual shift of focus from the study of rule systems, which have
increasingly been regarded as impoverished (as we would hope to be the
case), to the study of systems of principles, which appear to occupy a much
more central position in determining the character and variety of possible
human languages.” (Chomsky, 1982, pp. 7-8.)

5 Even though we tend to write such compound words with space between them, this ortho¬
graphic convention should not mislead us into thinking that we are not dealing with a single
word. For one thing, this word takes plural endings just like other single words. Just like other
single words, we cannot put things inside it. Thus, words like taffeta cannot be interrupted and
neither can words like lion hunter. Hence, we cannot say lion big hunter even though we can
say big hunter. Moreover, we cannot place the plural ending inside the word; that is, the plural
of lion hunter is lion hunters and not lions hunter. We see, then, that in a number of ways lion
hunter behaves like a single word.
Novel forms come from a variety of sources. Errors that speakers make is one common
source. Thus, when a child assumes that the plural of mouse is mouses and not mice, the child’s
behavior can best be understood as resulting from an overgeneralization of the rule for the
regular formation of plurals in English.
RULES AND PRINCIPLES IN PHONOLOGY AND SYNTAX 331

Linguistic theory is now characterized by levels of representation and

principles appropriate to those levels. Within this framework the principles
act like filters on representations, indicating whether a given representation
is well formed or not.7 Consider the following sentences by way of illustration:

(a) Mary bought a hat.

(b) What did Mary buy?
(c) Whati did Mary buy [t;]?
(d) Mary mentioned that Sally bought a hat.
(e) What- did Mary mention that Sally bought [tj]?
(f) Mary mentioned the fact that Sally bought a hat.
(g) *What-{ did Mary mention the fact that Sally bought [t; ] ?

Note, first, that (b) is related to (a). In particular, (b) contains a fronted
question word what. However, while this word appears at the front of the
sentence, it is, nonetheless, the object of the verb buy. To capture this
intuition, we place an abstract element called a trace (= t) where the object
of buy normally appears, and we index the question word and the trace to
indicate the relationship that exists between them. This is indicated in (c),
which therefore constitutes the appropriate representation of the question
corresponding to (a) at one level of grammatical representation.8 Consider,
now, the question that corresponds to (d) at this same level of representa¬
tion, namely, the sentence indicated in (e). Here, too, a trace element relates
the fronted question word what with the position that the object of buy
normally occupies. This brings us to the final pair of sentences, namely (f)
and (g). Notice that whereas (f) is a perfectly well-formed sentence in En¬
glish, it does not have a corresponding question form of the pattern indicated
in the earlier questions, namely, in (c) and (e). When we attempt to repro¬
duce that pattern, we get an ungrammatical sentence such as that in (g). On
the surface of things this is a remarkable fact. Why, after all, should it not be
possible to question (f) just as it was possible to question (a) and (d)? To
account for this fact, a principle is assumed to apply at the level of structure
corresponding to (g). This principle, called subjacency, states that a fronted
element may not occur too far away from its original position (where, for
purposes of the present exposition, we leave unspecified what too far away
means formally):

Subjacency: In a sentence that contains question words and traces,

the question word may not occur “too far” away from its trace.

7The shift from rules to principles in no way undermines the argument in (1) that knowledge of a
language means knowledge of a grammar. Rather the argument goes through but with principles
replacing rules as the dominant mechanism of grammatical knowledge.
8This level is called 5-structure, and it corresponds roughly to the notion of surface structure.
332 SAMUEL JAY KEYSER

The role of subjacency is as follows: Given a representation such as that in

(g), the principle checks to see whether the distance between a questioned
element and its trace is too great. If it is, the representation is treated as
ungrammatical. In this sense, the principle acts as a filter on the level of
representation of (g).
The model of syntax that emerges from this example is one that is static in
nature. According to this model, there are levels of representations, and
there are principles (or filters) that apply to representations at those levels. If
the principles are met, the representation is well formed; if they are not, it is
not. While this static model appears to be true of syntactic and even seman¬
tic levels of representation, it is not true of the phonological level. The latter
level is, in marked contrast, characterized by rules and not by filters. More¬
over, these rules perform a wide variety of functions. Thus, there are rules
that (1) insert sounds, (2) delete sounds, (3) relate sounds to one another in a
variety of ways, (4) change values of phonological features, and (5) add new
phonological features to a representation.9
In short, while the higher levels of contemporary linguistic theory have
become much more static in character, the lower levels—specifically, the
phonological and phonetic levels—have become much more dynamic in
character. Even though there is not sufficient space to illustrate this differ¬
ence between the various levels of linguistic theory, the difference is one
that contemporary researchers in the field would probably not find contro¬
versial.

Why the Difference?

The final portion of this piece is devoted to a speculation about why

phonological and phonetic components are dynamic in character while the
higher levels are static in character. One possible answer to this question lies
in the fact that the phonological component is deeply embedded in the physi¬
ological mechanisms of the vocal tract and auditory system. For example,
crucial to phonological theory is the notion of distinctive features, and these
features correspond in a rather direct way to gestures of the vocal tract.
Thus, the feature nasality plays a prominent role in the phonology of most of
the languages of the world. This feature refers specifically to whether the
velum is open or closed during any given speech gesture.10 For example, the
velum is open during the first sound of the word man and it is closed in
the first sound of the word can. The active character of the phonological
component may well be a direct consequence of phonological theory being
closely related to the mechanisms that actually produce the sounds of lan¬
guage.

This brief enumeration barely does justice to the active character of the phonological com¬
ponent.
The velum is the soft fleshy valve at the roof of the mouth, which, when shut, closes off the
nasal cavity and, when open, allows the nasal cavity to resonate.
RULES AND PRINCIPLES IN PHONOLOGY AND SYNTAX 333

The situation is quite different when we come to syntax and semantics.

Here, there is absolutely no notion at all of what the mechanisms are that
correspond to such entities as nounphrase, verbphrase, sentence, com¬
plementizer, and specifier (such as articles and adjectives). While we have a
clear idea of what the feature nasal refers to, namely, the position of the
velum, we have no clear idea of what grammatical category might refer to in
the brain. Similarly, there is no clear idea of what the mechanisms might be
that correspond to such notions as coindexing traces with question words as
was illustrated in (c) and (e).

CONCLUSION

I conclude with a question. Higher levels of grammatical theory have not yet
achieved the marriage of the formalism with the mechanism that phonology
has at least begun to achieve. As we learn more about mental mechanisms,
will syntactic and semantic theory change in ways that will reflect those
mechanisms just as phonological theory has done, or is the present nature of
syntax and semantics reflective of those mechanisms already?11 If a final
speculation is in order with respect to this piece, then that speculation would
be that syntax and semantics will follow phonology as more is learned about
higher order mechanisms of the brain. Given the rightness of this specula¬
tion, we can expect radical changes in syntactic and semantic theory to
confront us in the future.

POSTSCRIPT

Before his sudden and unexpected death, Fritz Machlup requested that I
might add a word about the relationship between my remarks here and
studies in the information sciences such as those represented in the present
volume. At the risk of going beyond the space limitations imposed on me by
the editors, I would like to attempt to comply with his request.
During the past quarter of a century, the study of linguistic theory has, to
a great extent, been carried on independently of the study of the computa¬
tional mechanisms that enable the sentences of a language to be both pro¬
duced and recognized. Thus, linguistic theory studies human language com¬
petence, while the study of computational mechanisms, which might be

"James Higginbotham (personal communication) has suggested that the static character of
higher level components has resulted from a shift away from conditions on rules (a dynamic
device) to filters, or conditions on well formedness at a linguistic level (a static device) as a
mechanism of grammatical description. This shift has come about for reasons internal to lin¬
guistic theory itself; in particular, because of the ease of description that results from the use of
filters as opposed to rules. Phenomena under current study would require significantly more
complicated statements if formulated in terms of rules rather than filters.
334 SAMUEL JAY KEYSER

supposed to instantiate this competence, deals with human language perfor¬

mance. In recent years, however, a trend has been mounting that suggests
that these studies should no longer be carried out independently of one
another. Rather, the view is growing that each study can, in fact, illuminate
the other.
There are at least two reasons why this trend is desirable. The first is an
intellectual one. The question of how human knowledge is represented men¬
tally is a profound and important one. The more we understand about what
grammars are, the more we will come to understand what we are, since
grammatical knowledge is at the very heart of human mentality.
The second reason is a practical one. The introduction of the personal
computer has brought to light the fact that computers are, for the most part,
very difficult to use without a great deal of training and sophistication. We
have only to consider how easy it is to master driving an automobile to see
that the computer has a long way to go before it can be equally accessible.
One way to facilitate this process might well be to endow the computer with
a language that resembles our own as much as possible. This means that we
need to understand how parsing mechanisms can best be implemented in
computers. However, it may well be that the best way to understand how to
build parsers into computers is to understand how parsers are already built
into people. The study of how to make computers more accessible is, in
other words, a form of applied cognitive science and the more we under¬
stand how we work linguistically the better we may be able to design com¬
puters for us to work with.
 COMPUTATION DOES MATTER
TO LINGUISTICS
Robert C. Berwick

A view widely held by academics over the past few decades has been that
computer science and modern linguistic theory do not mix. Several years
ago, so the story goes, linguists driven by Noam Chomsky’s distinction
between linguistic competence and linguistic performance proposed that the
study of what people know about their language should take precedence
over the study of how people actually process or produce utterances or
actually learn language. This distinction between what and how has attained
a kind of self-perpetuating status as the driving wedge between linguistics on
the one hand and artificial intelligence and computer science on the other—
see Newell’s characterization of it as a “field-splitter.” [Newell in his lead
paper in Section 3 of this volume.]
But times change. I would like to argue that we would be hard pressed to
find a more complementary marriage of disciplines than that between lin¬
guistics and computer science and that such a marriage has recently borne
and will continue to bear great fruit. Linguistics and computer science are
partners of a very special kind, because the study of human cognition is now
widely equated with the study of information-processing models. But, as
every computer scientist knows, such models consist of algorithms that
juggle very specific information structures. Algorithmic procedures for using
knowledge do not make sense without a specification of the form in which
that knowledge is couched. Similarly, an understanding of how language is
used hinges on an understanding of what that knowledge looks like—the
way it is represented. So the joint work of linguistics and computer science
is like the partnership between data structures and the algorithms that use
them. In the remainder of this paper I would like to make plain just what the
nature of that partnership is and what one can expect to gain from it.
When talking about a living organism—any organism—there are two
broad kinds of “how” questions, and each has its place in the explanation of
human information-processing abilities. The first of these questions, raised
in Dougherty’s paper, is couched at the level of the species: How is what the
individual organism knows (a body of information) passed on to its descen-
335
336 ROBERT C. BERWICK

dants? Or, to put the question from the perspective of the offspring, how is
knowledge of language acquired? This is a particular kind of performance
that is rarely mentioned by researchers concerned with the analysis of the
day-to-day use of language—for example, it is not even mentioned in
Newell’s lead paper in this volume as a goal of the computational study of
linguistic performance. The second kind of how is, in fact, just this more
familiar one of individual language use: How do people actually use informa¬
tion about language to speak or understand sentences?
The linguist’s answer to both questions, as exemplified in the papers by
Bever, Dougherty, and Keyser, is to characterize what an individual knows
about language as a system of rules, a grammar. As an example of what a
part of a grammar looks like, consider the following sentences (like those
described in Keyser’s paper):

(a) I believe Mary likes Bill.

(b) Who do / believe Mary likes?

Now, as Keyser observes, what modern linguistic theory does is tell one that
(b) really has the underlying form,

(c) Who do I believe Mary likes [t]?

where [t] acts like a variable bound to the value of who, so that the question
is really to be construed as,

For which X, X a person, do / believe Mary likes X?

Further, what people seem to know about forming questions is that one can
take a statement like (a) and turn it into a question by imagining that who is
the object of like:

I believe Mary likes who.

and then moving who to the front of the sentence, inverting the auxiliary
verb and subject, and leaving behind the (unpronounced) variable X. Appar¬
ently, though, we cannot move who very far. We can move it out of a lower
sentence (for example, Mary likes who) into the next higher sentence but no
farther. Look what happens when we try to move who out of a sentence that
is in turn embedded in a nounphrase, the claim that . . . :

I believe the claim that Mary likes who —>

Who do I believe the claim that Mary likes [t].

The sentence is ill-formed, and we know this, even though we can make
out what it would mean if we could say such a thing. This is part of our
COMPUTATION DOES MATTER TO LINGUISTICS 337

knowledge of English, part of our grammar. Because this knowledge has the
effect of restricting the radius of action of the movement of who, we might
call it a locality principle, PI:

Who cannot be moved across more than one sentence boundary.

Now, so far, all this is a fairly abstract characterization of what people

know. The burning question for the computationalist is whether all this
machinery—the notion of leaving behind variables and moving elements—is
actually involved when people use language. Or is it all just a picturesque
metaphor, useful for exposition perhaps, but not really a part of what people
actually do?
It is here that computational arguments come to the fore. The beauty of
computer science is that it gives us a useful way of looking at complex how
questions. It is an approach that is perhaps best illustrated by a simpler but
more familiar computational problem.
Consider the following task. Suppose we wanted to add up the first 100
integers, 1 + 2 + . . . + 100. Let us call this the abstract description of a
computational problem. We have described what the problem is, not how to
solve it. Now let me give two simple algorithms for solving the abstract
problem. The most obvious way to proceed would be to add 1 to 2 and get
the sum 3; then add the next integer 3 to this sum, and get 6; add 4 to this,
and so on. On the other hand, if I were clever, I could pair up 1 and 100, 2
and 99, 3 and 98, . . . , 50 and 51, noticing that each of these 50 pairs add to
101. So all I have to do is to multiply 101 by 50.
The point is that any single abstract computational problem could have
more than one algorithmic solution. Further, it is plain that the two al¬
gorithms just given differ greatly in the kind and number of operations that
they use. Algorithm 1 uses 99 additions, while Algorithm 2 uses just one
multiplication. While the second seems vastly superior to the first in terms of
the amount of time it would take to carry out, the specific or particular
algorithm we actually use depends on the nature of the machinery at our
disposal. In general, there could be many ways of carrying out or implement¬
ing one and the same algorithm, and implementation details could make a
difference to how fast the algorithm works. Suppose, for example, that we
had a pocket calculator with a working addition button but a broken multipli¬
cation button. Now, in order to carry out Algorithm 2, we would have to add
101 up 50 times—still better than Algorithm 1 but not by so wide a margin.
What this means is that a complete computational analysis depends on
juggling at least three levels of theory: the abstract characterization, the
algorithm, and the implementation. (This way of analyzing computational
problems has been most clearly stated by Marr and Poggio. [Marr and Pog-
gio, 1977.]) While theoretical analysis of each level cannot be isolated from
the others, logically the specification of the abstract problem is prior to that
of the algorithmic level or a detailed implementation. Once the basic outline
338 ROBERT C. BERWICK

of all three levels has been determined, then we can use what we know about
the implementation and algorithmic constraints to tell us something about
the abstract problem, and vice versa.
Turning now to the situation of language, the first thing we notice is that
what plays the role of the abstract computational problem are forms such as
(c). If one thing that we do in order to understand a sentence is to recover its
underlying form, then we must convert the string of words,

(d) Who does Mary like?

into the form

Who does Mary like [t]?

That is, linguistic theory tell us what the abstract computational problem is,
since it tells us what the units of representation are that must be recovered.
Now let us consider the how of it all. Just as with the 100-integers problem,
there could be many ways of actually performing the computations required
to get to the underlying form. And there is one additional factor to consider:
While we are free to speculate about algorithms and implementation
methods to solve the integers problem, in the case of linguistic algorithms
and implementations we are shackled not by our own creativity but by
Nature’s—at least if we want to explain what it is that people do. The chains
are wound doubly tight because what we know about the machinery that the
brain has at its disposal is quite limited; we know that it does not have add
and multiply buttons, but we know very little more than that.
Even so, we can still advance computational arguments for and against
certain ways of characterizing the what of linguistic theory. Let me give just
one example of this style of argumentation here, an example showing that
there is some computational evidence for the abstract form (c) posited by
modern transformational grammar. In passing, the argument will illustrate
one way in which a theory of computation can support the study of linguis¬
tics.
To begin, consider the following two algorithms (machines) for analyzing
sentence (d). Each machine takes as input the sentence to be analyzed and
writes as output a desired underlying form corresponding to the input sen¬
tence.
Algorithm 1. Imagine sliding a window across the input sentence. De¬
pending on what it sees in the window, the analysis machine makes a deci¬
sion as to what output to produce next. The window is fixed in width, in the
sense that it can hold a finite number of words (such as did) or phrases (such
as a sentence or nounphrase) but no more. This means that we can express
the operations of the machine in terms of finite patterns and actions, where a
pattern is simply a finite string of words or phrases, and an action is what the
machine writes as output. In our example, if the window of the machine
holds the items
COMPUTATION DOES MATTER TO LINGUISTICS 339

Who does Mary like?

then the machine writes out the same string as it sees, with the [t] marker
inserted as desired

Who does Mary like [t] ?

It is important to observe that the representation of the operating rules must

be finite if they are to be literally stored in a finite machine.
Algorithm 2 works differently. It assumes an underlying form distinct
from that described by (c). Instead of inserting the element [t], suppose that
we simply put back the displaced who, relocating it in its original position:

Who does Mary like? —>

Does Mary like who?

Computationally, this form demands a very different algorithm than that

of Algorithm 1. As we “read in” the input sentence, we save who in a special
memory location, set aside for that purpose. The remainder of the sentence
is simply written out directly, word for word. Then, when we reach the word
like followed by the end of the sentence, we retrieve the saved who and
literally place it after like—as desired.
Which method most accurately reflects what people actually do? I would
argue that the second algorithm does not mirror what people do when they
process language as accurately as does the representation posited by current
transformational grammar combined with Algorithm 1. Simply put, if repre¬
sentation (c) and Algorithm 1 are assumed, then we should expect a locality
principle like PI. In contrast, there is no such expectation with the memory-
store representation. To the extent that computational Theory 1 —
representation plus Algorithm 1—explains this descriptive fact about human
language, and computational Theory 2 does not, Theory 1 is to be preferred.
The reason for the difference is simple. Recall that Algorithm 1 demands
that operating rules be stated in a literally finite pattern-action format, so
that the rules can be stored in a physically finite machine. But this means
that there can be no rules that move elements in one fell swoop across an
unbounded number of sentences—lest the associated pattern be infinite in
length. Therefore, given that pattern-action rules are literally stored and that
the [t] elements are (properly) inserted by their operation, we would expect
that their radius of action should be strictly limited. That is to say, we would
expect the existence of something like principle PI. But there is no such
distance barrier with the memory-store method, and so there is no reason for
a locality principle. Since the who can be stored away indefinitely, we can
wait forever before putting it back. Therefore, if this method were actually in
use, there should be no reason, at least on grounds of sentence analysis, why
the following sentence should not be permitted:
340 ROBERT C. BERWICK

Who do I believe the claim that Mary likes?

In short, we have used computational methods about how something might

be computed to argue for a particular kind of linguistic theory about what is
represented.
Let me summarize the main point. Assume that what we know about
language includes, for example, knowledge that the underlying form of Who
do / believe Mary likes is

Who do I believe Mary likes [t]?

and that this knowledge is quite literally engaged in computational processes

when sentences are analyzed. These two assumptions, combined with the
functional (computational) demand that people be able to process sentences
of their language efficiently, are sufficient to derive the existence of locality
principles that are actually observed. In contrast, under alternative concep¬
tions of what this knowledge of language looks like and how it is put to use—
the memory-store method—the reason for the existence of such principles
remains a mystery. (Of course, there could be another functional argument
for these principles—in particular, they could make learning easier. But that
would be another story. See Berwick and Weinberg [1983] for a full ac¬
count.)
My tentative conclusion, then, is that rules and representations—the
“stuff” of modern transformational linguistics—are quite real, since by as¬
suming that they are engaged in mental computations, we can actually ex¬
plain part of what we observe to be true about our linguistic behavior. It is
important to note that we achieve this explanation only by marrying specific
linguistic results about what with specific computational techniques about
how. It is time, then, to put to rest the shopworn slogan that linguistics and
computer science are at odds with one another. At least for the study of
human cognition, they are, and ought to be, inseparable.
SECTION 5
 LIBRARY AND
INFORMATION SCIENCES
Disciplinary Differentiation, Competition, and
Convergence

W. Boyd Rayward

The purpose of this paper is to explore some of the relations between librari-
anship, bibliography, documentation, library science, and information sci¬
ence. All are considered part of a historical process that has led to different
ways of envisaging, creating, and investigating the interrelations and relative
effectiveness of formalized modes of access to recorded knowledge.
Underlying this approach are two observations. First, libraries represent
long-established and complex arrangements of a variety of these formalized
modes of access to knowledge. Second, not long after librarianship emerged
as a profession whose practitioners were chiefly absorbed by the adminis¬
trative and operational challenges presented by libraries, there arose a
countervailing resistance to the conceptual narrowness that such a profes¬
sional preoccupation involved. Important as libraries had become, histori¬
cally they represented only a partial solution to the constellation of problems
related to the conservation, organization, and diffusion of recorded knowl¬
edge and information. If libraries were to be improved or if other solutions to
the problems for which they presented partial solutions were to be invented
and tested, then it seemed necessary continually to seek new ways of look¬
ing at libraries themselves and at the broader environment of which they are
only a part. Librarianship, bibliography, documentation, library science,
and information science may be considered as incorporating modes of study
and investigation that not only express subtle occupational distinctions, but
also represent attempts at obtaining these new and increasingly general per¬
spectives.
In the second part of this paper, I examine five developments in the field
that have occurred in the United States in the last ten to fifteen years. I
suggest that these developments, relatively insignificant in themselves, are
important as indicators that librarianship and information science, the latter
arising in part from the documentation movement, have certain formal con-
343
344 W. BOYD RAYWARD

nections. These connections may well imply that there has been a disciplin¬
ary convergence of librarianship toward information science. The develop¬
ments to be examined in this connection are some recent changes in the
names of library schools; the employment of librarians in information cen¬
ters and other less conventional contexts; the growth in the number and the
diversification of the specialized content of journals in librarianship and
information science; the appearance of an important body of new terminol¬
ogy in librarianship; and finally, the formal structural recognition of mutual
interests that has taken place within the major professional associations in
librarianship and information science.
In relating the first part of the paper to the second, I take the somewhat
simplistic view that the prominence of the computer in the last ten or fifteen
years has given librarians themselves a new power of seeing beyond the
conventional boundaries of their thinking and activities traditionally im¬
posed by their concern for individual libraries. Because of the increasing
sophistication and accessibility of computer technology, librarians have
been able to develop library and information services and to organize, man¬
age, and exploit library and information resources in a manner and to an
extent not possible before.
It is also my view that information science—almost a phenomenon of the
computer age—represented in part, at least initially, systematic attempts to
generalize from and to explore in as rigorous a manner as possible questions
and answers that were expressed in what had become accepted library or¬
ganization and practice. This was not at first clearly recognized or at once
accepted.
In the third and final part of this paper, asking a question posed by Don
Swanson [1980/)]—What are the problems libraries were created to solve?—
I try briefly to show that there is a disciplinary continuum between librari¬
anship and information science with no easily identifiable boundary separat¬
ing them, though the difference between the extreme ends of the continuum
are clear and even dramatic. I also suggest that the disciplinary movement
toward more general and inclusive perspectives discussed in the historical
part of this paper may still continue to express itself in a tendency to seek a
redefinition of the objects of study of information science.
Underlying this paper is a simplification, which emerges in various ways,
that should be acknowledged at the outset. I consider the formation of
stable, long-lived scholarly and professional associations and societies as
key events. They are culminations in a developmental process of differentia¬
tion in librarianship, bibliography, documentation, library science, and in¬
formation science that I single out from a complex history that involves
much more than them. On the whole, I view the fields designated by these
appellations in terms of the occupational characteristics, and a related schol¬
arship, of various organized groups of individuals. I focus on some of the
changing structural and organizational expressions of relations between
these groups and the scholarship to which they lay claim, rather than on the
LIBRARY AND INFORMATION SCIENCES 345

content of the scholarship itself. By confining myself to a general discussion

of research and scholarship and to the nature of scholarly and professional
literature in the field, I deliberately eschew discussions of whether, or to
what degree or according to which criteria, library science and information
science can be considered science.

LIBRARIANSHIP

Let us now turn to a historical analysis of the emergence of a group of

related but, at least initially, self-consciously separate disciplinary and pro¬
fessional trends. Librarianship, an ancient occupation, clearly should be the
first of these to be dealt with. Its origins lie close to the origins of writing, to
the discovery of relatively permanent media on which writing could be in¬
scribed, and above all to the social and economic pressures that led to the
preparation, preservation, and retrieval of permanent written records. These
were necessary to processes of stable government and commerce, to orga¬
nized religions with commonly accepted rituals and beliefs, and to the cumu¬
lation and transmission of the knowledge of the times. Thus, we may speak
of librarianship in the ancient world as well as in the modern world.
Nevertheless, it was not until the last quarter of the nineteenth century
that modern librarianship began to emerge as a professional occupation. As
such, it was characterized by the invention of tools, techniques, and organi¬
zational structures that were widely adopted in the then rapidly multiplying
university, college, public, school, and special institutional libraries. Strong
formal associations of librarians were created at this time. Indeed, we may
date the beginning of the professionalization of modern librarianship as an
occupation in the English-speaking world from the foundation of the Ameri¬
can Library Association (ALA) in 1876 and the Library Association (of the
United Kingdom) in 1877. Among the responsibilities of these associations
were the oversight and continuing modification and refinement of codes of
accepted practice and dissemination of commonly held beliefs about the
goals and functions of various kinds of libraries. These associations helped
provide a context for the emergence of an apparatus of systematic profes¬
sional communication: regular conferences, journals, a small monographic
literature in which knowledge in the field was codified, disseminated for use,
and made available for public discussion. Processes of formal professional
education were created, institutionalized, and gradually standardized.

BIBLIOGRAPHY

But within a generation, it became clear that neither libraries nor the existing
organizations of librarians were able to initiate or even contribute usefully to
the study of solutions to certain general bibliographic problems that lay
346 W. BOYD RAYWARD

outside the walls of these individual libraries. This was as true in the United
Kingdom as in the United States. Thus, the creation of the Bibliographical
Society in 1892 was proposed and discussed at the Library Association
meeting in 1891. An issue debated at length was just what the relation of the
Bibliographical Society of Chicago, founded in 1899 (to become the Ameri¬
can Bibliographical Society in 1904) would be with the American Library
Association (ALA); indeed, the fourth general session of the 1902 confer¬
ence of the ALA was placed “in charge” of the Bibliographical Society of
Chicago for a discussion of the founding of a Bibliographical Institute.1
These several bibliographical societies were interested in bibliography as
an intrinsically rewarding area of study that had produced notable achieve¬
ments in its already distinguished history. Bibliography as these societies
conceived it had broad consequences for scholarship quite apart from librar¬
ies. In England, there was in the early days of the Bibliographical Society
the intoxicating idea of such a society creating a great catalog of English
literature and preparing a revision and supplement to Hain’s Repertorium
Bibliographicum. [Copinger, 1892; Ray ward, 1967a.] In both England and
the United States, membership in these early bibliographical societies com¬
prised, certainly at first, many librarians but also, and perhaps predomi¬
nantly as time went by, book collectors, antiquarians, professional literary
and historical scholars, and amateur men of letters who incorporated the
results of their specialized research in arcane papers that were com¬
municated to each other at meetings and in the journals of their societies.
Gradually, interest in what we would today call general issues of biblio¬
graphic control tended to subside in these societies, though never com¬
pletely in the American Bibliographical Society, to be replaced by a schol¬
arly preoccupation with books as physical objects.
The history of books, printing, and publishing, the description of old and
rare books, special collections of books and manuscripts, the implications of
the physical characteristics of codices and books for various disciplines, and
perhaps above all problems in the transmission of texts, such as the relation
of manuscript to printed versions—these gradually became the preoccupa¬
tion of the members of these societies. As specializations of a highly techni¬
cal kind developed, bibliography unadorned became too vague, too much a
portmanteau term to be useful. Nowadays, the scholarly activity of bibliog¬
raphers is labeled descriptive, historical, analytical, or critical bibliography
according to particular, perhaps not precisely or consistently distinguished,
criteria of interest, affiliation, and technique.
Bibliographers asserted the overall independence of their own area of
intellectual endeavor in new journals, among which The Library in Great
Britain, the Papers of the Bibliographical Society of America in the United
States, or, later, the Proceedings of the Bibliographical Society of Virginia
are important. Such journals help to define the boundaries between biblio-

' Yearbook of the Bibliographical Society of Chicago 1901-1902, p. 26.

LIBRARY AND INFORMATION SCIENCES 347

graphical scholarship and librarianship, and the continuing existence of the

journals asserts their continuing importance and separateness. The early
specialized bibliographical journals very soon abandoned discussions of gen¬
eral library problems and issues and even bibliography itself broadly con¬
ceived; increasingly they reported the narrowly focused and highly special¬
ized scholarship previously described.
At the turn of the nineteenth century, then, librarianship was differ¬
entiated from a certain kind of variously designated bibliographic scholar¬
ship, though some librarians have always maintained a close contact with it.
Indeed, nowadays, aspects of it are sometimes studied in library schools.
Nevertheless, differences in orientation between librarians and bibliog¬
raphers became clear very early. The professional interests of librarians
tended to be of a practical kind, firmly focused on the functions and opera¬
tional requirements of libraries, the organizations within which librarians
were employed. Bibliographers, however, tended to be unconstrained by
such organizations and cultivated a general academic orientation toward
books as complex cultural, intellectual artifacts. This difference in orienta¬
tion was reflected, for example, in this remark made at the time the transfor¬
mation of the Chicago Bibliographical Society into a national body was
contemplated: “Only a limited number of librarians of the country are likely
to become members of any organization whose aim is other than a purely
practical one. It has also been pointed out it would be more difficult, if not
well nigh impossible, to bring the scholars of the country into membership in
a section of a popular organization like the American Library Association.

DOCUMENTATION IN EUROPE

A different view of bibliography, one that resembles our notion of biblio¬

graphic control, was held at the turn of the century by Paul Otlet and his
colleagues in Brussels. They had founded the International Institute of Biblio¬
graphy in 1895 to create a universal bibliographical catalog organized on
new lines. Eventually Otlet developed his idea of bibliography in a series of
now neglected but seminal papers into what he called “documentation. His
concern was not to limit bibliographical control to cataloging books in local
institutions, but to find a way of creating a universal index of all documents
that constituted records of knowledge; books, yes, but also parts of books,
journal articles, brochures, industrial catalogs, patents, certain kinds ot ad¬
ministrative records of governments, the archives of municipalities, photo¬
graphs, post cards, newspapers. The problem was to find out what each
contributed to the sum of knowledge, what each contained of potentially

2 Yearbook of the Bibliographical Society of Chicago 1901-1902, p. 48.

348 W. BOYD RAYWARD

useful information, and to express and connect this flexibly, creatively to

what already existed. Otlet had a clear idea of what he wanted from libraries
in the new bibliographical order he was attempting to create. He believed
that if the bibliographical apparatus or equipment of individual libraries were
inproved, better books could be created. Thus would libraries actively con¬
tribute to the advancement of knowledge. Otlet wanted libraries to cease
being mere depositories and to become vital institutes of documentation
providing special information services on all matters of interest to all mem¬
bers of the public who might wish to use them. He wanted to see all libraries
transformed into what he called offices of documentation and these to be
joined together in a national and international system of communication for
sharing all existing resources, an international documentary network. He
envisaged an important role for microfilm in this. (His earliest paper on
microfilm, libraries, and scholarship was published in 1906. [Otlet and
Goldschmidt, 1906.]) A central preoccupation of his voluminous writing was
the delineation of documentation as a separate and developing field of study
that might ultimately lead to a transformation in the ways in which informa¬
tion was generated, recorded, transmitted, organized, and used. [Otlet,
1903; Ray ward, 1975.]
In the view of Otlet and his colleagues, though general bibliographic
organization and control—what they tended to designate as the new field of
documentation—encompassed traditional librarianship, it went far beyond
it. Traditionally, libraries were concerned with only a portion of the docu¬
mentary record; for documentalists, the whole of the record lay within their
scope. The methods to be used to exploit the record, and the institutional
settings in which this work could take place, they believed could be more
various and innovative than conventional libraries.
Librarians reacted to the apparent competition and threat of assimilation
by what was much touted as a new discipline essentially by denial. In conti¬
nental Europe, just after the first World War, social relations between librar¬
ians and documentalists broke down; they have never been fully repaired,
[de Costa, 1982; Rayward, 1967/?; 1977/?.] In England, there was some mild
general interest by librarians before the war in continental European docu¬
mentalists and their work. This became stronger after the war, with the
foundation in 1927 of an English member-organization of the International
Institute of Bibliography: the British Society for International Bibliography.
The English special library movement, which culminated in the foundation
of the Association of Special Libraries and Information Bureaus (ASLIB) in
1924, shared many interests with British documentalists, and in 1937, the
British Society for International Bibliography was absorbed by ASLIB.
Thus, in England, the documentation and special-library movements were
mutually reinforcing, had an early accommodation, and operated quite inde¬
pendently of the more traditional interests and activities of the Library
Association.
LIBRARY AND INFORMATION SCIENCES 349

LIBRARY SCIENCE

In the United States, the situation was a little more complex. Here, dissatis¬
faction with traditional practices of librarianship led not merely to what
might be called academic bibliographic scholarship, on the one hand, and to
the special-library movement with its interest in the special information
needs of business and industry (which I neglect in this paper [see Johns,
1968]), on the other; it also led to the creation of what was called library
science.
Appalled by the absence of rigorous research useful to libraries, C. C.
Williamson in 1930 asked “Can it be that there are no problems in library
service that call for scientific research? Nothing more to learn? No unsolved
problems?” (Williamson, 1931, p. 9.) And he quoted and slightly modified
John Dewey in describing how a library science might emerge:

There is no subject matter intrinsically marked off, earmarked so to say, as the

content of library science. Any methods and any facts and principles from any
subject whatsoever that enable the problems of administration and service to
be dealt with in a bettered way are pertinent. ... It may be doubted with
reference to some aspect or other of library service there is any organized body
of knowledge that may not need to be drawn upon to become a source of
library science . . . .” (Ibid., p. 12.)

In effect, the idea of a library science dates from the late 1920s and
specifically from the founding of the Graduate Library School at the Univer¬
sity of Chicago. This school was established in order to bring the best stan¬
dards of higher education to bear on the education of librarians and, through
the introduction of the PhD degree and the work of the school’s faculty, to
promote research in the field. In a statement of policy for the school, Doug¬
las Waples listed what the faculty considered to be the distinctive functions
of the school:

1. The most important single responsibility of the school is to meet the stan¬
dards of scholarship and research maintained by other graduate departments of
the university, both in the character of work undertaken by the staff and by the
research interests of its graduates. 2. The major aim is research, defined as
“extending the existing body of factual knowledge concerning the values and
procedures of libraries and their many aspects, and including the development
of methods of investigation whereby significant data are obtained, tested, and
applied. . . .” 6. Not all of the studies undertaken by the school need be
confined to research in its restricted meaning of “search for abstract princi¬
ples.” In many instances, they may more properly be called service studies,
studies intended to increase the effectiveness of library service. ... 9. An
important function of the school is the preparation, collection, and publication
of monographs whereby the results of significant studies are made available to
the library professions. (Waples, 1931, pp. 26-27.)
350 W. BOYD RAYWARD

It is not by chance, either, that a minor classic of librarianship was written

by an early member of the faculty of the Graduate Library School—Pierce
Butler’s Introduction to Library Science. [Butler, 1933.]
Certainly, there was an attempt at the Graduate Library School to find
new perspectives from which to study libraries and their problems. These
perspectives were provided by the rigorous, academically respectable
methods of the social sciences and historical scholarship. These methods
were introduced to challenge the existing reliance on shared practical experi¬
ence as the basis for the field’s professional knowledge, with all that this
reliance implied of a stultifying orthodoxy of dogma and belief. The impor¬
tance in the United States of the Graduate Library School for the profes¬
sionalization of—and the appropriate institutionalization of an educational
apparatus in the nation’s universities for—librarianship has been widely
acknowledged. [Carroll, 1970, pp. 50-56, 264-266; White, 1976, pp. 231—
237; Churchwell, 1975, pp. 60-61, 98-101.]
The Graduate Library School embodied what was for a time a unique and
controversial approach to professional education; its professors and stu¬
dents steadily produced a formidable body of research where before practi¬
cally none existed. The school issued the first scholarly journal in the field,
The Library Quarterly, and graduated those who became influential librari¬
ans or faculty members in other library schools. It was not until the late
1940s that two other library schools introduced doctoral studies in librari¬
anship—the Universities of Illinois and Michigan. The 1950s saw a fairly
rapid increase in doctoral programs. These programs by definition empha¬
sized the importance of research, as opposed to training practitioners, and
increased the volume and quality of the field’s literature by reporting this
research. Thus, the idea of a library science is hardly more than fifty years
old, and any great volume of research in the field is relatively recent.
It should be noted that the approach to creating a library science outlined
by Williamson some fifty years ago still obtains: It is still to take the substan¬
tive problems of libraries and librarianship and apply to them research
methods that seem appropriate from other disciplines. Much of the research
in the field has employed, and continues to employ, the methods of the social
sciences or is historical in nature. Latterly techniques of systems engineer¬
ing and mathematical modeling or analysis (such as the techniques of opera¬
tions research and the related and sometimes overlapping area of bibliomet-
rics) have proved invaluable in conceptualizing library problems in a new
way and providing powerful new techniques for analyzing research data.
While the idea of a library science is now widely accepted within the
library profession as desirable, it is only a little exaggerated to say that we do
not yet have an extensive, distinctive, coherent, well-integrated or sys¬
tematized and scientifically derived body of theoretical and practical knowl¬
edge in librarianship. This leads to criticism of library science as a science
and to suspicion of librarianship as a profession. [Rayward, 1980/?.] That it
is, however, exaggerated should be acknowledged because of the existence
LIBRARY AND INFORMATION SCIENCES 351

of cumulatively important subsets of library literature in the areas, for ex¬

ample, of systems engineering, operations research, use and user studies,
classification and indexing, management, and measuring effectiveness,
among others.
Nevertheless, it seems clear to me, despite the research that exists and
the academic location of professional education, that the empiricism of
nineteenth-century librarianship—which the new library science was to test
and perhaps ultimately supersede—remains strongly present in the library
profession today. That is to say, much of the librarian’s knowledge con¬
tinues to be arrived at, and quite generally accepted, in a pragmatic way
through library practice and the shared experience of librarians. Our catalog¬
ing codes and major classification systems, for example, are of this nature. It
is probably true that the central core of knowledge underlying librarianship
is still of this empirical kind and that consequently library science has failed
to lead to an extensive and solid corpus of objectively derived, system¬
atically examined and accumulated knowledge. Whatever knowledge of this
kind exists, it may be thought of as peripheral or incidental to the central
business of librarians—continuing to try, against increasing odds, to keep
their organizations running reasonably effectively.

DOCUMENTATION AND INFORMATION SCIENCE

In 1933, Watson Davis proposed that a scientific information institute be set

up to help foster communication among scientists and improve existing
methods of bibliographical control of scientific literature. [W. Davis, 1935;
Schultz and Garwig, 1969.] Such an institute would, toward these goals,
especially exploit the developing technology of microphotography. His pro¬
posal elicited widespread comment at home and from the European docu-
mentalists. As nothing came of his proposals, he created in 1935 a Documen¬
tation Institute within Science Service, an agency for the popularization of
science, of which he was the director. He used the word documentation to
include “all phases of issuance, use and interchange of recorded informa¬
tion.” The word attracted him as a term because it had a wide circulation in
Europe as a result of the work of Otlet and his documentalist colleagues,
while having no particular meanings associated with it in the United States.3
The American Documentation Institute was created in 1937 as a kind of club
to involve—through their directors or other senior officers—a wide range of
libraries, learned societies, and government agencies in microfilm and other
projects at first insufficiently supported and then more or less repudiated by
Science Service.
Though founded in 1937, the American Documentation Institute became a

31 owe much of this information to the unpublished work of Irene Farkas-Conn, a doctoral
student in the University of Chicago Graduate Library School.
352 W. BOYD RAYWARD

professional and scholarly force only with the publication of its journal
American Documentation in 1950 and with the institute’s transformation
into a general-membership organization in 1953. Nevertheless, the early
emphasis on technology, scientific communication and publication, and new
services offered through new kinds of institutions, anticipated later develop¬
ments within the institute. Here, essentially, was an attempt, stimulated by
technological innovation, to place library and library-related problems in the
context of scientific communication.
Vernon Tate’s editorial description of the scope of American Documenta¬
tion makes it clear that he, like Davis, saw the European documentation
movement as at last crossing the Atlantic to the United States and, from his
point of view, finding in the journal a vehicle that would ensure its subse¬
quent development. [Tate, 1950.] The journal was to deal with “The totality
of documentation,” and the definition of documentation given was that
which had been accepted, Tate noted, by the International Federation for
Documentation,4 for which the American Documentation Institute had be¬
come the American affiliate. “The term documentation refers to the crea¬
tion, transmission, collection, classification, and use of ‘documents’; docu¬
ments may be broadly defined as recorded knowledge in any format.” Tate
also went on to observe of the new work being done by documentalists:
“Traditional methods of communicating and recording knowledge are being
studied, reorganized and supplanted and in some cases revolutionized.”
[Ibid., p. 3.]
In the following decades, it became clear that the documentalists consid¬
ered themselves a breed quite apart from librarians. Unlike librarians, they
were not institution-bound. They belonged to a different environment; they
came to documentation from a wide variety of scientific and technical fields.
They were responding exuberantly to powerful social forces. As Jesse Shera
has said of this period: “First of all interest in the problem of information on
the part of the scientific world and the federal government reached an unprec¬
edented level. Coupled with this was a period of accelerated technological
development.” (Shera and Cleveland, 1977, p. 258.) The exploding world of
scientific literature; new communications and information-processing tech¬
nology; mammoth programs of governmental research calling for new sys¬
tems of information organization, storage, and retrieval—these were the
winds and waves of these new mariners, the documentalists, and the infor¬
mation scientists they formally became in 1968, when the American Docu¬
mentation Institute became the American Society for Information Science.
They invented a new language in which to discuss their work. They became
increasingly involved with the computer and mathematics. Machine-stored,
manipulated, and transmitted data inevitably involved a reassessment of
what constituted a document and what documentation was. The computer
raised the possibility of a wide range of indexing and retrieval experiments

4The new name that The International Institute of Bibliography adopted in 1937.
LIBRARY AND INFORMATION SCIENCES 353

hardly possible before. And, indeed, it may well be that retrieval experi¬
ments in the 1960s constitute a first flowering of information science as a
science.
Certainly, information science as a discipline nourishing certain profes¬
sional activities developed rapidly in the 1960s. In 1965, the first Annual
Review of Information Science and Technology appeared. It is of particular
interest because it represents a relatively early survey of the literature of the
field and its segmentation into what appeared at that time to be the accepted
major disciplinary components. These components vividly exemplified the
importance of the computer in giving shape and direction to both discussion
and research in the field. The following chapters were written explicitly in
relation to the computer: “File Organization and Search Techniques,”
“Automated Language Processing,” “New Hardware Developments,”
“Man-Machine Communication,” “Information-System Applications,” and
“Library Automation.” Two chapters were concerned with aspects of in¬
dexing and abstracting: “Content Analysis, Specification and Control, and
“Evaluation of Indexing Systems.” Another chapter was on “Information
Centers and Services,” central to which was a comparison of work carried
out in special libraries; one was on “Professional Aspects of Information
Science and Technology,” in which were examined problems of terminology
(briefly) and developments in educational programs (at length); and finally,
there was a chapter on “National Information Issues and Trends.”
The change of the name of the American Documentation Institute to the
American Society for Information Science was intended to reflect the chang¬
ing orientation of the society’s members. The adoption of the slippery term
information led to many theoretical and frequently tortuous attempts at
definitions. During the decades of the 1960s and the 1970s, there was also
much discussion about what an ideal curriculum for information science
might look like as opposed to those already in existence for librarianship.
Many of the developments of the late 1960s may be interpreted as the
latest culmination of ongoing attempts to create a more general understand¬
ing of how to facilitate access to recorded knowledge than was possible in
librarianship, which was perceived as being limited to creating collections of
books and related library materials and to managing libraries, or to docu¬
mentation, which focused on a more abstract notion of records of knowledge
or documents. In its turn, information science represented an approach to an
even more abstract level of analysis; it included books and documents while
not being limited to them. Information science has permitted the develop¬
ment of research that is more generally cumulative and generalizable, more
readily enriched by importations from other disciplines. It has led to the
development of new options, procedures, and organizational arrangements
for facilitating access to information recorded, or manipulated, in new ways.
Let us say that by 1970 this was the situation. On the one hand was a
vigorous librarianship, its practitioners still blissful in their relatively undis¬
turbed pragmatism, despite the existence of an important scholarly
354 W. BOYD RAYWARD

apparatus for the profession involving, in part, the nation’s graduate library
schools and despite increasing lip service to the importance of research for
the development of their field. Their journal and monographic literature had
become voluminous, though fragmented. They were organized into a num¬
ber of large, well-established, professional associations with a wide range of
competing, sometimes conflicting, professional functions. On the other hand
was the new, equally vigorous discipline of information science, with a
strong research orientation underlying its professional and service compo¬
nents, its own small fairly homogeneous learned society, a generalized field
of study (whose precise boundaries, however, were in constant dispute), and
a recently established bibliographical and educational apparatus. How did
what was a certain combativeness on the part of information scientists and
librarians change, insofar as it has changed, to mutual accommodation?

DISCIPLINARY CONVERGENCE

It is too simple to say that library schools, aware of the limitations of library
science and recognizing that information science might provide a more re¬
spectable disciplinary base for the profession of librarianship, enticed or co¬
opted information scientists as faculty. Or that it was hoped that if a little
information science were added to their library studies, library-school grad¬
uates might find the job market opening up. Or that information scientists
could not really find a more congenial academic home (or parking lot) in
engineering or computer science or mathematics departments than in library
schools. There may be some truth in all of these statements.
I attribute a disciplinary convergence, insofar as it has occurred, to essen¬
tially two factors. The first was the thoughtful recognition by leaders in the
fields of library science and information science that they were committed to
finding solutions to the same general problems, despite differences in ter¬
minology and orientation. The second was the increasing importance of the
computer in libraries; the need to exploit computers has brought about a
convergence in language and orientation discussed in the next part of this
paper.
In the 1950s, Jesse Shera and Margaret Egan, students of the work of
Bradford and the European documentalists, in a series of still valuable pa¬
pers developed new ideas about librarianship, bibliography, and informa¬
tion. [Shera and Egan, 1953.] They spoke of bibliographic organization
[Shera and Egan, 1951], speculated about classification, sought an intellec¬
tual foundation for librarianship and bibliography in a theory of social epis¬
temology [Egan and Shera, 1951], and tried to understand the significance
and implications of the rapidly changing information technology of the
times. [Shera, 1953.] During his tenure at Western Reserve University and
later, Shera was a witty, thoughtful observer of new applications of the
computer to information handling. He was a librarian eager to reconcile
LIBRARY AND INFORMATION SCIENCES 355

what seemed unnecessarily divergent movements; he was a synthesizer, a

gadfly, a theoretician. He held librarianship, documentation, and informa¬
tion science together intellectually in a broad and scholarly perspective and
encouraged their joint institutionalization in the library school at Western
Reserve, of which he became dean. [Shera, 1968.]
In 1963, Don Swanson, a physicist who had become an information scien¬
tist, was appointed dean of the Graduate Library School at Chicago. In
1964, he organized the 29th annual conference of the Graduate Library
School on the subject of the intellectual foundations of library education. In
his introduction to the conference, he observed:

The fields of information science, information technology, information re¬

trieval and documentation will not be treated as separate topics but rather as an
integral part of library science, . . . thus we shall not be concerned with
whether information science is part of library science or vice versa, but rather
we shall assume that the educational process in a graduate library school must
adequately take both into account. (Swanson, 1965, p. 2.)

Swanson then went on to outline a group of substantive questions that might

affect the immediate future of library education and research and that en¬
compassed much of what information science dealt with at the time.
These two instances, and I should stress that many more could be ad¬
duced, are of leaders in the field alive to new possibilities. They were indiffer¬
ent to, or at least undisturbed by, jurisdictional disputes between certain
librarians of the time who were fearful of computers and scientific method
and some information scientists who were disdainful of the pragmatism and
apparent parochialism of librarians. These leaders had a broad, generalized
perspective that allowed them to discern the general but common problems
that underlay librarianship and the new developments in information pro¬
cessing that were sparked by the computer revolution.
The computer eventually gave their arguments overwhelming force in two
ways. Its exploitation in individual libraries led to an almost universal inter¬
est of librarians in systems analysis that required identifying and recording
library operations at a level of detail appropriate for possible computeriza¬
tion. And this inevitably called into question much of the conventional wis¬
dom underlying library organization and administration. [Flood, 1965.] The
development of the Online Computer Library Center (OCLC) and later of
other bibliographic utilities and networks and online indexing search ser¬
vices (Dialog, Bibliographical Retrieval Services (BRS), etc.) required li¬
brarians to move beyond the confines of their own libraries, using a sophis¬
ticated electronic technology. Moreover, it became clear that problems
involved in creating and exploiting these systems both locally and centrally
were of a complexity and magnitude and yet of a solvability previously
unparalleled in librarianship.
Of course, the computer in and of itself was not responsible for these
356 W. BOYD RAYWARD

developments. The exploitation of the computer in developing new institu¬

tional relations between libraries and in improving local library processes and
procedures was encouraged by the growing volume and complexity of mate¬
rials that had to be dealt with and by a fiscal plight that has become increas¬
ingly severe. Nowadays, computer-based systems that have been created as
a response to these pressures have become one of the major foci of develop¬
ments in librarianship. Technical requirements, operational refinement, and
improvement of various systems have brought professional librarians and
professionally active information scientists together in a way that makes
them virtually indistinguishable. Librarianship as librarianship in this con¬
text typically concerns itself with the organizational, political, jurisdictional,
and economic aspects of the problems that these systems create and to
which, in part, they attempt to respond.

DEVELOPMENTS IN LIBRARIANSHIP

Let us now examine five developments that have taken place in librarianship
during the last several decades. These developments support the idea that,
at least from the librarian’s point of view, there is an emergent disciplinary
integration between librarianship and information science, the historical di¬
vergence and subsequent accommodation of which were discussed in the
preceding section of this paper.

Name Changes

First, and perhaps most superficial, are changes in the way library schools
name themselves. Recently, the Library School at the University of Illinois
changed its name to Graduate School of Library and Information Science.
At Denver, the school became the Graduate School of Librarianship and
Information Management. At Brigham Young University, the school is
called the School of Library and Information Sciences. At Berkeley, it is the
School of Library and Information Studies, while at Chicago, though a range
of information science courses is taught and the school is reported as one of
the first to attempt the systematic integration of a library-and-information
science curriculum [Taylor, 1966], the school continues to be called the
Graduate Library School. Despite these differing appellations, collectively,
we still tend to call all of these schools library schools, but that is no longer
precisely accurate in terms of their interests and scope and how they wish to
present themselves to the rest of academe. The importance of the name
changes lies in what they represent—a shift of emphasis from libraries to
something more general. The curriculum in most of these schools now in¬
cludes courses dealing with computers and computer programming; data¬
base creation, management, and development; information storage and
searching more generally; and behavioral characteristics of various groups
and kinds of information producers and users.
LIBRARY AND INFORMATION SCIENCES 357

Employment Trends

The changes in the names of library schools reflect, in part, another contem¬
porary trend. The graduates of our library schools as well as becoming
librarians now frequently go into information processing or information
management, and they are called information specialists, analysts, or con¬
sultants. This is the second development I wish to mention. To conventional
library jobs, librarians now sometimes add aspects of records management,
special indexing services, media management, and research analysis, and
they often work in what is called an information center. It has become fairly
generally accepted that there is a wide range of professional library and
librarylike work that library-school graduates can perform successfully out¬
side conventional library settings. [Sellen, 1980.]

Journal Proliferation

The next change I want to mention is the enormous proliferation of journals

in the library and information science fields. Some of these journals have
gone through interesting name changes: Information Storage and Retrieval
has become Information Processing and Management; American Documen¬
tation has become the Journal of the American Society for Information
Science; UNESCO Bulletin for Libraries has become the UNESCO Journal
of Information Science, Librarianship and Archives Administration. One
group of new journals is clearly library related: Collection Management, The
Serials Librarian, The Behavioral and Social Sciences Librarian, Public
Library Quarterly, Library Research, and the Journal of Academic Librari¬
anship. Other journals, of no less interest to librarians, have in their titles the
new language of information work: On-Line, On-Line Review, Program, and
Data Base.
This phenomenon of a rapidly growing number of journals, insofar as it
does not represent some form of collective madness, given the economic
conditions of the times, must mean something. We can argue that the phe¬
nomenon is, in part, the result of the commercial exploitation of institutional
pressure on librarians and library-school faculty to increase their publishing
activities in order to achieve academic and professional respectability. But
we may also argue that this literature suggests that a closely related range of
subspecializations has developed and, in librarianship especially, achieved a
certain maturity—a range that progresses from matters of quite conven¬
tional librarianship through key areas in information science. Journals that
tend to draw the boundaries of the fields as a whole, such as the Library
Quarterly or the Journal of the American Society for Information Science,
include articles over the whole range of issues and problems dealt with by
separate journals, though we assume that there are different frequency dis¬
tributions of papers across the subspecializations included in the general
journals. Neither the field of librarianship nor information science, given
this variety and overlapping of journals, can be perceived as unitary and
358 W. BOYD RAYWARD

strictly exclusive of the other. Rather, each is an agglomeration of parts,

many of which may be regarded as common to both.

Terminology

The new language of librarianship is important because it is a terminological

reflection of all the changes that have been occurring. It is this body of
relatively new terminology that constitutes the fourth change that I wish to
identify. Fifteen years ago, when we spoke of a catalog, we meant some¬
times a book but usually a card catalog. We looked to the Library of Con¬
gress to supply catalog copy in the form of cards or proof sheets. Now we
speak, in addition, of COM (Computer Output Microform) and online
public-access catalogs, of MARC (Machine Readable Cataloging) data dis¬
seminated on magnetic tapes. In the mid-1960s, new technology consisted of
keypunch machines, sorters, printers, and a monstrous engine called a com¬
puter, to which indirect access was had in batch mode. Nowadays, there is
an increasingly large group of librarians who can speak with some familiarity
of CPUs (Central Processing Units) and peripherals; mainframe, mini- and
microcomputers; Baud rates and memory capacity; hard discs and floppy
disks; and bubble memory. Reference librarians, sensitive to the importance
of understanding database structures and search strategies because of new
kinds of cost considerations, are expected routinely to conduct what is
variously called online or machine-assisted searching of bibliographic and
nonbibliographic databases, access to which is purchased from database
vendors in dial-up mode using the Tymnet or Telenet communications sys¬
tems. Other librarians buy software and hardware and turnkey systems from
commercial vendors who are part of something we now designate as the
information industry. Once, cooperative schemes led to shared storage facil¬
ities and union catalogs that were kept up to date and reasonably complete
only with enormous difficulty and at a prohibitive cost. Now, we have biblio¬
graphic utilities and networks for which the Union Catalog function is but
one, and that more or less incidental, of a great range of capabilities. The
new language is essentially derived from computer applications in libraries
and, as a consequence, is frequently the same as, or at least similar to, that
used by information scientists, for whom, historically it can be argued, the
computer provided a major disciplinary impetus. The shared language has
helped identify shared technical and professional problems.

Structure of Associations

The final development to which I wish to refer is the structural recognition

within the formal professional association in both librarianship and informa¬
tion science of major areas of both common and divergent interests. As early
LIBRARY AND INFORMATION SCIENCES 359

as 1951, the Special Libraries Association established a Documentation Di¬

vision; this division is now the Information Technology Division. The
Special Libraries Association has a wide range of specific subject divisions
and a Publishing Division as well. In 1962 the American Library Association
created an Interdivisional Committee on Documentation, which became, in
1966, the Information Science and Automation Division, now the Library
and Information Technology Division. For most of its life, the Resources
and Technical Services Division has had a Section for Classification. In
1968, when the American Society for Information Science set up Special
Interest Groups (SIGs), one of the first was for Library Automation and
Networks and another for Classification Research. Some of the Special In¬
terest Groups, particularly in scientific and technical subject areas, paral¬
leled subject divisions in the Special Libraries Association. Most recently,
while the number and configuration of subject areas that are the object of
Special Interest Groups have changed, the Library Automation and Net¬
work Group and the Classification Research Group have been maintained.
Among new groups are those for Community Information Services, Com¬
puterized Retrieval Services, Nonprint Media and Reprographics, and Infor¬
mation Generation and Publishing. These all have counterparts within li¬
brary associations. The Public Library Association of the American Library
Association, for example, has a Community Information Section; the Refer¬
ence and Adult Services Division has a Section for Machine Assisted Refer¬
ence (also called Computerized Search Services); the Resources and Techni¬
cal Services Division has a Reproduction of Library Materials Section; the
Library Information Technology Division has sections for Information Sci¬
ence and Automation (the old name of the division), Audio Visual Materials,
and Video and Cable Communications. These all reflect developing areas of
mutual interest between librarians as an occupational group and information
scientists.
The formal associations also reveal areas that do not overlap in so direct
or important a way. The differences are clearest where library associations
focus on quite specific library matters. These are represented structurally
within the American Library Association, for example, by the type-of-
library divisions. On the other hand, the information science association
seems most different when it focuses on information and information pro¬
cessing in a most general way: represented structurally within the American
Society for Information Science, for example, by special interest groups on
numeric databases, the foundations of information science, and information
analysis or evaluation. But librarians increasingly consider that they must be
able to use effectively nonbibliographic databases; that the foundations of
information science must inevitably encompass recorded information and its
“potentiation” for society in organizations such as libraries. Information
analysis or evaluation may well deal in a generalized way with issues that a
large literature in evaluating library and information services has already
attempted to deal with.
360 W. BOYD RAYWARD

A DISCIPLINARY CONTINUUM

In the preceding sections of this paper, I discussed first the emergence of

four major disciplinary trends that represent, in the case of documentation
and information science especially, attempts to find an increasingly general
perspective from which to conceptualize, investigate, and find new and im¬
proved operational responses to problems that underlie libraries and librari-
anship. Next, I described certain contemporary phenomena that suggest a
convergence of librarianship toward what has become known as information
science. In the third section of the paper, I want to take a more speculative
approach to the relation between the definition of librarianship and informa¬
tion science.
As a point of departure, it is useful to examine the question posed by
Swanson: What are the problems libraries are intended to solve? [Swanson,
1980a.] An immediate implication of this question is that, despite their antiq¬
uity, libraries may not be the only, the best, or even a lasting solution to
these problems. Furthermore, if there are other solutions imaginable, it is
important to be able to assess their relative effectiveness in the endless
search for an optimal configuration of solutions.
Libraries may be regarded as a major institutionalized response to the
problems of providing generalized access to the record of what is known.
They are a complex organization of access mechanisms. A collection of
books and other traditional library materials organized according to the
conventions of librarianship by a specially trained and deployed staff for the
use of a defined clientele—this is a typical way in which libraries provide
access to a portion of the record of knowledge. The explicit or implicit
definition of a clientele provides boundaries as to what will be collected and
how what is collected will be organized and made available. In addition, an
individual library usually provides a range of staff, a variety of bibliographic
and other tools, and creates formalized procedures for gaining both biblio¬
graphic access and sometimes physical access to that portion and form of the
documentary record that is not directly incorporated into the library.
As time has gone by, the ways in which what is known has been recorded,
on the one hand, and what constitutes effective access to this record, on the
other, have changed. As well as the variety of forms of printed record, there
are now other formats in which knowledge or information is recorded; per¬
haps the latest of these are electronic. Access is required not merely to
conventional library materials or to other formats of the record, but to what
they contain as well, to their components, or elemental parts that can be
described as facts, ideas, data, information. In some circumstances, access
is provided not merely on demand but in anticipation of demand through
selective dissemination of information (SDI) services. Furthermore, access
is also seen as increasingly desirable in a limited number of areas of knowl¬
edge to the restructured content of what has been recorded in these areas.
The need to provide this kind of access has led to setting up information-
LIBRARY AND INFORMATION SCIENCES 361

analysis centers and social-science-data archives, for example, and em¬

ploying research analysts. This kind of access has, as a rule, required the
participation of kinds of personnel different from those traditionally em¬
ployed professionally in libraries; special training in the disciplines involved
and in computer-based statistical procedures have been thought to be neces¬
sary. Schools of library-and-information science nowadays, however, can
and do provide some of this training.
As the volume, complexity, and variety of formats of the record of knowl¬
edge or recorded information have increased and the number of users and
the range of potential requirements of access to this record have expanded, a
variety of responses to these circumstances has occurred. Some of the re¬
sponses have been made commercially by what we now call the information
industry, which creates and sells bibliographic services and products to
libraries, which constitute a major segment of their market, and also inde¬
pendently of libraries. Libraries have been organized locally, regionally, and
nationally into an enormously various range of configurations of systems,
networks, and consortia. Developments in telecommunications capabilities
have been seized on as facilitating new kinds of institutional arrangements
for providing different levels of access both through libraries and outside of
them, as in the experimental home-based VIEWDATA or PRESTEL sys¬
tems. New kinds of organizations have been created to provide new kinds of
access, such as information centers and data archives.
A narrow and conventional definition of a library expresses a limited view
of the format of the records involved and modes of access provided to them.
But if these limitations are abandoned, libraries may be thought to embrace
not only the arrangements and services of traditional libraries but of infor¬
mation centers and data archives as well. Traditional libraries fall into place
on a historical continuum that has arisen as institutionalized responses to the
need to provide access to information in recorded form have been devel¬
oped, examined, and modified.
Such an approach to a definition may seem too broad to be useful. It
includes bookstores and office-filing and records-management systems. It
may be argued that it extends to genetics and incorporates museums. While
bookstores, office-filing and records-management systems involve what may
be described as librarylike documents, they have in the past tended to fall
outside the scope of librarianship. In none of these cases were the systems
for, and modes of, access particularly complex and interesting. In book¬
stores, in addition, access is consummated only on a basis that librarians
have tended to repudiate (except in the case of the rental collections some
public libraries still maintain)—the payment of a fee. Nevertheless, as book
jobbers, for example, have developed extensive librarylike systems essen¬
tially for the use of libraries, book jobbers have become less like bookstores
and more like the libraries they serve. And insofar as in hospitals, law firms,
and other institutional settings, filing and records systems have grown in
scope and complexity and in the sophistication of the technology through
362 W. BOYD RAYWARD

which they are maintained, they have begun to attract the interest of librari¬
ans and information scientists.
It is not implausible, either, to think of a museum as a library of artifacts.
Suzanne Briet, writing from the point of view of the European documental-
ists, espouses a definition of documentation almost as broad as this. “An
antelope that has been ‘cataloged’ [in a zoo] is a primary document and other
documents are secondary and derivative.” (Briet, 1951, p. 8.) It seems more
clearly figurative to speak of chromosomes as libraries of genetic informa¬
tion: Certainly neither man nor nature promises generalized access to the
information they contain, despite the advances that have been made in
genetic engineering.
But it is at this point of seeking appropriate definitions for libraries, docu¬
mentation, and information science and of exploring their ramifications that
we may suggest the continued relevance of the historical development ex¬
plored earlier by asking the question: What are the phenomena that informa¬
tion science studies? The preceding discussion has focused on a changing
understanding of the scope and limits of a professional occupation that at¬
tempts to provide information services through relatively permanent and
formal structures. What underlies these changes has been an increasingly
complex and sophisticated view of what constituted recorded information.
First came books; they are documents, and documents contain information,
and access to information is provided for by the systematic manipulation of
the contents of documents and books. Despite all the unknowns, such rela¬
tions are relatively clear, and the study of the processes involved and the
larger environment in which they take place are relatively well cir¬
cumscribed.
When, however, the focus becomes not library and information work, but
information science, the clarity, such as it is, is lost. Information becomes a
phenomenon for scientific investigation, and the contexts within which it is
examined and the investigative techniques to be employed depend very
largely on how information is defined. Information as electrical impulses in
the brain or as represented by certain biochemical phenomena or as the
content of disciplines—this set of different meanings raises a serious prob¬
lem of disciplinary encroachment. When is information science information
science and not epistemology, psychology, biopsychology, physiology or
physics or journalism or the sociology of groups or anthropology?
Thus, an ultimate consequence of the process of seeking an ever more
general perspective for information science, for the problems of access to
recorded and potentially consultable or public information underlying the
creation and development of libraries, may be the loss of all sense of disci¬
plinary coherence in the field. We might argue from the point of view of the
continuum I described earlier that at one end is the librarian constructing a
catalog of books; at the other, because of the way in which information can
now be conceived, is the genetic engineer consulting the information re¬
corded in and then restructuring DNA. Perhaps, however, there is a discon-
LIBRARY AND INFORMATION SCIENCES 363

tinuity between the two involving notions of symbolic representation and

public consultation that more careful analysis and definition would reveal.
The point here is not so much the difficulty of definition or the slipperiness
of terminology as the implication that the generalizing process described
early in this paper may still be at work. It is a source of innovation and
strength, but it blurs traditional distinctions and can unsettle professional
convictions.
 KNOWLEDGE AND PRACTICE
IN LIBRARY AND
INFORMATION SERVICES

David Batty and Toni Carbo Bearman

Librarianship, as Boyd Rayward says, is an ancient profession, as old as

Babylon. But the government of society and its commerce has depended on
the collection and use of information for just as long—from Babylon through
the Domesday Book to the United States Census and beyond. Is not then
information science equally ancient? We know about the Babylonian librar¬
ies by their catalogs; we know about the use of information in government
and commerce from the existence of inventories. The principal characteriz¬
ing principle of the organization of information seems to be list-making; that
is, we might say that the kind of list-making characterizes the kind of general
information activity. This thesis will be addressed passim in the paper.
Rayward quite rightly describes the recent convergence of hitherto dis¬
parate disciplines or practices, but in referring to library and information
science and bibliographers and documentalists, he seems to present a pic¬
ture that does not always do justice to the complexity of the nature of the
component forces. Further, his historical backdrop, with 1876 as the begin¬
ning of formal librarianship and the founding of the Graduate Library School
in Chicago as the beginning of research in library science, does not really
allow a large enough stage to accommodate the true range of significant
characters.
In short, while we agree with Rayward's general thesis, we would like to
extend it, with reference to some earlier history and other phenomena in the
last century. We would like to begin by making a few formal points:

There was a long-standing library tradition before 1876, admittedly (un¬

ashamedly?) pragmatic, of academic libraries, and later, subscription and
circulating libraries. Those in charge of libraries like the Smithsonian
(Jewett), the British Museum (Edwards, Panizzi), and the Boston
Athenaeum (Cutter) were learned and articulate.
There was a long-standing bibliographical tradition, which began in the
365
366 DAVID BATTY AND TONI CARBO BEARMAN

late renaissance, was cultivated in the eighteenth century, flourished in

the nineteenth century, and continues today. But the concern of classical
bibliographers (for example, McKerrow) has always been for scholarly
description of books in such detail that individual copies could be
identified.
Scientists in the late nineteenth century quite outside either bibliographic
or library traditions were already compiling indexes and catalogs of
scientific material, because neither librarians nor bibliographers provided
them. These were the protodocumentalists—they were not bibliog¬
raphers.
From the eighteenth century onward, new libraries and new library prac¬
tices veered between the bibliographic and documentalist traditions, as
they tried, on the one hand, to be scholarly in their catalog descriptions
and, on the other, to provide easily used “finding-tools” for their patrons.

Rayward is quite right when he says that “[librarianship’s] origins lie

close to the origins of writing.” However, we should remember that the
earliest writings were inventories—of property, goods, and laws in Babylon,
Mycenae, and so forth. Early library catalogs were only another form of
inventory—even the celebrated Alexandrian catalog of Callimachus. Simi¬
larly, in the Middle Ages, monasteries kept inventories of their collections
and sometimes used them as catalogs. Even as late as the end of the eigh¬
teenth century, libraries kept an internal inventory and perhaps an internal
slip catalog arranged by author/title and occasionally published a meager
author/title entry catalog for the library patrons. For example, between 1667
and 1840, even Harvard College produced a public catalog only three times.
[Weber, 1964.]
The library catalog as we know it was really the product of the book trade:
Aldus Manutius and his contemporaries in the late fifteenth century for
bibliographic description; Andrew Maunsell and the English booksellers of
the early seventeenth century for the subject catalog; and the early eigh¬
teenth-century Paris booksellers like Marchand and Brunet, for the idea of a
general bibliographic classification.
We should not assume, however, that booksellers gave rise to the biblio¬
graphic tradition. Their aim was to sell books, which they did successfully
by grouping similar books and subjects together in their catalogs and de¬
scribing them explicitly and practically to their customers. On the other
hand, the classical bibliographer’s delight is to record the diversity and
differences of books in order to identify individual copies. Indeed, Rayward
says, “These several bibliographical societies were interested in bibliogra¬
phy as an intrinsically rewarding area of study that had . . . broad conse¬
quences for scholarship quite apart from libraries.” Of course they were
interested—the eighteenth century had been a century of gentleman’s
individual scholarship. Now there was a hope of institutionalism—perhaps
even a society like the Royal Society.
LIBRARY AND INFORMATION SERVICES 367

Ray ward asserts that “we may date the beginning of . . . modern librari-
anship ... in the English-speaking world from the foundation of the Ameri¬
can Library Association (ALA) in 1876 and the Library Association (of the
United Kingdom) in 1877. . . . But within a generation, it became clear that
neither libraries nor the existing organizations of librarians were able to
initiate or even contribute usefully to the study of solutions to certain gen¬
eral bibliographic problems that lay outside the walls of these individual
libraries. This was as true in the United Kingdom as in the United States.”
This is an almost Hegelian interpretation of library history. Lor the Amer¬
ican Library Association to be formed in 1876 meant the coming together of
a large number of like-minded professionals. What had Charles Jewett,
Charles Cutter, Justin Winsor, and Melvil Dewey been doing in the years
before 1876? What had Edward Edwards, Antonio Panizzi, and Andrea
Crestadoro been doing in England before 1877? Britain had passed a Public
Libraries Act nearly thirty years before. The years 1876/1877 do not mark
the beginning of librarianship; they mark its formalization in two national
associations after at least a generation of practice. The international confer¬
ence of 1877 was a declaration of the existence of a solid body of knowledge
and experience.
Jewett had already laid out in the 1840s the three factors essential for
solving the problem of universal bibliography: access to material, reproduc¬
tion technology, and consistent descriptive practice. [Jewett, 1851.] Jewett,
and later, Dewey and Cutter, tried to solve the problem from within the
library world. Ray ward ignores the very real efforts made toward the stan¬
dardization of bibliographic description by librarians like Panizzi, Jewett,
Cutter, and others.
It is true that conventional libraries did not produce a solution to general
bibliographic problems in the last quarter of the nineteenth century—nor did
bibliographers. The reason was not a failure due to any one part of our
general profession, but the effect of historical development—Jewett’s three
problems were never all solved satisfactorily at the same time.
Indeed, as the bibliographers delved even more into the scholarly
minutiae of the differences between copies of early printed books, they
inhibited the production of large-scale scholarly catalogs. It fell to the very
different documentalists (sometimes called scientific bibliographers, to add
to the confusion) to try to collect and organize all the citations in the
scientific literature.
Bibliographic control has never been a central interest of classical bibliog¬
raphy. Classical bibliographers are concerned with differences between
copies of a published work, whereas librarians are concerned with
similarities in order to collect them into subject groups on shelves and into
author/title and subject groups in catalogs and indexes. Documentalists have
always been concerned with specific detail but have customarily used
librarylike techniques to organize their material. They are on the other side of
the librarian from the bibliographer but perhaps not so far away.
368 DAVID BATTY AND TONI CARBO BEARMAN

As Ray ward says,

In the view of Otlet and his colleagues, though general bibliographic organiza¬
tion and control—what they tended to designate as the new field of documenta¬
tion—encompassed traditional librarianship, it went far beyond it. Tradi¬
tionally, libraries were concerned with only a portion of the documentary
record; for the documentalists, the whole of the record lay within their scope.
(Rayward in his paper in this volume.)

We should remember that, although Otlet certainly addressed general

bibliographic control in 1895, so had Melvil Dewey, as evidenced in the
preface to the first edition of his classification scheme in 1876. [Dewey,
1876.] Actually the problem had been that librarians, influenced by classical
bibliographers, wanted to include a kind of formalized record in the manner
of the bibliographers. This had begun under the influence of Panizzi’s cry for
“a full and accurate catalog” [Report of the Royal Commission, 1850] and,
by 1908, with the joint Anglo-American code [Cataloguing Rules, 1908], was
somewhat divorced from the indexing tradition of Poole s Index [Poole,
1882], and so forth. Early documentalists wanted comprehensive coverage,
sufficient (i.e., perhaps minimum) record, and precise location.
Unfortunately, in his discussion of library science, Rayward neglects spe¬
cial libraries, the true inheritors of the documentalist tradition. In the United
Kingdom at least, in the 1920s/1930s (precisely the period Rayward ad¬
dresses), there were strong forces binding together the disparate elements of
scientific and technological information and formal (mostly public and spe¬
cial) libraries. Regional networks like Sheffield Interchange Organization
(SINTO), Liverpool and District Scientific, Industrial, and Research Librar¬
ies Advisory Council (LADSIRLAC), and so forth, were based in public
libraries (with the public library seen as a kind of honest broker) and linked
all kinds of library and information units for the exchange of nonbiblio-
graphic as well as bibliographic information. To understand the true
significance of these networks, and the position of the public library within
them, the reader should realize that for at least the first half of this century,
public libraries in the United Kingdom represented the most powerful sector
of the library profession. It was not unusual to find city reference libraries
richer in their collections and more active in their reference service than
university libraries in the same cities.
Rayward’s suggestion that the formation of a theory of librarianship and
library research began with the founding of the Graduate Library School in
Chicago seems less cause and effect than a post hoc ergo propter hoc argu¬
ment. There had been many scholars, thinkers, and workers concerned with
many aspects of library science between 1840 and 1940, most of whom owed
no debt to the Graduate Library School. It is true that Jewett, Cutter, Bliss,
Sayers, and Ranganathan addressed individual aspects (although Cutter and
Ranganathan came close to addressing the whole field). It is probably true to
LIBRARY AND INFORMATION SERVICES 369

say that they did not develop a theory of librarianship—but we wonder if

one has ever been enunciated. It is true also that much of the work in the late
nineteenth century addressed practical aspects of library science—but there
are many fields in which research customarily addresses practical aspects of
the discipline.
We have mixed feelings about some of the customary attitudes toward or
definitions of research especially in library-and-information science. Li-
brary-and-information science is in an uncomfortable position vis-a-vis re¬
search for two reasons: The discipline is structural rather than substantive,
and (partly for this reason and partly for historical reasons), it never
amassed, or had time to amass, a solid quantitative base of empirical obser¬
vations in a period of its development early enough to avoid doubts being
cast on that kind of research or observation. Librarians and information
scientists very soon found themselves alongside researchers in other disci¬
plines who had that quantitative base and who were now exploring theoret¬
ical implications. The profession may do itself a disservice if it seeks to
emphasize only “respectable” research. Ray ward alludes somewhat dis¬
paragingly to the empiricism of the nineteenth century—yet, like medicine
and law, we are a profession with a job to do. We hope to improve by
learning from research, but we can also improve through carefully consid¬
ered practice.
The best hope we have is precisely the tenor of Ray ward’s paper: That we
are now seeing the convergence of closely related bodies of knowledge and
practice that for long, like siblings with territorial problems, have viewed
each other at best with suspicion and at worst with outright paranoia. Infor¬
mation has already become the late twentieth century’s biggest industry.
With a clear view of the nature, origin, and potential of the components (that
may include ourselves), we may see our way to the best resolution.
 LIBRARY SCIENCE AND
INFORMATION SCIENCE
Broad or Narrow?

Manfred Kochen

It is tempting to begin with a discussion of the nature of information and

some elementary distinctions. The disciplines of information are still in too
early a stage of development for any kind of consensus on this question to
exist. Yet, it may help for the purpose of discussion to offer at least my own
views as working definitions. To those of us who were educated in mathe¬
matics, physics, chemistry, mathematical social sciences, or engineering
since the middle of this century, the concept of information was associated
with entropy, order, energy, organization, and control. It aroused our
scientific curiosity and excited our imagination as being a concept at the
frontier of the unknown. We thought its explication might for the first time
unlock some deep secret of nature—how organized systems are formed and
maintain themselves—and shed light on tantalizing mysteries of life and
mind.
I believe that this curiosity accounted for the unexpected popularity of
Wiener’s highly technical Cybernetics and the enthusiastic reception of the
mathematical theory of communication. [Wiener, 1948 and 1961; Shannon,
1948.] Program-steered electronic computers, which made their appearance
as practical devices about that time also, were regarded as logical, symbol¬
processing information engines. Their enormous potential for revolu¬
tionizing our conceptions of problem-solving and information-processing
also stimulated a great deal of speculation and research.
It is fruitless to engage in semantic disputes over when a discipline of
information is not epistemology, psychology, biopsychology, and so forth.
What matters is that investigators who identify with the information disci¬
plines formulate researchable problems and make discoveries, and contrib¬
ute insights that clarify the nature and dynamics of information and knowl¬
edge. Their identification with a scientific community that calls itself
information science may, presumably, provide them with ideas, methods,
paradigms, and interactions. It may lead to success in ways that have not
371
372 MANFRED KOCHEN

succeeded in other communities of inquiry. To suggest that the primary

focus of information science should be library and information work is sti¬
fling and unproductively restrictive. It is not likely to attract scientists with
imagination, ambition, and attitudes similar to those of the best scientists in
other areas.
The conceptualization of information1 as the reduction of uncertainty by a
receiver about which message, from an ensemble of messages, was transmit¬
ted over a noisy transmission channel was a major scientific achievement.
The identification of uncertainty with entropy, and Claude Shannon’s
theorem about the existence of codes that make it possible to transmit infor¬
mation at rates that come arbitrarily close to a specified upper limit with
arbitrarily low probabilities of undetected error, rooted what others called
information theory firmly as an engineering discipline. It placed information,
in its technical sense, in the repertoire of major concepts of modern science.
In this sense, it was definitely divorced from meaning and the less precise
interpretations that could connect it to uses such as information in books or
documents.
Shortly after large-scale computers began to challenge the talents of en¬
gineers, mathematicians, and others—those who later called themselves
computer scientists—to explore and realize their potential, information re¬
trieval became one of the first application areas. It was a solution in search
of a problem, motivated by the urge to explore all possible uses of com¬
puters. It required a broadening of the information concept to include its
semantic, pragmatic, and syntactic aspects. Very few of these precomputer-
scientists/engineers were affiliated with the American Documentation Insti¬
tute (ADI) in the mid-1950s. Taffee T. Tanimoto and Hans Peter Luhn were
the exceptions, and many of the early pioneers, such as Calvin C. Mooers,
still are not active in the American Society for Information Science (ASIS),
the present name of ADI.
In 1960, as a research mathematician at the IBM Research Center, I was
asked by the director of research to participate in a team to survey the
possibilities of automation at the Library of Congress. We became deeply
immersed in the problems of librarianship. We came to appreciate them as
real problems to which the use of computers, among other possible solu¬
tions, might be applicable. For the first time, excepting the problems of
intelligence, we reversed the pattern of a solution in search of a problem.
Also for the first time, we became aware of the difference in attitude
between professional librarians and ourselves. They had to maintain such
standards as the print quality of catalog cards and descriptive and subject
cataloging, and the librarians expected the use of computers to increase their
productivity, decrease the time it takes to produce cards, and increase the
volume and quality of books cataloged. We were looking for greater

1 According to Myron Tribus in his contribution to this volume, Shannon had serious misgivings
about using such an overworked term and did so against his better judgment.
LIBRARY SCIENCE AND INFORMATION SCIENCE 373

justification than that to spend $60 million on powerful computing systems.

The tool or solution was far more powerful than that needed to solve such
problems, and in another sense, it was not appropriate. Somewhere in the
contents of the library’s totality of books must be a treasure, a potential
synthesis of knowledge so valuable that with the help of computers, it could
make a major difference.
It was during the discussions in 1960-1962 about these potential functions
of an automated Library of Congress that the idea of a continually updated,
on-line encyclopedialike knowledge-base occurred to me. [G. W. King,
1963, p. 22.] Herbert Bohnert pointed me to H. G. Wells’s 1936 world
encyclopedia proposal, and I began to see an intellectual path—from Wells
to Vannevar Bush, Alvin Weinberg, Watson Davis, and Charles Van
Doren—that pointed toward what I felt should be a reversal of the priorities
then in force. These priorities had stressed improved general access to the
record of knowledge by specialists.
It became clear that the needs of our age intersected with the potentials of
emerging information technologies in stressing screening, evaluation, and
synthesis of what is known and ensuring that continual additions to knowl¬
edge fit into a coherent image that can be brought to bear on the day-to-day
problems of all people. Such services are active rather than only passive.
They intervene when intervention is needed. The new priorities stressed a
kind of system that primarily teaches, is a work of electronic art, espouses a
human viewpoint, addresses and arouses the curious average person, stimu¬
lates his or her curiosity, and aims to improve the world. Only secondarily
does such a system seek to inform, perform a reference function, aim at the
specialist, and accurately reflect the knowledge, opinions, prejudices of the
time. The proposed system, now called WISDOM (Worldwide Information
Service for the Development of Omniscience in Man), is still an idea, though
pilot versions have been implemented.2 As a system, it is a social organ that
is eventually to supplant the minds of individuals, a kind of community mind
that may be the next stage in the evolution of mind. But WISDOM is also an
emerging movement for the previously mentioned reversal of priorities, and
several organizations have been created to further this movement.
Paul Otlet may have had a similar vision in 1895. Indeed, some docu-
mentalists and information scientists, such as Eugene Garfield, embraced
variants of the idea underlying WISDOM in independent discovery, which
led to contributions of revolutionary impact, such as citation indexing and
clusters based on coupling profiles (bibliographic coupling plus cocitation

2The system was at first called WISE [Kochen, 1972] and elaborated subsequently at a Sym¬
posium of the American Association for the Advancement of Science. [Kochen, 1975.] WISE
was changed to WISDOM at the suggestion of C. T. Larson, who has been working toward
various concrete versions of this idea, such as networks of centers for monitoring and assessing
the state of the built environment, exchanging data, ideas, reports of exemplary practices and
research results in a coherent, integrated form that will lead to improvement, based in part on
his 1953 work with Lonberg-Holms on The Development Index.
374 MANFRED KOCHEN

coupling). Such techniques apply not only to documents that comprise the
written record but also to the individuals who use knowledge as well as to
the fields of knowledge themselves that are only partially embodied in the
record.
Librarianship, library science, documentation, bibliography, and infor¬
mation science in the narrow sense used by Ray ward, have in common a
focus on the written record and the physical documents—books, journals,
manuscripts, musical scores, prints, and so forth—that it comprises. Librar¬
ians are concerned with organizing collections of such documents and
facilitating their use by library patrons. As such, they need to cope with at
least three tasks: (1) continually selecting from the growing record what to
collect (possibly weeding little-used items to make room if there are space
limitations); (2) bibliographic control or organizing and maintaining the col¬
lection as well as maintaining the tools enabling them and their patrons to
determine at any time what is in the collection and where it is; and (3)
reference, to guide patrons in their use of records in the library’s collection.
Information science, in the broader sense in which I interpret it, is, in
contrast, concerned with information, knowledge, understanding, and wis¬
dom. Here, information is used in its technical, scientific sense to denote
what is transmitted over a communication channel to remove a receiver’s
uncertainty about an ensemble. It is both a flow and a pattern that flows over
a channel. This channel may have a memory that stores the encoded infor¬
mation pattern of bits as meaningless codes. The channel may contain a
transducer and transformer that rearrange, encode, decode, or otherwise
process patterns of bits, as in a computer. As a flow, information is a tempo¬
ral change in a pattern. To be informed is to experience a change in some
cognitive structure; to inform is to effect such an experience.
Information becomes knowledge when it is given meaning, usually by a
human mind. A person (or machine) may be said to know the answer to a
previously unencountered question that is not merely a syntatic transforma¬
tion of the statement through which that knowledge was acquired. Knowing
something requires the existence of a knowledge space containing niches for
concepts, ideas, facts, and so forth, that are acquired. These niches are
organized into a web, and the way a niche is embedded into this structure,
relating it directly to many other niches and indirectly to many more, is what
gives it meaning at any time. Knowledge is stock; it may accumulate, but
mere accumulation of knowledge is neither understanding nor wisdom. It is
the latter that can help us to survive and learn to improve our lives.
Knowledge becomes understanding when it contributes to changes in a
knowledge space, adding new connections among niches and new niches,
perhaps revising the entire or large parts of the structure. Certain new items
of knowledge enable the knower to think of and ask questions never thought
of before; they lead to awareness that not all is known and important new
knowledge is needed. Understanding leads to deeper answers to questions
LIBRARY SCIENCE AND INFORMATION SCIENCE 375

regarding how and why, while knowledge is mostly know-what and know¬
how.
Wisdom, in one sense the ultimate purpose of our intellect and the
scientific enterprise, and therefore of central concern to information disci¬
plines, means bringing knowledge and understanding to bear on shaping our
world for human ends by human means. Wisdom governs communication of
information for purposes of control. It includes know-when, that is, the
choice of strategic times for action based on or justified by knowledge and
understanding.
Clearly, this emphasis is far from concern with access to the written
record and even further from understanding such institutions as libraries,
media centers, and so forth. That understanding is certainly necessary but it
is far from sufficient as a professional and scientific foundation for the
knowledge industries. The convergence hoped for by Ray ward is certainly
desirable, but it is not feasible so long as schools of library science take an
attitude of the kind expressed in the following quotation from a resolution by
the faculty of a school of library science:

In its Standards for Accreditation, 1972, the American Library Association s

Committee on Accreditation (ALA/COA is the only group which the Council
on Postsecondary Accreditation [COPA] recognizes for purposes of accredita¬
tion in librarianship) defines libraries and librarianship in the “broadest sense
as encompassing the relevant concepts of information science and documenta¬
tion,” including “media centers, educational resource centers, information,
documentation, and referral centers.” . . . The School is now actively looking
into additional means of enhancing the information science aspects of the
AMLS program . . .

These are parts of a resolution passed unanimously by the faculty of the

School of Library Science at the University of Michigan on October 19,
1978, in formal opposition to informal suggestions for establishing a campus¬
wide doctoral program in the information disciplines with the broader and
deeper interpretation sketched here.
In doing this, the faculty erroneously judged my interpretation of informa¬
tion science to be consonant with that of the Council on Postsecondary
Accreditation and the American Library Association’s Committee on Ac¬
creditation. Note also the phrase, “information science aspects of the
ALMS program,” an exceedingly narrow topic of no interest or relation to
the information disciplines. It is a typical instance of the confusion generated
by superficial interpretations of “information science and documentation
in its popular sense and with reference to such institutions as libraries or
media centers. It is also the typical defensive response of an institution that
feels threatened with extinction. Whether that fear is justified or paranoid
depends on the ability of such institutions to adapt, as in the case of the
376 MANFRED KOCHEN

Baxter School of Library and Information Science at Case Western Reserve

University.
In conclusion, I suggest that the present state in the development of the
five fields under discussion is characterized by strongly divergent orienta¬
tions to information science. Mine differs greatly from the library and infor¬
mation sciences in the title of Rayward’s paper. I may have been among the
first to use the term information science in the previously mentioned broader
sense. [Kochen, 1965.] It may be true that many information scientists,
however they interpret their field, found welcome homes in library schools.
Librarians and library scientists are an amiable group, and the meetings of
the American Library Association are full of vitality and excitement. Their
hospitality to, and acceptance of, information scientists should certainly be
gratefully acknowledged even if there are barriers to understanding and
attitude.
If people trained in physical sciences do not borrow models and
paradigms from their colleagues in librarianship, library science, documen¬
tation, or bibliography, then this is reciprocated. A recent survey of librari¬
ans’ attitudes toward technology showed that they like technologies but not
technologists. [Galvin, 1981.] The forces leading to this state were the rela¬
tively slow rate of progress in basic information science, the corre¬
spondingly slow development work in the field, the slow growth of the
American Society for Information Science, the search by information scien¬
tists for appropriate intellectual homes, the expanding need for graduates of
library schools during the 1960s, and the need to adapt to new technologies
and new ways of thinking.
These forces have now subsided. Replacing them is the more rigorous
development of database-management systems, expert-knowledge systems,
the use of distributed networks and parallel arrays by computer scientists.
Graduates of library schools are no longer in great demand; public libraries,
competing for increasingly scarce municipal funds with police, fire, and
other vital services, are not receiving enough support to stay even, much
less keep up with the growth of the written record at inflated prices. Pub¬
lishers can no longer count on libraries as certain minimal markets for books
and journals. Some employers who support library schools are demanding
that graduates be trained in wholly new ways—with stress on the broader
interpretation of information science, and some even request that library be
entirely deleted from the name of the degree.
The trend is almost certainly away from access to such physical items as
printed paper or microforms and away from emphasis on bibliographic con¬
trol of a physical inventory. Printed paper will probably continue to coexist
with alternative forms of recorded knowledge but lose its predominance.
[Kochen, 1982.] We can no longer refer to electronically recorded databases
as the written record except in a most general sense. To be sure, it is as
necessary to organize knowledge in the newer recorded form as it is in
LIBRARY SCIENCE AND INFORMATION SCIENCE 377

traditional forms, but the principles of cataloging or indexing taught in

library schools may no longer apply.
Attention will, I hope, shift to key problems in information science. More
and better researchers may be attracted to information science, and
their contributions will lead to vigorous enterprises offering new products
and services that meet the needs of knowledge-based societies. When the
growing demand for a firm professional and scientific foundation underlying
the information sectors of such societies is recognized, those engaged in
information will occupy numerous and leading roles and contribute to in¬
creasing the productivity of those sectors.
 LIBRARIANSHIP AND
INFORMATION SCIENCE

Jesse H. Shera

As I anticipated, Boyd Ray ward has written an excellent general statement,

well documented, of the historical development of librarianship in relation to
the emergence of documentation and the transformation of the latter into
information science. Therefore, about the most that I can do is to amplify
certain points he presents and add some commentary of my own. Admit¬
tedly, the commentary is highly subjective and based mainly on more than
fifty years of experience in library practice and teaching. Moreover, it at
times represents a point of view that is not universally held by librarians and
so-called information scientists.
I shall not deal with the stubborn problem of what constitutes a discipline.
Suffice it here to say that it is difficult to deal with the interdisciplinary
relation between librarianship and information science until a decision has
been reached as to whether these areas of activity are, in themselves, true
disciplines. Though, over the years, I have referred to librarianship as a
discipline, I did so mainly because I did not know how else to classify what
librarians do and the technologies by which they have striven to achieve
their ends. Information science has yet to prove itself as a discipline. Like
librarianship, it is still largely an agglomeration of technologies drawn from
other areas of study, particularly, in the case of information science, from
mathematics and electrical engineering. In the final analysis, we are still
compelled to say, I think, that librarianship is what librarians do and infor¬
mation science is defined by the operations of the information scientist. We
could say with, I think, some degree of accuracy that librarianship is a
service and information science is an area of inquiry that seeks to measure
and improve the efficiency of the librarian; but such a statement is inade¬
quate because information science deals with only a part of what the librar¬
ian does.
Viewed from the perspective of history, then, as admirably presented by
Rayward, Paul Otlet and his associates who formed the Brussels Institute

Some parts of this paper were used in an article entitled “Information Science and the Theory of
Librarianship,” published in the Indian journal, International Information and Communica¬
tion, vol. 1 (March 1982), pp. 3-15.
379
380 JESSE H. SHERA

and subsequently the International Federation for Documentation (FID),

were basically librarians. Librarianship has always been interdisciplinary in
that its goal was to make the total record of the human adventure readily
accessible to all. Otlet and his followers merely strove to expand the area of
library activity to encompass a wider range of materials and improve access
to those records by using a new technology. They adopted the classification
scheme devised by Melvil Dewey to form the Universal Decimal Classi¬
fication (UDC) and supplemented it by using certain signs and symbols
of interdisciplinary relations. Neither traditional librarians nor these in¬
novators exemplified a new field of academic endeavor; both were essen¬
tially centripetal—drawing to themselves the products of scholarship and
the accumulated recorded knowledge of many fields. Thus, these pioneers
relied on librarianship for their basic principles and structures. They
elaborated what had already been done. That they called themselves docu-
mentalists and their field of activity documentation did not change their
basic objectives. When the documentation movement came to the United
States, American librarians rejected the UDC even though it was based on
Dewey’s system, because they believed it to be too complex; they turned,
instead, to microphotography as a technology that would improve the acces¬
sibility and availability of graphic records as well as provide a substantial
reduction in costs. Librarians even foresaw a time when microfilm would
replace the conventional book. On this side of the Atlantic, documentalists
were mainly librarians who had adopted a new technology; there was no
alteration in the basic goals.
Because these American documentalists, most of whom were librarians,
based their technology on the science of photography, they tended to regard
themselves as scientists, tended to avoid the term librarian, and formed the
American Documentation Institute (ADI). Thus began the fragmentation
that has plagued librarianship ever since.
By mid-century, microphotography had become firmly established as an
important addition to the librarian’s arsenal of technologies, and most of the
members of the American Documentation Institute had turned to other inter¬
ests. Those few who retained their concern for the further development of
microphotography seceded to form the National Microfilm Association,
composed mainly of commercial interests organized for the promotion and
manufacture of photographic equipment.
The remaining members of the ADI turned to new techniques for the
organization of library materials—which they characterized as information
retrieval. These new interests were largely devoted to the application of
electronic and mechanical systems for improving the accessibility of graphic
records. Thus, there evolved such systems and mechanisms as Calvin
Mooers’s Zato coding, Mortimer Taube’s coordinate indexing, Ralph
Shaw’s rapid selector, and the searching selector fabricated by J. W. Perry
and Allen Kent at Western Reserve University. The ADI was then largely
composed of librarians, with the addition of a number of commercial organi¬
zations, such as Eastman Kodak, Remington Rand, and General Electric,
LIBRAR1ANSHIP AND INFORMATION SCIENCE 381

which had research resources far beyond those of librarians. The line of
demarcation between documentalists and librarians was even then beginning
to be drawn, and there were some suggestions that library schools should
change their names to schools of documentation. Librarianship was becom¬
ing a tarnished word, and many wanted to dissociate themselves from it as
much as possible.
This movement was intensified by the influence of Claude Shannon and
Warren Weaver and their work at the Bell Telephone Laboratories on the
communication of information. The term information was misleading, but it
was used by members of the American Documentation Institute as being
descriptive of their major concern. Again, the influence of Shannon and
Weaver became prominent in attracting mathematicians to their field, creat¬
ing such technologies as bibliometrics and citation analysis. These phenom¬
ena had long been known to librarians, but their rebirth gave the phenomena
an air of novelty and innovation.
By the late 1960s, the term documentation had become old hat, and the
ADI changed its name to the American Society for Information Science.
Today, even the Special Libraries Association, founded in 1909 by John
Cotton Dana (librarian of the Newark, New Jersey Public Library) to serve
the bibliographic needs of business and industry, is contemplating a change
of name to something akin to an association of information managers. The
flight from library goes on, leaving the old and respected name to typify, in
the main, public library service. The trend would make an interesting study
in linguistic deterioration. It is quite evident that information science is
rooted in attempts to extend the boundaries of library technology and give it
revived respectability by endowing it with a unique name.
The American Society for Information Science and its preceding organi¬
zation, the American Documentation Institute, could provide an excellent
subject for the study of that peculiar amalgam that is librarianship-
information science. Unfortunately, a thorough inquiry of this kind has not
yet been made. However, in 1979, Donald W. King and two graduate stu¬
dents at the library school at Drexel University interrogated the entire mem¬
bership of ASIS in an attempt to provide a profile of the membership of that
organization. [King et al., 1980, pp. 9-17.] The study does provide some
interesting insights. Of the 3894 regular members of the association, 52 per
cent responded. Of the respondents, 24.2 per cent, the largest single group,
were engaged in library administration. The second largest group was 20.3
per cent, composed of managers, directors, or coordinators. Information
technologists, analysts, and scientists came third, with 18.6 per cent, at
which point the percentages drop off sharply. Unfortunately, comparable
data are not available for the early years of the American Documentation
Institute, but I can testify from my own knowledge that the configuration of
the membership was quite different. Most of us would have characterized
ourselves, without hesitation, as librarians.
It is also interesting to note that in the survey of the primary employers of
ASIS members, 35.6 per cent were from industry, 26.9 per cent were from
382 JESSE H. SHERA

colleges or universities, and 12.1 per cent were from the federal government.
In the sample, 56 per cent were women, while in the earlier ADI years, there
was no more than a handful of women. Whether or not these statistics reflect
the actual situation is problematical in that women may be more responsive
to questionnaires than are men, but it does indicate that the new technology
is increasingly of interest to women.
In an editorial on the study, Donald King points out that ASIS members
are concerned with four primary professional areas: operational information
functions, including searching, preparing, and analyzing data on behalf of
others—1490 ASIS members; management of information operations, pro¬
grams, services, or databases—1160 members; information science and
technology, including research and development, systems analysis, and sys¬
tems design—530 members; and educating or training information profes¬
sionals or other workers—360 members. [King, 1980.] We somewhat glibly
assert that librarianship is a field of practice whereas information science is
an area of research—and it was in the early days. But today, if the member¬
ship of ASIS is indicative, information science is approaching an area of
service not unlike that of librarianship. Both activities are concerned with
the transfer (I prefer the term communication) of certain types of informa¬
tion (and again, I prefer knowledge). Information science, however, gets
bigger headlines because of its use of engineering and mathematics in its
technology.
I have indulged myself in this recapitulation of Ray ward’s historical sum¬
mary of the origins of information science in part to amplify some of the
points he makes and also to set the stage, so to speak, for some of my own
argument.
In his summation, Ray ward evinces a certain doubt in his own mind about
an organic interdisciplinarity between librarianship and information science.
Thus, he writes: “When, however, the focus becomes not library and infor¬
mation work, but information science, the clarity, such as it is, is lost . . .
[Immediately there arises the] problem of disciplinary encroachment. When
is information science information science and not epistemology, biopsy¬
chology, physiology or physics or journalism or the sociology of groups or
anthropology?” When he sees information science broadly defined, he dis¬
covers a loss of all disciplinary coherence. He goes on to see this library/
information science relation as a continuum with

at one end ... the librarian constructing a catalog of books; at the other,
because of the way in which information can now be conceived, . . . the
genetic engineer consulting the information recorded in and then restructuring
DNA. Perhaps, however, there is a discontinuity between the two involving
notions of symbolic representation and public consultation that more careful
analysis and definition would reveal.
The point here is not so much the difficulty of definition or the slipperiness
of terminology as the implication that the generalizing process described earlier
L1BRARIANSHIP AND INFORMATION SCIENCE 383

in this paper may still be at work. It is a source of innovation and strength, but
it blurs traditional distinctions and can unsettle professional convictions. (Ray-
ward in his lead paper in this volume.)

Twenty years ago, I thought of what is now called information science as

providing the intellectual and theoretical foundations of librarianship, but I
am now convinced that I was wrong. A building is not a library, and its
organization, operations, and services not a machine. I seriously question
whether there is a true interdisciplinary relation between librarianship and
information science; rather, it is only a series of borrowings of the technol¬
ogy of one for the use of the other. Because librarianship is much more than
the mechanized access to data banks or networks that provides efficient
access to institutional borrowing, we must look to other disciplines for its
interdisciplinary relations and the core of its theory.
The enthusiasm with which many librarians, especially academic librari¬
ans, rushed to embrace new technologies with only the most casual consid¬
eration of their appropriateness seemed to substantiate Kaplan’s “Law of
the Instrument.” Simply stated, this so-called law holds that people tend to
formulate their problems in such a way as to make it seem that the solution
to the problem demands precisely what they happen to have at hand. He
illustrates this law by saying that a small boy given a hammer immediately
concludes that everything needs pounding. [Kaplan, 1964, p. 303.] A more
sophisticated example might be that an executive who has acquired a photo¬
copying machine almost inevitably concludes that all of his or her immortal
prose should be distributed throughout the organization in multiple copies.
There are times when it seems that the goal of science is not to know or to
understand but to use. In the American character, there has been a strong
strain of utilitarianism, or, rather, pragmatism, and this is perhaps no more
clearly evident than in librarianship. The major figures in the history of
American librarianship were doers rather than thinkers; they were con¬
cerned with process rather than purpose. They devised and taught in their
library schools routines and procedures, and with the advent of online net¬
works and access to data banks, they are doing it more than ever today. We
need not be surprised, therefore, that librarians rushed to embrace science
and began to talk glibly of library science, turning their backs on the library’s
humanistic origins. Librarians even went so far as to evolve a new discipline
called information science. They misinterpreted Shannon and Weaver’s spe¬
cialized use of the noun information and assumed that it related to the
communication of knowledge rather than the transmission of signals. Infor¬
mation theory, in contrast to the theory of librarianship, is severely limited
in its communicative potential. Its objective is to provide the librarian with a
system giving functional or operational access to the contents of documents.
Information scientists give all of their attention to design, production, im¬
plementation, and control of the system. Thus, as H. Curtis Wright has so
well pointed out, our schools of library science are busily engaged in produc-
384 JESSE H. SHERA

ing “control artists,” technicians and mechanics who tinker with informa¬
tion systems but are unable to solve the problems of access and retrieval
because they cannot control ideas. These control artists constantly mistake
the symbol for reality and believe that counting or figuring with these sym¬
bols solves the bibliographic access problem. [Wright, 1981.] This confusion
between data and ideas leads to the confusion between data systems and
idea systems. Our so-called information systems are actually nothing but
data systems. Thus, Wright is close to what Robert Fairthorne has called
“marking and parking” in document retrieval rather than information or
knowledge retrieval.
Librarians have become so concerned with process that they have con¬
fused substance with instrumentation. Processing data can be performed by
machine, but only the human mind can process knowledge or even informa¬
tion. Information science is based on data and their manipulation, not on
ideas.
Science deals with things, things that can be measured, weighed, poured,
fastened, or mixed together, whereas librarians are dealing only incidentally
with things but primarily with ideas, concepts, and thoughts. Librarians are,
or should be, characterized by their knowledge, not by their instrumenta¬
tion. We conventionally say that the computer is here to stay and we must
learn to adapt to its powers and capabilities. I submit that the situation is
quite the reverse: The computer is here to stay, therefore it must be made to
adapt to the librarians’ problems and needs. It must be kept in its proper
place as a tool and a slave, or we will become sorcerer’s apprentices, with
data data everywhere and not a thought to think.
Librarians, ever sensitive to the somewhat meager social esteem ac¬
corded their activities, devised the term library science and eagerly seized
information science as potential supports to their claims to professionalism.
But science does not a profession make, and an overlay of scientific opera¬
tions is not a sine qua non for professionalism. The term profession derives,
of course, from the root verb to profess, to believe in, and etymologically, a
professor is one who acts according to his or her beliefs or convictions.
Wright has recently reviewed the development of the concept from the
Greek arete, which meant skill or proficiency in a particular job and a knowl¬
edge of the job in hand, to the well-known characteristics of a profession as
enunciated by Abraham Flexner. [Flexner, 1915.] Wright cites Socrates to
the effect that if we wish to know what a good shoemaker is and wish to be a
good shoemaker, the first essential is to know what a shoe is and what the
purposes of the shoe are. Only then can we decide what materials are best
for making shoes and what tools should be employed and the best techniques
and methods for using these materials and tools. But such skills and tech¬
niques cannot be successful without an understanding of what is to be pro¬
duced and what functions it is supposed to perform. Wright quite properly
sees a direct connection between knowledge of a task and its performance,
citing the so-called Socratic paradox that “virtue is knowledge,” which was
LIBRARIANSHIP AND INFORMATION SCIENCE 385

interpreted by his contemporaries to mean that we cannot be efficient unless

we take the trouble to learn our job. Stripped of the verbiage that has
accumulated around the paradox over the centuries, what Socrates meant
was that when there is a specific occupational function to perform, the
worker must perform that function knowing what the task is, what its pur¬
pose is, and the best way of performing it. This, says Wright, is the true basis
of professionalism. [Wright, 1980.] We might also cite the statement attrib¬
uted to Michelangelo, that the apprentice sculptor should learn anatomy
thoroughly and then forget it. By this admonition, he enunciated the belief
that a piece of sculpture should not do violence to anatomical laws, but, at
the same time, it should be more than an exercise in anatomy. He was
subordinating process to purpose. Librarians would do well to remember
occasionally Moses or a Pieta and think somewhat less frequently of Shan¬
non and Weaver.
What, then, is the “shoe” that is the end product of the “shoemaker,”
the professional librarian. I submit that there are three components in the
concept of a library. First, there is acquisition, which involves knowing
what to acquire for a given clientele and how to acquire it. Acquisition
implies substantive knowledge of the materials and the uses to which that
knowledge can and should be put. The second is organization, that is, ways
in which the accumulated materials can be arranged and processed for max¬
imum convenience and efficiency of use. It is here, and only here, that
information science makes its contribution to librarianship. The third is
interpretation and service, which is the raison d’etre of the library, and it is
here that information science reveals its greatest threat of dehumanizing the
library’s function. To express the same idea in another way, the library can
be seen as three interrelated spheres: the sphere of optimum content; the
operational or mechanistic sphere; and the sphere of maximum context. If
librarians persist in sublimating librarianship to the lure of the machine and
the all importance of data retrieval their “shoes” will be very uncomfortable
to the user, and they will leak at certain vital points.
Janus-like, the library looks in two directions simultaneously. It looks
toward the social sciences, because it is a creature of society, evolved to
meet the needs of human beings working toward the solution of certain
problems. Thus, the library qualifies as a social agency or instrumentality.
But the library is also humanistic in that its characteristics, modes of access
to its resources, uses, and values are humanistic. It is not, and never has
been, a scientific enterprise. Except for some tinkering with its processes,
scientific technology has made but little contribution to the development of
the library. A catalog entry on the face of a cathode ray tube is essentially
the same entry devised by Charles A. Cutter, and the fact that it has been
transmitted across the continent in a fraction of a second may be a conve¬
nience to the user, but it does not alter the character of librarianship or
transform it into a science.
Of the importance of the humanities to a world of science, a professor of
386 JESSE H. SHERA

history at the University of Michigan has recently written: “Our society is a

seamless web, and ultimately science and technology share the same values
and are dependent upon the purposes which derive from the study of what it
means to be human. The neglect of the humanities will bring in its train
inevitably the decay of science.” He goes on to characterize an excessive
dependence on science alone as “the higher barbarism, unless there is some
human purpose that looks beyond the immediate technical and economic
applications. The day will come when the hand will lose its cunning and life
be found to be devoid of purpose. We are more dependent upon our poets
than we realize.”1
It is perhaps reassuring that all of us know successful scientists who
enrich their professional reading with generous additions from the
humanities; who are accomplished amateur musicians; or who, like the dis¬
tinguished astronomer Fred Hoyle, have made reputations in creative
writing.
Information theory is severely limited in its communicative potential. Its
only aim is to provide efficient access to knowledge, but the design of
mechanisms, their production, and use of the system are its areas of concen¬
tration. We are, as Wright has emphasized, turning out control artists and
data mechanics. This confusion of ideas with data can be dispelled only by
distinguishing between data systems and idea systems. Thus, it follows that
information science cannot qualify as a theoretical base for librarianship,
and calling it bibliometrics or informatics does not alter the situation. We
librarians must constantly remind ourselves that our concern is with
sociological and psychological phenomena, not physical objects and pro¬
cesses. Moreover, information science itself does not know what it is or
where it is going. It cannot tell whether it belongs in a school of library
science, engineering, management, operations research, or some loosely
defined area of its own. There is already distinct evidence that the American
Society for Information Science is fragmenting itself into a variety of activi¬
ties, some of which are mutually contradictory.
I submit that librarians must look to “symbolic interactionism” for the
proper foundation of a theory of librarianship. This term, first named by
Herbert Blumer in 1937, is rooted in the social psychologies of William
James, Charles S. Peirce, Charles H. Cooley, John Dewey, and George
Herbert Mead.
The term symbolic interaction refers to the process by which people
relate to their own minds and the minds of others; the process by which
individuals take account of their own or others’ needs, desires, means and
ends, knowledge, and like motivations. Among sociologists this phenome¬
non is frequently known as social interaction. In the definition of symbolic
interaction, the word mind denotes instrumental activities that animals as
well as human beings direct toward their environment. These instrumental

‘Stephen J. Tonsor, letter in Wall Street Journal, August 31, 1981, p. 13.
LIBRARIANSHIP AND INFORMATION SCIENCE 387

activities, sometimes referred to as action or as psychological activities, are

related to the organism’s requirements and the conditions and resources in
the environment that are relevant to meeting those requirements. The term
symbolic includes the representation of those requirements through a system
of commonly understood linguistic, or other representational activities, such
as gestures, whereby the content of the communicated message is transmit¬
ted to, and understood by, the intended recipient.
Such a discipline of symbolic interaction reminds librarians that their area
of concern is based on social phenomena and it provides for the contribu¬
tions of philosophy, linguistics, and psychology in the formulation of its
unique purpose. That purpose is to make accessible the graphic records of
human culture, so that people may understand the totality of the environ¬
ment in which they find themselves and their own place in it.
In summary, we who are librarians must constantly remind ourselves that
information science is an area of inquiry, of research. It is not, as is librari-
anship, a service or a practice. Information science can provide the librarian
with some important and useful tools to expedite library services, but the
ability to communicate a message with incredible speed over long distances
through the use of glass fiber bundles or laser beams or to store vast quan¬
tities of recorded knowledge in computerlike mechanisms does not in any
way alter the purpose of the library. That the internal-combustion engine can
move vast quantities of materials and people over longer distances and with
greater efficiency than the horse does not alter the need for transportation.
The social purpose of the library remains unchanged—to bring the human
mind and the graphic record together in a fruitful relation—for the growth of
knowledge is still the goal. Administration, management, architecture, and
many other disciplines can contribute to the effectiveness of the library, but
they are not librarianship.
We who are engaged in library education must be aware that the librarian
needs to attain certain specific skills in the use of the tools that all of these
disciplines, including information science, have given us. But the hard core
of librarianship remains basically as it has always been—the mastery of the
substantive content of graphic records. Librarians should be characterized
by their knowledge and not by their instrumentation. The most effective
librarians enter the profession armed with broad generalized knowledge or
with knowledge in a chosen field. Librarians must remain the guides to the
mastery of the substantive content of graphic records. The idea, not the
process, is our primary concern.
The great danger with which information science threatens librarianship is
the loss of control of the library profession to other and less competent
hands. Of the dangers inherent in this threat, we do not now seem to be fully
aware. The basic purpose of librarianship is not encompassed in the ma¬
chine, and there is much more to librarianship than is envisaged in informa¬
tion science. If we permit ourselves to be mesmerized by the gadget, if we
accept the flickering image of data on a fluorescent screen as knowledge, we
388 JESSE H. SHERA

will soon become like those mythical people of many centuries ago who
mistook for reality the passing shadows reflected on the walls of a cave. So I
conclude with the words of that distinguished librarian of Stanford and sub¬
sequently dean of the library school at Berkeley, Ray Swank: “The long
experience and tested values of librarianship are still our most reliable and
sophisticated resources for the solution of information-handling problems in
most contexts whatever additional help may now be needed.” (Swank,
1981.)
 BIBLIOGRAPHICAL R&D

Patrick Wilson

Boyd Rayward says that librarianship has converged toward information

science, which was born in the 1960s out of documentation by the computer.
That is not how I would describe the situation. To show what I think wrong
with his story and what I think the correct story is, I will have to go over
some of his ground again, redescribing the situation from a somewhat differ¬
ent angle and with a different distribution of emphasis.
Let me use the phrase the bibliographical sector for the assemblage of
institutions and organizations that collectively take the output of the publish¬
ing industry and try to make it accessible for public use. This sector includes
the wholesale and retail book trade, libraries, and various agencies—
scholarly, professional, and commercial—that produce such bibliographical
instruments as abstracting and indexing services, general periodical indexes,
lists of newly published books and books in print. The job of making the
output of the publishing industry accessible has two logically distinct parts:
first, making it possible to discover the existence of a publication; and sec¬
ond, making it possible to get our hands on a copy of the publication. The
first is the problem of intellectual access, the second that of physical access.
Until the middle of this century, the bibliographical sector was a relatively
tranquil sector of society, but soon after World War II it became an agitated
and problem-ridden one. First, the spectacular growth of science and tech¬
nology led to a vast expansion of the output of scientific and technical
literature, constituting what was frequently referred to as an information
explosion. To some, it appeared that traditional methods of providing intel¬
lectual access to scientific and technical literature were incapable of working
adequately in the changed circumstances. Second, the equally spectacular
growth of higher education put great strains on libraries in colleges and
universities and made it imperative that new methods of handling very large
amounts of bibliographical information be developed. Happily, the concur¬
rent development of computer technology promised to provide the neces¬
sary new technology to solve the problems of both the size of scientific and
technical literature and the pressures on academic libraries.

389
390 PATRICK WILSON

RESEARCH AND DEVELOPMENT ON BIBLIOGRAPHY

During the last 25 years, an increasing number of people went to work on

research and development centering on the problems of the bibliographical
sector and focused mainly on libraries and bibliography-producing agencies,
to the exclusion of the book trade. Pending a later discussion of the right
name for this group, let us call it for now the bibliographical R&D group.
Some of the members of this group came from the library profession and its
educational institutions (the library schools), others from the group calling
itself documentalists or documentation specialists (but considered to be sim¬
ply special librarians by other librarians). But a large number came from
outside: from the fields of mathematics, logic, physics, chemistry, linguis¬
tics, industrial engineering, electrical engineering, computer science, and so
on. They worked in various settings: some in newly created research insti¬
tutes affiliated with universities, some as members of newly created R&D or
“systems” staffs in libraries and bibliography-producing agencies, some in
independent research organizations and industrial organizations, some as
faculty members of library schools or engineering schools or computer sci¬
ence departments. The entrants from other fields naturally brought their
tools with them as well as their expectations about the proper way of tack¬
ling problems in the bibliographical sector. They did not, of course, bring
with them an information science, for none existed. They hoped to create
one. The question is whether or not they did. To answer this, let us start by
sorting out the different kinds of work done by the bibliographical R&D
group over the last quarter-century. The work falls roughly into six catego¬
ries, of which three are of special interest.

First Category

The first and most successful category includes work aimed at improving the
means of storage, manipulation, transmission, and display of bibliographical
information, based on the application of computer technology. Recent years
have seen the development of important new products, systems, and ser¬
vices, such as computer-based services for literature search, computer-
produced citation indexes, and computer-based support systems for catalog¬
ing. While it is impossible to draw a sharp line between research and
development work, it seems clear enough that the work done by the R&D
group leading to such new products, systems, and services has been pre¬
dominantly development work. It is work for the designer of computer sys¬
tems and the programmer. It is technology, not science; the goal has been,
not to produce true statements about the world, but to produce better de¬
vices and systems offering new or better services—product and process
innovations. There are plenty of published documents describing the work,
but the documents are incidental. The literature resulting from this work,
BIBLIOGRAPHICAL R&D 391

like the literature resulting from work on improving technology generally, is

not a body of findings but of reports of makings. The work has increased the
available stock of know-how, much of it embodied in devices and systems.

Second Category

The second category includes work aimed at improving techniques for creat¬
ing the bibliographical information to be manipulated, stored, transferred,
and displayed. The improved products and processes already mentioned are
advances in operation on given bibliographical information. Attempts to
improve processes of originating the information have been strenuous but
much less successful. At the technical heart of the job of providing intellec¬
tual access to publications are the operations of bibliographical description
(description of everything but the content of a publication) and content
representation through indexing, abstracting, and classification. Work on
problems of bibliographical description has been essentially organizational
or managerial: trying to get agreement on international standards for de¬
scription and trying to reduce unnecessary duplication of effort. Work on
problems of content representation has been theoretical and experimental
rather than managerial: trying to find new principles and radically different
techniques for the analysis and representation of intellectual content. At¬
tempts first to evaluate and then to improve on the performance of tradi¬
tional methods of content representation have been frustratingly difficult and
on the whole disappointing. Experimental attempts at computer-based auto¬
matic indexing and classification have so far led to some advances in under¬
standing but few practical applications. (Computer-produced indexes where
words in titles and abstracts are taken as content indicators without further
analysis are simply ways of avoiding the problem of content representation.)
Abstract analysis of indexing, employing probability and utility theory, has
resulted in formulations of normative conditions for optimal indexing, but
not in altered practices. Improvement in techniques for the representation of
intellectual content is fundamentally different in character from, and enor¬
mously more difficult than, improvement in manipulating strings of charac¬
ters after it is decided what the strings are to be. Strings of characters can be
manipulated, transmitted, and displayed without any attention whatever to
the meanings they convey; producing a correct representation of semantic
content essentially requires understanding the content at some level. Index¬
ing and classification still depend on the rapid intuitive grasp of the scope of
the intellectual content of a text. Numerous novel indexing and classification
schemes for displaying the results of that intuitive grasp of content have
been devised, and many of them put into effect, but it is still true that the
subject indexing and classification schemes most widely used by libraries are
those proposed, and already well developed, a hundred years ago. It is
commonly agreed that these old schemes do not work very well, but improv-
 392
PATRICK WILSON

ing on them in practice would seem to require some improvement of, or

substitute for, that intuitive grasp of intellectual content, and no one has
good practical ideas about how that might be done.

Third Category

The third kind of work done by the bibliographical R&D group is not norma¬
tive but descriptive, devoted to studying some of the characteristics of the
literature that constitutes the input to the bibliographical sector and the use
of that literature. Much of this is quantitative in character, consisting of
bibliometric studies and studies using citation analysis; for example, studies
of the size and growth rates of bodies of literature, concentration and dis¬
persal of literature of various subjects in a population of serial publications,
and obsolescence or decline in use of literature over time. Bibliometrics and
citation analysis are relatively straightforward unobtrusive measures of
some aspects of human communication behavior. [Webb et al., 1966.] Other
quantitative but obtrusive studies of communication behavior are based on
observation, questionnaire, or interview; for instance, the number of hours
spent reading, sources from which people learn of publications useful to
them, the use made of library catalogs, and so on.
There are those who think, or once thought, that such work would lead to
the discovery of laws governing information behavior, such as “laws ac¬
cording to which different kinds of information lose their meaning, validity,
relevance, or value” (Zunde and Gehl, 1979, p. 69); or laws governing the
growth of knowledge; or laws that would stand to information retrieval as
physics stands to engineering. [Weiss, 1977, p. 2.] I take it that we can safely
say that no laws have been discovered: interesting empirical regularities, yes
(Bradford s Law of Scattering being the most prominent example) but laws,
no. [Bookstein, 1976.] A natural law expresses a constraint, a natural neces¬
sity; it tells us what in some sense (hard to clarify precisely) must happen,
and what cannot happen. Simple statements of empirical regularity do not do
that. And it is no more likely that information laws will be discovered by
looking at observable features of documents and uses of documents than
that sociologists or political scientists will discover laws of social and polit¬
ical behavior by looking at what people say and do. [P. Wilson, 1980.] It is
more reasonable, if much more modest, to look on these studies of docu¬
ments and communication behavior as a form of social intelligence, useful to
those concerned with the workings of the bibliographical sector for the
enlightenment they give, if for no other reason. Social intelligence is unlikely
to be directly applicable in the development of new products or processes,
but it can, in the form of market research, indicate the likely reception of a
new service. It can help identify lacunae in services, suggest places where
practical improvements are desirable, give at least rough guidance in the
formation of plans and policies, and spot social changes that are likely to
BIBLIOGRAPHICAL R&D 393

have an impact on the workings of the sector. Research of the sort indicated
here can, of course, be undertaken for its own sake, simply because it is
interesting. There is, in that case, no external criterion of significance, one
fact being as good as another if found to be as interesting. If undertaken to
provide social intelligence to the bibliographical sector, however, research
has an approximate criterion of importance, namely, how much difference it
makes to our plans or expectations or simple understanding of the situation
in which we work. On that criterion, it would have to be admitted that much
of the work actually done is of little or no significance, but that is hardly
unique to the work of this group.
The gathering of social intelligence quite clearly overlaps with work done
by other researchers unconnected with, and only incidentally interested in,
the bibliographical sector. Both in techniques used and in phenomena inves¬
tigated, this work is similar to work done by some sociologists and com¬
munications researchers. More generally, it can be seen as belonging to the
sociology of knowledge. The sociology of knowledge was for a long time too
closely identified with the work of Karl Mannheim to be very productive;
more recently its interests have widened, and a good deal of sociological
inquiry is directed at questions that are also of interest to those gathering
social intelligence for the bibliographical sector. [Holzner and Marx, 1979;
T. D. Wilson, 1981.] Of course, sociologists are not the only social scientists
interested in information and its producers and consumers. How could there
be a branch of the study of social life that did not, sooner or later, have to
investigate information and communication? And bibliographical R&D
might overlap with work in any of the social sciences; as time goes on, we
would expect more overlapping, not less.

Fourth, Fifth, and Sixth Categories

The last three categories can be very briefly described as consisting of stud¬
ies of the bibliographical sector itself as it has been, as it now is, and as it
ought to be. Actually, the fourth category, which includes historical studies,
contains a great deal of work consisting of the study of old books as physical
objects, but it does also include studies of the history of libraries, publishing,
and bibliography. The fifth category includes analysis and description of
present conditions in the bibliographical sector, much of which is simply
social reporting, useful since few people are in positions to know much at
first hand about the actual workings of the sector as a whole. Outsiders, or
occasional temporary members of the bibliographical R&D group, have
made important contributions to this category, notably of economic analy¬
sis. [For example, Machlup and Leeson, 1978-1980; Baumol and Marcus,
1973.] The sixth and final category includes studies aimed at improving the
organization and management of the sector (apart from its central biblio¬
graphical components) and analysis of alternative policies for service. The
394 PATRICK WILSON

bibliographical R&D group is certainly not unified by shared problems or

methods, but is divided into numerous small groups working in very differ¬
ent ways at strikingly different sorts of tasks.

CONVERGENCE OF LIBRARY SCIENCE AND INFORMATION SCIENCE?

Now we are in a better position to assess the merits of Rayward’s claims. Is

it correct to say, as Rayward does, that librarianship is converging toward
information science? The question would have been easier to answer if
Rayward had said instead that library science was converging toward infor¬
mation science. Let us see why. Rayward tells us surprisingly little about
information science. But if it is, indeed, a separate discipline, as he thinks,
and if its content is approximately represented by the Annual Review of
Information Science and Technology, as he suggests, it is certainly not the
only discipline concerned with information. Computer scientists think they
are concerned with information, as do cognitive psychologists, communica¬
tions researchers, and numerous others. We have to say that if Rayward’s
information scientists are correctly named, they have a generic claim to the
title but not a specific claim. They may be information scientists, but they
are not the only ones; they constitute only a particular species of the genus
information science. Now in that sense, library scientists, if there are any,
might also claim to be information scientists; library science would be the
name of the species, information science the name of the genus. Then Ray¬
ward’s thesis could be expressed as follows: One species of information
science, namely library science, was converging toward another species, as
yet unnamed. This might mean that library scientists were increasingly im¬
itating the research style of the unnamed information scientists or perhaps
that they were one by one joining the other group. This would be under¬
standable and might be true. But this is not what he claims is happening. It is
the profession, librarianship, not its corresponding science, that he has do¬
ing the converging. Presumably, this does not mean that librarians are ceas¬
ing to be members of a service profession and are taking up scientific re¬
search; rather, it must mean that librarians increasingly look to this unnamed
branch of information science rather than to library science for guidance and
help. This interpretation seems consistent with what Rayward says of library
science: that there were attempts to create such a science but, to put it
crudely, library science was a flop. Information science looked vigorous and
apt to provide “a more respectable disciplinary base for the profession of
librarianship,” so library science was dethroned and information science
installed in its place.
Is that what has happened? I do not think so. We have to face squarely the
question of the actual content of the unnamed branch of information science.
For some, this is unproblematic; it is the second of our six research catego¬
ries, mainly concerned with content representation, and might best be
BIBLIOGRAPHICAL R&D 395

named information-storage-and-retrieval theory. For others, it is equally

unproblematic but the third rather than the second category, empirical, look¬
ing for behavioral laws rather than practical rules. For still others, it is the
first category, principally design of computer-based information systems.
Now let us agree that there is a difference, however hard to define exactly,
between research and development and between science and engineering
and that the largely developmental and engineering-oriented first category
cannot furnish the scientific content for a discipline, however practically
important and useful it is. No one calls the work in our fourth, historical,
category by the name information science, nor are the fifth or sixth catego¬
ries likely to be so identified. So if this branch of information science can be
identified by its subject matter, it must be the subject matter of categories
two or three or both. But the work done in these categories is done both by
people calling themselves information scientists and people calling them¬
selves librarians {not library scientists—I know of no one who calls himself
or herself a library scientist).
Content representation is a major practical and theoretical concern for
librarians, and quantitative studies of the literature and its users have long
been established topics of investigation in librarianship—Bradford of Brad¬
ford’s Law was a librarian, even if a special librarian. So information science
(or this species of it) cannot be distinguished from library science by its
special subject matter. Perhaps it cultivates a shared subject matter in a
distinctive way? Perhaps the work done by information scientists is theoret¬
ical while that done by librarians is not? Or mathematical while that of others
is not? Or experimental? Or exact? Or simply, good? No, none of these, not
even the last one, will provide a clear criterion for demarcation. No
methodological or topical criterion will provide a clear division between
work done by those who think of themselves as librarians and those who
think of themselves as information scientists. Of course, we are at liberty to
say that it matters not at all what people call themselves or how they think of
themselves; we can simply define information science by reference to
specified subject matter and methods. Anyone can do that if he or she
wishes, but there does not seem to be any general agreement on how to draw
the boundaries. And doing so turns an empirical question about the relations
of two groups into a matter of definition. Some library science will turn out
to be information science, and some information scientists’ work will turn
out to be library science, thus yielding a convergence by definition. But then
we are no longer talking about history. Since Rayward was talking about
history, I take it that he would not want to take this line.

THE TRANSFER OF RESEARCH STYLES

The fact is that those newcomers to the bibliographical R&D community

from mathematics, physics, and other fields brought their characteristic
396 PATRICK WILSON

techniques and expectations and enlarged the available repertory of research

styles in the community, styles that were then available for others to borrow
or copy. The newcomers naturally affiliated themselves with documentalists
rather than with librarians, since their initial interest, like that of the docu¬
mentalists, was in problems of access to the scientific literature. At first,
they were, indeed, a distinct group, socially as well as intellectually, from
the (relatively few) people doing library-related research. As time has gone
on, though, the distinctions have become blurred. It is harder, not easier, to
tell the difference between a piece of information science and a piece of
library science. [It is useful to look at Peritz, 1980-1981 in this connection.]
Information science did not gel into a distinctive, coherent research area
with its own subject matter and research methods. In particular, it did not
arrive at a distinct “solid corpus” of “ scientifically derived” theoretical and
practical knowledge. And it did not replace library science as the disciplin¬
ary base for the profession of librarianship. That profession has no unique
disciplinary base but draws whatever it finds useful from research anywhere
in the bibliographical R&D sector and anywhere outside—sociology, man¬
agement science, communication, and other fields. At it happens, librari¬
anship can draw only a small amount from the research done in categories
two and three; as noted earlier, practical applications from category two
have been few. Working professionals want practically usable results, but
those have been few in the categories that are the chief contenders to consti¬
tute the content of information science. So, despite Rayward’s evidence
from name changes in library schools and the like, I think this convergence
of librarianship is mythical.
Shall we say, instead, that library science has converged toward informa¬
tion science? I think that could be made out if all it meant were that styles of
work in library science had been influenced by the work of those newcom¬
ers. But I would much prefer to avoid talk that suggests two distinct and
distinguishable groups. The bibliographical R&D community is not com¬
posed of two groups, information scientists and library scientists. It is a
single but wildly heterogeneous group, a motley lot of engineers, techni¬
cians, professional librarians, historians, classification-and-indexing theor¬
ists, and the like. Its work is similarly motley in character. That is what we
should expect of people studying and trying to improve different aspects of
the operations of a practical sector of society like the bibliographical sector.
And we should also expect that this group would have complex relations
with other research groups concerned in one way or another with informa¬
tion. As we have seen, its work overlaps, in part, with the social sciences; it
is itself, in part, an area of social science research and can make use of work
done in other areas of the social sciences. As a branch of technology, the
group is a borrower, not a contributor to other branches, dependent on
developments in computer technology, telecommunications, and the tech¬
nology of reprography in particular. Its historical parts are simply special¬
ized branches of intellectual or social history, and the history of technology.
BIBLIOGRAPHICAL R&D 397

The second research category, content representation, is related to, and

stands to gain from advances in, linguistics, cognitive psychology, and
artificial intelligence, but help may be a long way in the future. It seems that
there ought to be a close relation with linguistics, but linguists’ lack of
success in dealing with semantics, and their fixation on the single sentence
rather than long discourse (a situation that shows signs of changing), do not
encourage much optimism about help from that quarter. Cognitive psychol¬
ogy is still so primitive a subject that we cannot expect usable results for a
long while. Artificial intelligence seems hardly more likely to help. What we
would like, for its possible help in improving techniques for content repre¬
sentation, is deeper understanding of the phenomenon of understanding it¬
self and computer systems that at least simulate understanding at a sophis¬
ticated level. We are not likely to get them soon.

COHERENCE: TOO LITTLE OR TOO MUCH?

The information explosion is now in the past, leaving us at a much higher

plateau in terms of volume of publication. There are still plenty of problems
for research and development in the bibliographical sector. At present, the
chief motivation for support of bibliographical R&D is cost containment or
reduction. Rayward’s fears that information science might lose “disciplinary
coherence” seem to me unfounded; the whole bibliographical R&D group is
already without disciplinary coherence, but its main practical effort is in
product and process development, and I see no reason why that should not
continue. Given the rather disappointing results of research in content repre¬
sentation, I should think that the worry would be over too much coherence,
that is, continued concentration on a few unsuccessful lines of work. We
should rather hope for a profusion of attempts to go in new directions. As for
the empirical study of published documents and their use, while it is cer¬
tainly futile to pursue these studies in the hope of discovering information
laws, there is plenty to be learned that might contribute to the development
of a really interesting sociology of knowledge.
 LIBRARIANSHIP AND
INFORMATION RESEARCH
Together or Apart?

W. Boyd Rayward

Some of the assumptions that underlie my paper, assumptions essentially of

a simplifying nature, have quite properly caused its commentators some
misgivings. I have taken the view that the formation at various times of
institutions, associations, and societies, especially those that have proved to
be relatively permanent, represents important historical culminations of
scholarly and professional activity. Insofar as these bodies can be shown to
be related in terms of memberships and missions, I also infer from them the
existence of an increasingly complex historical development, of which they
represent notable stages. My initial points of reference, then, were the for¬
mation of the American Library Association, the Library Association (of
Great Britain), various bibliographical societies, the International Institute
of Bibliography, the Graduate Library School at the University of Chicago,
and the American Documentation Institute (later the American Society for
Information Science). I did not intend to present a more general history and,
indeed, explicitly eschewed discussions of the special libraries movement in
the United States (the Special Libraries Association, still very active, was
founded in 1909) and Great Britain (save for a reference to ASLIB, founded
in 1924).

DAVID BATTY AND TONI CARBO BEARMAN

Batty and Bearman are troubled by my restricted palette and small canvas,
and they sketch in some of the historical detail I neglect. Elsewhere, how¬
ever, I have discussed at some length certain general bibliographical devel¬
opments in Great Britain through the latter part of the nineteenth centui y
[Rayward, 1967a], the work of a number of individual bibliographers
[Rayward, 1977a; 1981 b], of the Royal Society for Scientific Bibliography
[Rayward, 1976], and a number of relevant international trends, to which
399
400 W. BOYD RAYWARD

changing notions of bibliographical control of scientific literature especially

were central. [Rayward, 1980a; 1981/?.] My present concern is only to sug¬
gest a framework, a perspective, for examining and suggesting the interrela¬
tions of certain occupational developments over a period of a hundred or
more years. I could perhaps have taken one or more different or complemen¬
tary approaches. I could have examined, for example, professional or schol¬
arly work of significant figures or developments relating to certain phenom¬
ena such as subject analysis and indication or other bibliographic practices. I
chose instead to confine myself to organizational culminations as reflecting
occupational and scholarly or disciplinary maturations.
One apparent misapprehension, however, in Batty and Bearman’s com¬
ments should receive attention. Given my connection with the Graduate
Library School at the University of Chicago, I am very much aware that any
discussion that I offer of its past or present may appear if not self-serving, at
least necessarily biased. I have, however, alluded to some general studies of
education for librarianship that analyze its role in bringing to the phenomena
of interest in librarianship the attitudes and techniques of the social sciences
and historical scholarship. I nowhere suggest that the school attempted to
arrive at “a special theory of librarianship and library research,” as my
commentators allege. In the United States, in the period following the
school’s foundation, librarians and others praised or vilified the school ac¬
cording to their point of view for its emphasis on rigorous critical enquiry.
To recognize this is in no way to minimize the contributions of those figures,
from Jewett to Sayers and Ranganathan, whom Batty and Bearman identify
as active between 1840 and 1940. The point might be made in this connec¬
tion, however, that for all their writing on bibliographical description and
classification, none of these figures could remotely be described as having
taken a scientific approach to their subject. In taking this approach, the
Graduate Library School represented a new attitude and spirit, though, of
course, it, too, like all of the other phenomena mentioned in my paper, did
not spring full-formed, as Minerva did, from the brow of Jove. Richardson
has explored the background to the school’s foundation, the crises—some
nearly fatal—of its adolescence, and he soberly attempts to assess its contri¬
butions through 1951. [Richardson, 1982.] Karetsky has recently examined a
notable body of research produced by the school’s early faculty and stu¬
dents. [Karetsky, 1982.]

MANFRED KOCHEN

Kochen is writing in a long and distinguished tradition. He tells us that he

was influenced by H. G. Wells’s “world brain” idea, and he adverts to the
influence of some of the early documentalists and pioneers in information
science on his thinking, He mentions Paul Otlet but seems to underrate the
influence of Otlet’s ideas in Europe and those of men like Watson Davis in
LIBRARIANSHIP AND INFORMATION RESEARCH 401

the United States. And, of course, the paths of Otlet and Wells (and Watson
Davis) intersected on at least one notable occasion: The World Congress on
Universal Documentation held under the auspices of the Institute for Intel¬
lectual Cooperation in Paris in 1937.
But no doubt, Leibniz was a forerunner; in the last third of the seven¬
teenth century, he was struggling to find a universal characteristic, devise a
calculus of reason, and create an encyclopedia that would “be the great
instrument of bringing civilization to its highest powers.” (McRae, 1957, p.
38.) In this encyclopedia were to be unified all the arts and sciences; the
encyclopedia would not be merely ‘‘the elaboration of an ordered inventory
of various intellectual possessions. It was absolutely essential for the radical
reform and advancement of the sciences.” (Ibid. p. 42.) Similar to the world
brain and Leibniz’s encyclopedia was the notion of an Office of Publick
Addresse that animated a group of men of letters in mid-seventeenth-century
London at the time of the Civil War and the Commonwealth. How at last to
create Bacon’s House of Solomon; how to mobilize the intelligence of the
day, what we now generically call information, for purposes of social
amelioration; how to coordinate, unify, and allow for the orderly progres¬
sion of the sciences; above all, how to bring about the city of God on earth—
these were among the many religious, scientific, political, and social matters
that Samuel Hartlib, William Petty, and John Drury, among the forerunners
and founders of the Royal Society, wrote about and discussed among them¬
selves and with the great Comenius, visiting London from Bohemia. The
many functions of the Office of Publick Addresse that they variously de¬
scribed were a fascinating combination of a modern employment bureau, the
yellow pages of the telephone directory, classified advertisements in news¬
papers, the community information service offered today by many public
libraries, an information analysis center, a research library, a publishing
house, and a residential institute of advanced study. [Turnbull, 1947; Web¬
ster, 1970.]
Kochen’s WISE and WISDOM schemes are related to a long tradition of
utopian concern for the unification and mobilization of knowledge, of specu¬
lation about ideal forms and uses of encyclopedia. Kochen, however, has
turned away, it seems—with what comes across as dismissive impatience—
from recorded or public knowledge and from systems to make it accessible.
Unlike Leibniz and that group of energetic, enlightened men of the English
Commonwealth, unlike Otlet and Wells, his concern is not with what is
known and publicly available in a readable format in books, scientific pa¬
pers, documents, or even by extension on magnetic tapes, not with the
objective information they contain. His interest is aroused by something
more abstract and subjective, a particular, hitherto elusive quality of the
knower—wisdom. Moreover, he deals with the idea of wisdom in no tradi¬
tional way; his is not the Socratic ideal. For him, wisdom is the product of a
system. WISDOM is a system that will have the capability of making men,
presumably with or without their assent, all knowing and wise, within, we
402 W. BOYD RAYWARD

assume, the constraints of the system. There is something chilling, Orwel¬

lian about the notion of a wisdom system; can it, we wonder, entertain
doubt, as men do, as to what ultimately is wise and what foolish. Kochen,
however, speaks of the system as being able eventually “to supplant the
minds of individuals, a kind of community mind that may be the next stage in
the evolution of mind.” “Wisdom,” he observes later, “governs communi¬
cation of information for purposes of control.” While it is difficult to know
what much of this actually means, it provides a glimpse of a nightmare world
of social and psychological engineering.
Some of Kochen’s incidental comments suggest that statements I had
framed with some care are not clear. I do not suggest, for example, that
information science should focus primarily on library-and-information work.
I note only that information science raises problems of definition that infor¬
mation work does not. A number of his comments also suggest that my
conclusion that “a certain combativeness” on the part of information scien¬
tists and librarians has changed, is not justified in his case. He notes the
declining opportunities for library-school graduates, the fiscal plight of
public libraries, the declining importance of libraries as publishers’ markets,
the existence of new kinds of information jobs, the calls that have been made
for library schools to remove library from their names, and trends towards a
“paperless” society. I myself alluded to some of these complex matters
when I affirmed that change and accommodation were occurring. Kochen,
however, refers to them in such a way, and without much-needed explana¬
tion or further interpretation, that they seem like shots from an arsenal of
antilibrarianship rhetoric. Presumably, these matters are intended to show
that library schools are just about finished—unless they cease being library
schools; that information scientists in the company of librarians, amiable
and hospitable though librarians be, are strangers in a strange land; and
above all, that librarians do not, indeed cannot, partake of the central mys¬
teries of WISDOM. This last is explicit. Ultimately, we are told, “concern
for the written record,” and “understanding of such institutions as libraries,
media centers, etc.,” are remote from “bringing knowledge and understand¬
ing to bear on shaping our world toward human ends by human means.”
This, in Kochen’s view, is the central thrust of the information disciplines.
The information disciplines (it is comforting that there is more than one)
constitute, it seems, a new “pansophia” and in themselves presumably will
be able to do what in the past a multitude of special disciplines separately
aspired to.

JESSE SHERA

Shera provides a welcome relief from this kind of thinking. His comments,
graceful and penetrating, amplify the historical perspective I attempted to
provide and offer some interesting data about the composition of the mem-
LIBRARIANSHIP AND INFORMATION RESEARCH 403

bership of the American Society for Information Science. Shera’s chief con¬
cern is to assert the importance of libraries in just those areas that Kochen
denies them relevance, at just the point where WISDOM, a kind of deus ex
machina, supervenes. The library, in Shera’s view, is a social agency, an
instrument of social communication; it is also humanistic in that its “charac¬
teristics, modes of access to its resources, uses, and values are humanistic.’’
Shera is as disillusioned with information scientists as Kochen is disdainful
of librarians. Both scholars, it seems to me, adopt a narrow, perhaps stereo¬
typed, certainly prejudicial view, which I had hoped the perspective I at¬
tempted to provide might obviate. The reasons that Shera gives for abandon¬
ing his early hope that information science would provide “an intellectual
and theoretical foundation” for librarianship seem to me not to be particu¬
larly convincing. They are related to the uses to which the computer has
been put. Certainly, Shera seems to agree that the advent of the computer
precipitated the rapid development of information science and was wel¬
comed by many librarians. But he cites gloomily Kaplan’s Law of the Instru¬
ment and further suggests that information scientists are preoccupied with
the “design, production, implementation, and control of the system.
Under their influence, library schools have lost their sense of the historic
and humanistic mission of the library; they are turning out control artists
and “data mechanics.” Insofar as this has been true (and I am not sure that
it has been), it need not continue. Shera is offering more of a warning than an
argument, for he does accept that librarianship as a form of service a
practice—does interrelate with information science as “an area of inquiry,
of research.”
Libraries are concerned with “sociological and psychological phenomena
not with physical objects and processes,” Shera asserts. Their goal is the
growth of knowledge, and a theory of librarianship, he believes, should be
founded on the concept of “symbolic interactionisms.” If Shera’s position is
at all tenable, then as information science moves beyond data systems,
beyond the gadget and the flickering image, from the thing to the idea, from
information to a concern for understanding and knowledge, it approximates
closer and closer Shera’s ideal librarianship. We are presented with the
irony that Kochen’s WISDOM system may be no more than an idealized
description of the library, perhaps the library of the future, as Shera con¬
ceives it. Here, then, in the terms of my paper, is a convergence with a
vengeance.
Shera’s final warning that control of librarianship as a profession is
threatened by information science does not seem to me to be justified. In a
way, Kochen’s and Shera’s papers are complementary in illustrating the
misunderstanding, hostility, even a certain rivalry, that have existed be¬
tween library and information scientists, for reasons that I find difficult to
understand; there seem to be class or status elements of an interesting kind
involved. In any case, these two papers reaffirm my view, speaking from the
standpoint at least of librarians, that broader perspectives are needed to help
404 W. BOYD RAYWARD

librarianship and the enormously wide variety of libraries that now exist to
adapt, change, grow, even further diversify. Such perspectives will help
them in experimenting with, and in incorporating, new kinds of expertise
and technology; in taking shape according to new and reevaluated old
philosophies and ideals of service; and in drawing on a more developed kind
of scholarship. Let all of these come from where they may.

PATRICK WILSON

Before commenting specifically on Patrick Wilson’s paper, I should make it

clear that my view of information science and of librarianship is not that of
one hoping to make rigorous distinctions between professional occupation,
basic and applied research, science, and technology. I do not think of infor¬
mation science or library science as scientific disciplines. (Wilson suggests I
think of information science in this way.) I speak of occupations and a
related scholarship, of disciplines nourishing professional activities. Thus, I
have no quarrel with Wilson’s rather elaborately conducted refinement of
my argument to show that it is library science, not librarianship, that may be
conceived of as converging toward information science. He supports this
with a historical argument similar to mine.
Wilson, in effect, starts from an unusual and very general perspective. He
eschews various traditional distinctions that had to some degree been my
concern and puts forward the notion of the existence of an overarching
bibliographic R&D community. This is a striking and useful idea. Wilson’s
analysis of the kinds of work carried on within this community is masterly,
and the six categories of work he describes help organize conceptually a
scattered and confusing literature. He concludes that the R&D group is “not
unified by shared problems or methods, but is divided into numerous small
groups working in many different ways at strikingly different sorts of tasks.”
He is concerned “to avoid talk that suggests two distinct and distinguishable
groups, . . . information scientists and library scientists. It is a single but
wildly heterogeneous group, a motley lot of engineers, technicians, profes¬
sional librarians, historians, classification and indexing theorists, and the
like. Its work is similarly motley in character.” It has not attained, nor need
it fear attaining, disciplinary coherence; its aim is mainly practical, “product
and process development.”
This notion of Wilson’s deserves looking into. In effect, he uses it to avoid
any real definition or description of information science in general. He has
some fun distinguishing the kinds of information scientists whose work is
represented in the Annual Review of Information Science and Technology
from other species of the genus. Eventually, he argues that a branch of
information science that I refer to but do not name cannot be effectively
distinguished from library science. Within the bibliographic R&D commu¬
nity, however, it seems to me, are firmly located the various groups I dis-
LIBRARIANSHIP AND INFORMATION RESEARCH 405

cussed. Note that save for the engineers and technicians (and then presum¬
ably not all of them), those whom he mentions in that wildly heterogenous
group would probably describe themselves as librarians or information sci¬
entists. In schools of library and information science (as well as outside of
them, of course) work is regularly conducted in all of the areas Wilson
identifies. But is there more? Wilson speaks of the complex disciplinary
relations members of this group have with others concerned with informa¬
tion and of what may and may not be hoped from the latter who are clearly
not members of the bibliographic R&D community; they are those con¬
cerned with more conventionally conceived sciences and engineering. Their
interest in information seems to complicate for Wilson the question of
definition, or perhaps, more accurately, “designation.”
It seems to me that members of the bibliographic R&D community are
principally library and information scientists, that those who work within it,
whatever their origins, have come to be so designated. The literature pro¬
liferating within it, whether reporting research or development, furnishes the
basis for a multifaceted discipline. Many of those working in the community
have affiliated themselves in the ways I have described. In my view, this
community is not so unorganized, so apparently chaotic as Wilson suggests.
Its research component struggles to cumulate more and better data about a
range of constantly evolving problems, refine our understanding of these
problems, and translate what is known into better practice. (This is more
than process and product development.) Wilson raises, but does not solve,
the problem of how library-and-information science, whatever the extent of
its development, interpenetrates and is developed by the bibliographic R&D
community. It is not enough simply to say or imply that the problem does
not exist. The existence of this community—insofar as the disparate ele¬
ments Wilson describes are not so disparate as to make the idea of a commu¬
nity a ludicrous fiction—does not obviate the various concerns I dealt with
in the last part of my paper; it simply provides a different context for them.
The existence of such a community raises questions of definition, historical
issues of growth and change, contemporary problems of disciplinary assimi¬
lations, and so on. There are, too, occupational issues of how the biblio¬
graphic R&D community, and bibliographic R&D itself, are related to pro¬
fessional practice in libraries and other institutional settings. Wilson s is a
provocative paper built on an intriguing notion. For a community to exist, let
its members work on as many different problems, employing as many differ¬
ent methods as they can devise, the presence of certain commonalities is
required: shared beliefs as to how the problems interrelate, some sense of
common goals, however complex and fragmented. Wilson stresses differ¬
ences. The question of similarities is ignored. Maybe Wilson is simply pro¬
viding his own name for what nowadays would be otherwise described as
library-and-information science (and technology)!
SECTION 6
 CYBERNETICS

Murray Eden

Norbert Wiener, in his book entitled Cybernetics, or Control and Communi¬

cation in the Animal and the Machine, did not define explicitly the word he
believed he had coined. [Wiener, 1948 and 1961.] It had, in fact, been used
before: by Andre-Marie Ampere in his Essai sur la philosophie des sciences.
[Ampere, 1838 and 1843.] Ampere, in a work typical of the impulse for
classification that motivated natural scientists of his day, exhibited a taxon¬
omy of “all human knowledge” and in the same spirit provided neologisms
for the sciences he derived in his classification.1 However, Ampere, despite
his statement that he was generalizing the concept of steering, was not aware
that it could be extended to the regulation of organismic behavior. He as¬
signed to cybernetics the role of a subscience of government. His
classification of biology contained no niche for control, nor for that matter
did his classification of physics.
In the introduction to his intensely personal book, Wiener presented his
rationale for choosing to name a particular “no-man’s land between the
various fields [of science]”:

There are fields of scientific work, . . . which have been explored from the
different sides of pure mathematics, statistics, electrical engineering, and

'Ampere gave the following description of what he meant by cybernetics: “Cybernetics

[cybem6tique]. Relations between peoples, the subjects of study within the two sciences dis¬
cussed earlier [international law and diplomacy] are only a small part of that with which a good
government must concern itself. Maintenance of public order, administration of laws, equitable
distribution of taxes, selection of the people it must employ, and all the other considerations
that may contribute to improving the state of society require the continual attention of govern¬
ment. Choices must constantly be made, among diverse measures, about which measure is most
appropriate to achieve the desired goal. Only by intensive study and comparison of the various
elements that, for each choice, are provided by a knowledge of all that is relevant to the
nation—its character, customs, opinions, history, religion, way of life and property, institu¬
tions, and laws—can government create the general rules of conduct that must guide it in regard
to each particular case. Therefore, it is only after all the sciences that are concerned with these
various factors that one must place the science in question here. I would call this science
cybernetics from the word icu(3epvf|TTi‘;. From the restricted definition for the art of steering a
vessel, cybernetics took on a meaning—even among the Greeks—of the art of steering in
general.” (Ampere, 1838 and 1843, pp. 140-141; my own translation—M. Eden.)

409
 410 MURRAY EDEN

neurophysiology; in which every single notion receives a separate name from

each group, and in which important work has been triplicated and quadru¬
plicated, while still other important work is delayed by the unavailability in one
field of the results that may have already become classical in the next field.
(Wiener, 1948 and 1961, p. 2.)

CHIEF ELEMENTS OF WIENER'S CONCEPTION

In an historical account, Wiener enumerated the ingredients that led to his

notion of cybernetics, which forms the nexus for their application to descrip¬
tions of animate and inanimate systems. I sketch the line of his interests and
disciplines he coopted along the way. It begins in the late thirties by reason
of his association with the neurologist, Arturo Rosenblueth. In 1940, the war
required that he turn his attention to “the development [by Vannevar Bush]
of computing machines for the solution of partial differential equations” and
the design of electric networks [in collaboration with Yuk Wing Lee],
Wiener concluded that scanning as employed in television was destined to
be useful to engineering, especially for the representation of functions of
more than one variable. In modern parlance, it might well be called image
analysis or graphics. These interests taken together led him to the notion of
the modern ultra-rapid computing machine.” Another war project, the
prediction of aircraft flight paths, led him to work on the theory and technol¬
ogy of prediction [with Samuel Caldwell and Julian Bigelow], hence, to the
generalization of the notion of feedback and the belief that control theory
was applicable to the explanation of neuromuscular behavior. During the
course of this work, it became clear to Wiener and Bigelow “that the prob¬
lems of control engineering and of communication engineering were insepa¬
rable . . . .” (Ibid. p. 8) Wiener and Bigelow focused on the “fundamental
notion of message . . . precisely what is called a time series by the statis¬
ticians. . . . The solution of the problem of optimum prediction was only to
be obtained by an appeal to the statistics of the time series to be predicted
(Ibid., pp. 8—9.) And a bit later: “It occurred to us . . . that there was a whole
region of engineering work in which similar design problems could be solved
by the methods of the calculus of variations.” (Ibid., p. 10.)
Wiener and Bigelow often found value in the concept of “a message
contaminated by extraneous disturbances which we call background noise."
And thus, “we have made of communication engineering design a statistical
science, a branch of statistical mechanics.” This notion “attaches itself very
naturally ... to entropy” and to the second law of thermodynamics. The
second law enters “in the study of enzymes and other catalysts, and their
study is essential for the proper understanding of such fundamental phenom¬
ena of living matter as metabolism and reproduction. The third fundamental
phenomenon of life, that of irritability, belongs to the domain of communica¬
tion theory.” (Ibid., p. 11.)
CYBERNETICS 411

We may criticize Wiener for claiming too much in the way of priority for
himself and his collaborators. Here and there, he cites collateral and earlier
contributors, but he was not writing a history of scientific ideas. The pro¬
gression was natural enough for a mathematician and statistician in that
time. His continuing interest in biology and neurology—almost an avoca¬
tion—led him to the connections that are so obvious nowadays. They may
not have been obvious in the early 1940s when he was meeting with biolo¬
gists and physicians in the Josiah Macy Jr. Foundation meetings.2 3
At various places in his book, Wiener provides statements of immediate
relevance to the question of interdisciplinary points of contact.

Symbolic logic and semeiojogy:

If I were to choose a patron saint for cybernetics out of the history of science, I
should have to choose Leibniz. The philosophy of Leibniz centers about two
closely related concepts—that of a universal symbolism and that of a calculus
of reasoning. From these are descended the mathematical notation and the
symbolic logic of the present day. (Wiener, 1948 and 1961, p. 12.)

Automata theory:

The newer study of automata, whether in the metal or in the flesh, is a branch
of communication engineering, and its cardinal notions are those of message,
amount of disturbance or “noise”—a term taken over from the telephone
engineer—quantity of information, coding technique, and so on. (Ibid., p. 42.)

Ergodic theory and group theory:

... in order to appreciate the real significance of ergodic theory we need a

more precise analysis of the notion of invariant, as well as the notion of trans¬
formation group. These notions were certainly familiar to Gibbs, as his study
of vector analysis shows. Nevertheless, it is possible to maintain that he did
not assess them at their full philosophical value. (Ibid., p. 50.)

2 Ludwig von Bertalanffy the biologist and inventor of the term general system theory has a
somewhat different perspective. [Bertalanffy, 1968.] “It is true that cybernetics was not without
precursors [in biology]. Cannon’s concept of homeostasis became a cornerstone in these con¬
siderations. Less well-known, detailed feedback models of physiological phenomena had been
elaborated by the German physiologist Richard Wagner in the 1920s, the Swiss Nobel prize
winner W. R. Hess (1941, 1942) and in Erich von Holst’s Reafferenzprinzip. The enormous
popularity of cybernetics in science, technology and general publicity is, of course, due to
Wiener and his proclamation of the Second Industrial Revolution. (Bertalanffy, 1968, p. 16.)
3 Wiener was a bit wide of the mark here. The current progress of automata theory is to a very
large extent austerely mathematical and draws most heavily on mathematical logic and the
theory of computation.
 412 MURRAY EDEN

The computer as a model for the nervous system:

... it became clear to us [N. Wiener and Walter Pitts] that the ultra-rapid
computing machine, depending as it does on consecutive switching devices,
must represent almost an ideal model of the problems arising in the nervous
system. (Ibid., p. 14.)

The nervous system as a model for the computer:

... the modern ultra-rapid computing machine was in principle an ideal central
nervous system to an apparatus for automatic control . . . .” (Ibid., p. 26.)

Robotics:

To sum up: the many automata of the present age are coupled to the outside
world both for the reception of impressions and for the performance of actions.
They contain sense organs, effectors, and the equivalent of a nervous system
to integrate the transfer of information from the one to the other. They lend
themselves very well to description in physiological terms. It is scarcely a
miracle that they can be subsumed under one theory with the mechanisms of
physiology. (Ibid., p. 43.)

Pattern recognition:

How do we recognize the identity of the features of a man, whether we see him
in profile, in three-quarters face, or in full face? How do we recognize a circle
as a circle, whether it is large or small, near or far; whether, in fact, it is a plane
perpendicular to a line from the eye meeting in the middle, and is seen as a
circle, or has some other orientation, and is seen as an ellipse? (Ibid., p. 133. )

It is fair to say that here Wiener is not particularly successful, although he

touches on a variety of approaches that have been explored by later students
of pattern recognition. In fact, pattern recognition draws on some of the
mathematical disciplines listed by Wiener as relevant to cybernetics_
notably statistics, information theory, graph theory, and group theory.
Wiener even touches on artificial intelligence in the discussion of a chess¬
playing machine:

I think it is possible to construct a relatively crude but not altogether trivial

apparatus for this purpose. The machine must actually play—at high speed if
possible all its own admissible moves and all the opponent’s admissible
ripostes for two or three moves ahead. To each sequence of moves it should
assign a certain conventional valuation. Here, to checkmate the opponent
receives the highest valuation at each stage, to be checkmated, the lowest;
while losing pieces, taking opponent’s pieces, checking, and other recogniz¬
able situations should receive valuations not too remote from those which good
players would assign them. The first of an entire sequence of moves should
CYBERNETICS 413

receive a valuation much as von Neumann’s theory would assign it. (Ibid., p.
165.)

Not at all a bad start on the problem, even though Wiener did not foresee
that much more powerful computers (with the help of some programming
ingenuity) would be made to search the move tree to greater depths than two
or three.
To clarify this heterogeneous set of steps, I shall attempt a parsing of the
sequence in Table 1. The last five ideas of Table 1 appear to define cybernet¬
ics according to Wiener. Most important, therefore, are two related insights.
First, that there is an essential unity in the set of problems in communica¬
tion, control, and statistical mechanisms (noisy phenomena), whether they
are to be studied in the machine or in living tissue. Second, that “the com¬
puting machine must represent almost an ideal model of the problems arising
in the nervous system.” We shall return to this second issue.

RECEPTION OF WIENER'S IDEAS

Wiener went on to document by anecdotes the excitement and interest his

idea engendered in a number of eminent scientists of widely different back¬
grounds and competence. In the second edition of his book [1961], he was
able to add to the evidence for the widespread diffusion of the concepts
embodied in the term cybernetics. Indeed, the concept was embraced ea¬
gerly by many scholars, especially in psychology and the social sciences; its
acceptance was more muted in the purely physical sciences.
Wiener was himself uncertain about the applicability of his ideas to sociol¬
ogy and economics:

It is certainly true that the social system is an organization like the individual,
that it is bound together by a system of communication, and that it has a
dynamics in which circular processes of a feedback nature play an important
part. This is true, both in the general fields of anthropology and of sociology
and in the more specific field of economics .... Drs. Gregory Bateson and
Margaret Mead have urged me, in view of the intensely pressing nature of the
sociological and economic problems of the present age of confusion, to devote
a large part of my energies to the discussion of this side of cybernetics.

Much as I sympathize with their sense of the urgency of the situation, and
much as I hope that they and other competent workers will take up problems of
this sort ... I can share neither their feeling that this field has the first claim on
my attention, nor their hopefulness that sufficient progress can be registered in
this direction to have an appreciable therapeutic effect in the present diseases
of society. (Ibid., p. 24.)

Wiener goes on to explain why he is pessimistic about the use of cybernet¬

ics in these fields. Wiener’s argument is that “the main quantities affecting
 <D
D-
>> CQ
H

neuromuscular
a
u
p
Dh

Explaining

behavior
c

cd
o>
"O
<D
a

Af
>, hn
H

a>
s
a
o
<U
>

2 Control
theory
<D
Q 13
O
-Q
03

<u
Cl
>,
H

J*i
U a>
o >
<L>
o £ C/3
a
u
e cd o 01)
p
Oh
O .cd
c ao ^
c *5 Id <L>
22
c oj C 50
o cd
C/3 att *C <u • 2 x>
03 .S Cd
a f<u 'Zn o 5- 2 *c T3 u
-5 o ?£ <L> <D n> C *2
3 Cd v a
u. .2
D
■4—*
c/d -a Q co c: > u 0-. cd
C/D U3
a <D
<D JO
Q.
>>
H
•o
cd

Glj J2
c 2 ^ m
<D
P <u I u + ^ c
O Q
E "u c p
a D C cd
jo a- *s O c/3
o <N
cr'-S
<L> E ^
C/3
+ o -J3
> o « g > <u
u.
<L>
U z g H O*
Q

414
 <u
o
c
1)
<v G
> .o>
G-
c
o
V , ,
Q. o
C/5 <u
D
u fco
o
OD 0)
O X <3
<D
O -g
<u
X) X
C/5
c o
-o U
X
03 g
cd <u
C/5 <D X
o a. ■*“*
‘X a c
o3 as Gn

■■ ■
— O
a £
£
E
■— ■o
G<
C5JJ
o X
3
-a
c
o
C/5
a
c G
X X <u
o ■*—' c
<u a> <D

0) sC/5
£
o
c
c
<D 0
•o c
‘g
C/5
03 <u
*0.
T3 o
C “D E
o3 G
<D C/5
c
§ <u
C/3
X
f-T £ H
o
a -Cl

415
 416 MURRAY EDEN

society are not only statistical, but the runs of statistics on which they are
based are excessively short.” He hints at the issues of ergodicity, stationar-
ity, and changes of state:

There is no great use in lumping under one head the economics of steel industry
before and after the introduction of the Bessemer process, nor in comparing
the statistics of rubber production before and after the burgeoning of the auto¬
mobile industry and the cultivation of Hevea in Malaya. Neither is there any
important point in running statistics of the incidence of venereal disease in a
single table which covers both the period before and that after the introduction
of salvarsan, unless for the specific purpose of studying the effectiveness of
this drug. (Ibid., pp. 24-25.)

Curiously enough, he did not bring up the issue of purposiveness, which

plays such a crucial role in the early paper of Rosenblueth, Wiener, and
Bigelow [1943]. There, the authors argued for a particular restriction of the
idea of “intrinsic purposive behavior,” which they considered to be “tele¬
ological, and it was on this basis that they searched for certain restricted
comparisons of living organisms and machines. Purposiveness is a slippery
subject even within these two kinds of objects, but what can be said of
purposive behavior in the social sciences?4
Nonetheless, in the decade of the 1950s and most of the 1960s, the cyber¬
netics principles were taken up widely and many books were published.
Titles of books in print in 1968 include, according to a book index,5 Philoso¬
phy and Cybernetics; Cybernetics and Management; The Social Impact of
Cybernetics; Cybernetic Modelling: The Science of Art; Great Ideas in In¬
formation Theory, Language, and Cybernetics; Cybernetic Principles of
Learning and Educational Design; even The Cybernetics of the Sacred and
The Cybernetic ESP Breakthrough. Only one book is avowedly a textbook
for cybernetics: a book by Viktor M. Glushkov, Introduction to Cybernetics
[1966],
Despite Wiener’s caveats and those of others with respect to the inappro¬
priate use of the concepts of cybernetics, it seems abundantly clear that the
underlying ideas were so attractive that many scholars untrained in mathe¬
matics and statistics attempted to apply the techniques to which Wiener
alluded in ways that were wildly inappropriate. By the late 1960s, the term
cybernetics, though not the underlying concepts, had fallen into some disre¬
pute. John F. Young, a British psychologist, attempted to clarify the psycho¬
logical use of cybernetics, but in his book, Cybernetics, published in 1969,
he felt impelled to justify himself:

4 For a discussion of the philosophical aspects of their approach the reader may wish to consider
the comments of Berlinski [1976].
5Nina R. Thompson, ed., Cumulative Book Index 1967-1968 (New York: H. W. Wilson Co.
1971), vol. 1, p. 814.
CYBERNETICS 417

It is unfortunate that many of the books on cybernetics which have been

published fall into one of three categories. One type treats the reader as a
moron who is to be astounded and so, it is hoped, impressed. Another treats
the reader as though he were a potential employer who is to be baffled and so, it
is hoped, impressed (and also, if possible, be made apprehensive and therefore
generous). The third type of book includes the word ‘cybernetics’ in the title,
but inside it apparently deals with an entirely different subject. It is hoped that
the present book cannot be pigeon-holed into any of these categories.

A notable feature of cybernetics is the extent to which pseudo-mathematical

tracts have appeared. Strangely, much of this writing has been produced not by
mathematicians but by psychologists. Unfortunately, some of this ‘mathemat¬
ics’ appears to be intended merely to impress and sometimes it has little rele¬
vance to any practical results which have been obtained. Possibly this is partly
caused by the largely irrelevant mathematical chapters which Wiener inserted
in his book on cybernetics. Whatever the reason, there has been a great deal of
mathematical mumbo-jumbo associated with the subject. Fortunately, real
mathematicians, physicists and engineers concerned with cybernetics have
tended in general to be more restrained in their treatments.

The situation is by no means ideal, and there is a danger that cybernetics will
become generally regarded as an up-to-date form of Black Magic, as a sort of
twentieth century phrenology. If this unfortunately happens, then one whole
field of research, at the border of many disciplines, will be robbed of the
support it needs. (Young, 1969, pp. ix-x)

Fourteen years later, it must be said that the situation has not improved and
that in large measure Young’s melancholy prophecy in the last paragraph of
his preface has been largely fulfilled.
Although I have not reviewed in more than a cursory way Russian litera¬
ture available in English, it may be worth noting the somewhat different
course of cybernetics in the Soviet Union. Wiener, who was a humanist and
something of a socialist in his political leanings, was much disturbed by the
original reception of his work in the Soviet Union. It appears to have been
relegated to the category of bourgeois science and, hence, dismissed. How¬
ever, following the defrocking of Stalin as a scientist, Wiener’s cybernetic
principles were firmly endorsed. In a manner that may be typical within
Russia’s science establishment, cybernetics—almost overnight—became a
fundamental discipline; institutes were created, academic departments es¬
tablished, ajournal was initiated, courses and textbooks were prepared. One
striking difference remained. Western physical scientists and engineers
rarely used cybernetics in titles of papers and referred, instead, to the mathe¬
matical name of the technique being applied to their research: control, time-
series analysis, statistical communication theory, automata theory, and so
forth. Russian literature tended to use cybernetik whenever any of these
techniques were being applied. Science Citation Index for 1980 includes
papers of Russian origin, with such titles as “Making Ultrasonic Inspection
of Welded Joints More Efficient by Means of Cybernetics’’ and “Group
 418 MURRAY EDEN

Identification of Materials in Gas Chromatography on the Basis of Cyber¬

netic Models.”
Even in Russia the term is now not nearly so widely used as it was a
decade ago. For example, in the Science Citation Index for 1980, there are
only twelve papers indexed under cybernetics or one of its modifications. By
contrast, entries under cycles, cyclic processes, and cyclic have more than
1600 citations.6 Of the twelve papers cited under the keyword cybernetics,
eight are from the Russian literature.

LIVING SYSTEMS

Putting aside the curious story of the linguistic depreciation of the word
cybernetics, what of the ideas behind the term, especially with respect to
mind and the nervous system? Every age compares life with its contempo¬
rary machinery.
Consider how interpreters of mind and nerve attempted to explain these
phenomena. A central concept for Aristotle was pneuma. I draw the follow¬
ing quotations from the biologist, Thomas S. Hall, who has written what he
calls a ‘‘preliminary account” of the “difference between living and nonliv¬
ing things.” (Hall, 1969, p. viii.)

How Aristotle viewed pneuma has occasioned considerable scholarly debate.

... Its nature is partly suggested by Aristotle’s reference to it as “hot air” or
simply a “hot substance”—but with the understanding that its heat is no
ordinary fire-produced heat. . . .

A. L. Peck supposes that Aristotle means to make pneuma a quite general

intermediator between psyche and soma, a role which would give it great
importance in life-matter relations. (Ibid., p. 115.)

Further, Hall writes:

Diodes (ca. 300 B.C.) and Praxagoras (ca. 300 B.C.) had given the psychic
pneuma its source in the heart whose arteries distribute it to the body generally
whereas the veins carry blood. . . .

It remained for Praxagoras’ pupil Herophilus to take the giant stride of identify¬
ing the motor nerves and their role in efferent conduction, as well as the
sensory nerves which he probably regarded as conductors of a sensory
pneuma. The important point to note here is that Herophilus and his successor
at Alexandria, Erasistratus, now returned the brain to the position of central
organ of sensation and volition. Herophilus showed through dissection that
nerves arise from the brain. Erasistratus’s doctrine is already familiar to us.

I chose cycle because the concept of cycles is important within cybernetics and made a
correction for purely chemical terms such as cyclic AMP. Science Citation Index, part 10,
permuterm subject index, columns 14496-14497, ISI Inc., Philadelphia, Pa. [1981],
CYBERNETICS 419

The arteries distribute vital pneuma to the body generally, he said, including
the brain. Here psychic pneuma is formed and travels outward over the nerves.
Galen’s ideas are notable especially because of the extent to which actual
experimental evidence influenced his opinions. (Ibid., pp. 161-162.)

We are tempted to think that the Greeks purified their writings from the
taint of technology much as modern mathematicians delete their marginalia
and numerical examples, which detract from the elegance of their papers.
Nevertheless, there is at least circumstantial evidence of a technological
marker for pneuma. A few centuries after Aristotle, Hero of Alexandria, in
his work Pneumatica, is reported to have given several examples of a kind of
apparatus that may be the prototype of existing engines. Steam or steam-
water-mixtures operated mechanisms that could move temple doors without
human intervention. We may also note that Aristotle’s use of pneuma is pure
metaphor, but experimental findings were beginning to introduce their con¬
straints on the model. “As for method, although Galen’s experiments were
often inconclusive, they were also numerous and relevant and were not to be
matched in ingenuity until the seventeenth century.” (Ibid., p. 163.)
The age-old vitalist/mechanist argument in biology is well known. But we
are not here concerned with the argument per se but rather the role of the
automata or pseudoautomata that have been known in history and myth as
models for mechanistic thought.

An interesting point of scholarship is the extent to which mechanistically ori¬

ented physiologists were influenced, in constructing a new image of man, by
the automated imitations of vital action familiar, as we have just noted, since
earliest times. One recent careful student of this question, Derek J. de Sofia
Price, considers the connection a real and important one. (Ibid., pp. 223-224.)

Descartes was the most explicit early proponent of the mechanistic view.
He closes his Treatise of Man with the following paragraph:

I desire that you should consider that all the functions that I have attributed to
this machine [he here enumerates the functions] . . . imitate as perfectly as
possible those of a real man; I desire, I say, that you should consider that all
these functions follow quite naturally solely from the arrangement of its parts
neither more nor less than do the movements of a clock or other automaton
from that of its counterweights and wheels; so that it is not necessary to
conceive in it any other principle of movement or of life than blood and spirits
set in motion by the heat of the fire that burns continually in its heart and is of
no other nature than all those fires that are in inanimate bodies. (Descartes in
Le Monde, published posthumously 1664; as quoted by Hall, 1969, p. 262.)

Referring to Descartes, Harvey, and other seventeenth-century scien¬

tists, Hall writes:

... we need to realize that biology became mechanistic simultaneously on two

levels, the meso- and micromechanical. At the mesomechanical level, the
420 MURRAY EDEN

whole organism and its main working parts were viewed as machines. For
Descartes (1635) the whole animal was an automaton and so were such working
components as the heart and the various pairs of reciprocating muscles. Borelli
(1685) had a comparable attitude toward the bird and its wings; the whole bird
was a flying machine of which the wings are necessary working parts. Harvey,
writing earlier than Descartes, compared the propulsion of blood through the
heart to that of water through a water bellows.

At the m/'cromechanical level we encounter numerous devices, of which the

most elaborate are, perhaps, those which Descartes imagined in the brain,
apertures which, by opening and shutting and by being movable so as to face in
various directions, control the flow of spirits between the ventricles and the
nerves. In a somewhat more objective but still speculative spirit, Hooke (1665)
suspected that his cells might be equipped with small mechanical contrivances.

Some theorists carried the mechanical analogy even beyond the level of things
potentially visible through a microscope. Thus both Descartes and Giovanni
Borelli . . . extended mechanical models all the way to the unit particle (Hall
1969, pp. 228-229.)

The sources of analogies for neurology or psychology have undergone no

great changes from the seventeenth century to our own. In 1922, Knight
Dunlap, professor of experimental psychology at the Johns Hopkins Univer¬
sity, in a text entitled The Elements of Scientific Psychology, used two
analogies: (1) the burning of a fuse to describe the production of nerve
current and (2) a telephone exchange as the metaphor for the structure and
perhaps function of brain and nerve.

If we have a series of neurons, with a fiber of the first in contact with a fiber of
the second; a fiber of the second in contact with a fiber in the third, and so on;
and if we stimulate the first neuron properly, it will stimulate the second, the
second will stimulate the third, and so on. Obviously, some “process” passes
through the neuron, passing into the cell body through one fiber, and out
through another: because a neuron excited at the tip of one fiber will excite
another neuron with which another of its fibers is in contact. This “process” is
called “nerve current.” Its exact nature is unknown, and is the subject of
various theories, but it is analogous to the burning of the powder in a fuse.
Suppose we lay a number of short pieces of fuse end to end, and light the free
end of the first. The combustion process will run through the first piece, ignite
the second, and so on through the whole line. Yet nothing travels, except the
process of combustion. The chief differences between the action of this line of
fuses, and the action of a chain of neurons are that the neuron has a rapid
recovery, becoming quickly ready for another discharge; and that the neuron,
in many cases, has numerous branches of its fibers in contact with many other
neurons, and stimulates, at a given discharge, some of these without stimulat¬
ing the others. (Dunlap, 1922, p. 177.)

The functioning of afferent and efferent neurons is relatively simple. Current

due to stimulation of a given receptor can enter the cord or the brain stem at
CYBERNETICS 421

only one point. Current emerging from the cord or brain stem at a given point
can go to only one effector or effector group. The central neurons, however,
provide a multiple switching system, by which an afferent route can be con¬
nected with any one of several efferent routes. If we consider the cord, brain
stem and hemispheres together, they may be compared to the exchange of a
telephone system, through which any calling phone can be connected with any
other phone. The difference is that the “sending” (receptors) and “receiving”
(effectors) stations in the nervous system are distinct; whereas they are com¬
bined in the telephone system. The nervous system is a “one-way system.”
But within this system a given receptor can be connected with almost any
effector, or with several at the same time; and conversely, a given effector may
have connected to it any one of a wide range of receptors or various combina¬
tions of receptors. (Ibid., pp. 183-184.)

This was the state of affairs in neurology until the second half of the twen¬
tieth century. Mechanics was the metaphor, and contact between neurology
and psychology was limited largely to the fact that behavior is the expression
of nervous activity. Wiener, as we know, used the computer as a model.

THE COMPUTER ANALOGY FOR NEUROPHYSIOLOGY

Wiener was among the first to see the computer analogy, and he certainly
contributed to its popular diffusion.

Let me now come to another point which I believe to merit attention. It has
been clear to me that the modern ultra-rapid computing machine was in princi¬
ple an ideal central nervous system to an apparatus for automatic control; and
that its input and output need not be in the form of numbers or diagrams but
might very well be, respectively, the reading of artificial sense organs, such as
photoelectric cells or thermometers, and the performance of motors or sol¬
enoids. With the aid of strain gauges or similar agencies to read the perfor¬
mance of these motor organs and to report, to “feed back,” to the central
control system as an artificial kinesthetic sense, we are already in a position to
construct artificial machines of almost any degree of elaborateness of perfor¬
mance. (Wiener, 1948 and 1961, pp. 26-27.)

Wiener cites a number of mathematicians, engineers, neurophysiologists,

and physicians among his intellectual contributors and relates how he be¬
came acquainted with the work of Walter Pitts, who, together with Warren
McCulloch, had made the mathematical connection between logic and the
brain before him. [McCulloch and Pitts, 1943.] Wiener referred to their work
in the following statement:

At that time [1943] Mr. Pitts was already thoroughly acquainted with mathe¬
matical logic and neurophysiology, but had not had the chance to make very
many engineering contacts. In particular, he was not acquainted with Dr. Shan-
422 MURRAY EDEN

non’s work, and he had not had much experience of the possibilities of elec¬
tronics. He was very much interested when I showed him examples of modern
vacuum tubes and explained to him that these were ideal means for realizing in
the metal the equivalents of his neuronic circuits and systems. From that time,
it became clear to us that the ultra-rapid computing machine, depending as it
does on consecutive switching devices, must represent almost an ideal model
of the problems arising in the nervous system. The all-or-none character of the
discharge of the neurons is precisely analogous to the single choice made in
determining a digit on the binary scale, which more than one of us had already
contemplated as the most satisfactory basis of computing-machine design. The
synapse is nothing but a mechanism for determining whether a certain combi¬
nation of outputs from other selected elements will or will not act as an ade¬
quate stimulus for the discharge of the next element, and must have its precise
analogue in the computing machine. The problem of interpreting the nature and
varieties of memory in the animal has its parallel in the problem of constructing
artificial memories for the machine. (Wiener, 1948 and 1961, p. 14.)

PSYCHOLOGY

The effect of cybernetic ideas may be illustrated by looking at the history of

psychology in the twentieth century. Earlier, I described some of the meta¬
phors used throughout the intellectual history of brain and behavior, and I
cited a psychology text of 1922 that reveals the poverty of its analogic
armamentarium. Kurt Lewin’s closely reasoned program for a science of
(gestalt) psychology gropes for the right mathematical framework without
success. [Lewin, 1936.] No bridge to neurology is even hinted at. Lewin
does not include adaptation, memory, information, transformations, compu¬
tation, control, or any of their cognates among his definitions. He does talk
of structure, spaces, and states (the idea of state space is implicit), but he is
unable to achieve an integration that would support mathematical applica¬
tion.
In their introductory psychology text, Robert Woodworth and Donald
Marquis make no use of such concepts as message, information, feedback,
computation, or control. A discussion of neurology is included, but its rela¬
tion to psychology does not go beyond the description given by Dunlap at
least thirty-five years earlier. For these authors, the stimulus/response
paradigm is preeminent. “The preceding chapter made clear that the envi¬
ronment acts on the individual by stimulating his receptors, and that the
individual reacts to the environment by responses of various kinds. We
learned that behavior also depends on the individual—his structure, his
temporary state and his activities in progress.” (Woodworth and Marquis
1921 and 1949, p. 230.)
These fibers are conducting units like the insulated wires of a telephone
cable. They illustrate the all-or-none theory of nerve transmission as Dunlap
did: “[If] a charge of dynamite . . . explodes at all it explodes completely.”
CYBERNETICS 423

(Ibid., p. 238.) Woodworth and Marquis recognize the importance of organi¬

zation but offer no means of investigating this property of the brain: “But
these sensory data and motor responses must be organized if the individual
is to deal effectively with the objective environment. Most of the cortex is
concerned with this major job of organizing.” (Ibid., p. 259.) We note again
the analogic use of telephone cables and explosives. I find no reference to
the analogy of the computer. Furthermore, in the chapter on perception,
there is no bibliographic entry later than 1943. It is fair to conclude that the
authors did not believe that perceptual research during the twenty-four years
prior to this edition was of particular importance to beginning under¬
graduates.7
There is a notable instance of the converse situation, when a psychologist
writing much earlier came very close to current concepts but did not have an
apposite analogy at his disposal. In 1911, William MacDougall attacked the
mechanistic view of mind. [MacDougall, 1911, rev. ed. 1961.]

The history of thought upon the psycho-physical problem is in the main the
history of the way in which Animism, the oldest and, in all previous ages, the
most generally accepted answer to it, has been attacked and put more and more
upon the defensive in succeeding centuries, until towards the end of the
nineteenth century it was generally regarded in academic circles as finally
driven from the field. I have therefore given to the historical chapters the form
of a history of Animism. . . .

The modern currency and usage of the word derives chiefly from Prof. Tyler’s
“Primitive Culture,” and I use it with the general connotation given it in that
celebrated treatise. The essential notion, which forms the common foundation
of all varieties of Animism, is that all, or some, of those manifestations of life
and mind which distinguish the living man from the corpse and from inorganic
bodies are due to the operation within him of something which is of a nature
different from that of the body, an animating principle generally, but not neces¬
sarily or always, conceived as an immaterial and individual being or soul.
(Ibid., p. 14.)

The contemporary American psychologist Jerome Bruner has this com¬

ment on MacDougall’s contribution in his 1961 introduction to a new edition
of MacDougall’s work:

He [MacDougall] published a paper in 1901 in Brain, “On the seat of the

psychophysical processes,” in which it was plain that Mind acted not as a
capricious organ of intervention but in a highly systematic way to set and alter
the passage of neural impulses in the synapses of the central nervous system.

7My judgment has been made in light of the publishing history of this popular text. The copy
available to me is a version of the twentieth edition that appeared in a new size in 1963 and was
reprinted in 1967. The twentieth edition was copyrighted in 1949. However, the preface to the
edition is dated February 26, 1947.
424 MURRAY EDEN

His argument is anatomical (that it is in the synapses between neurones that

any shunting or selectivity of conduction must be determined), physiological
(that the evidence of sensory adaptation and contrast and other sensory phe¬
nomena required a selective interaction in the brain), and psychological (that
attention particularly required a theory of synaptic shunting). The paper is
exquisitely argued and documented, and while it contains some bad guesses
about eventual mechanisms of nerve transmission, the general outcome is still
interesting. For what he is saying, to put it in ultramodern dress, is that if Body
is the machine, Mind is the program that determines how the machine will
process its data and one cannot equate the machine and the program. The
series of three papers on such sensory phenomena as inhibition, facilitation,
and contrast in visual sensation published in Mind in the 1890s make it quite
clear that he is groping toward such a formulation. Mind, in short, uses the
nervous system. To describe the nervous system, then, does not constitute a
description or analysis of mind any more than the description of the use to
which a man puts a hoe represented a description of the nervous system or the
mind of the man that uses it. Mind, in effect, is the realm of purposes, plans,
intentionality. He wished to get Mind as Program into Body as Machine. The
misplaced concreteness of his solution is startling. The problem is real, and can
be stated better today. (Bruner in MacDougall, 1961, pp. xiv-xv.)

But consider the psychology text by Floyd Ruch revised for the ninth edition
in 1967. [Ruch, 1937 and 1967.]

The “knowledge explosion” in the area of physiological psychology in the last

fifteen years has produced much that is of direct relevance to a science of
behavior. So the discussion of this important area has been moved from the
Reference Manual into the text itself. (Ruch, 1967, p. ii.)

The new analogies are beginning to be used.

Although psychologists and neurologists still have much to learn about our
brains and nervous systems, they are sure that underlying every thought, every
perception and every action is a pattern of neural activity. As we encounter
stimulus situations of all kinds, messages are received, evaluated, integrated
and stored, and other messages are sent out to the various organs of response.
(Ibid., p. 41.)

Ruch cites Wiener for introducing the concept of servomechanism to biol¬

ogy. Moreover, the poverty of stimulus/response has been enriched to in¬
clude stimulus, organization (which he credits to gestalt psychology), re¬
sponse, feedback. Hence:

While driving, you hear an automobile horn honking—S. Past experience

causes you to seek an explanation—O. You glance in the rear view mirror—
R—but your response does not stop at that point. You see another car follow¬
ing closely, and you respond by pressing harder on the accelerator. Your
response of looking in the mirror supplied sensory feedback, which served as a
CYBERNETICS 425

new stimulus. Thus we see that our formula will not be really adequate unless
we add the element of sensory feedback. (Ibid.)

During the last decade, teaching of general psychology has been turned
upside down. Peter Lindsay and Donald A. Norman entitle their text Human
Information Processing: An Introduction to Psychology. They begin the
book with perception described in terms of pattern recognition, rules, and
feature analysis. They describe “neural information processing” in terms of
idealizations of the nervous system as logic elements and circuit diagrams.
Feedback control enters even in their discussion of motivation. The ter¬
minology of cybernetics as construed by Wiener is on virtually every page,
but neither Wiener nor cybernetics is mentioned. [Lindsay and Norman,
1972.]

BIOLOGY

It seems proper to conclude that the concepts of cybernetics have become

internalized in many aspects of psychology; much the same can be said of
neurology as well as other branches of biology. Biology remains largely an
experimental science and finds little place for the mathematics underlying
the cybernetic insights. Nonetheless, I cite a few examples from books I
have at hand to defend this position. I quote from Manfred Zimmermann
writing for a recent text in sensory physiology [Zimmermann in Schmidt,
1978]:

The nerve fiber can be compared, from the point of view of its function, with a
cable carrying information. The approach of the communications engineer—
information theory—can also be applied to the nervous system. This body of
knowledge, together with control theory, is often called cybernetics; especially
in Europe, its application in biology is termed biocybernetics, though such
general terms are used less in the United States. (Ibid., p. 68.)

Zimmermann then goes on to describe the rudiments of information theory

and relate it to notions of stimulus threshold and sensitivity.
More elaborate uses of control theory have been made by Lawrence A.
Stark in “The Control System for Versional Eye Movements.” [Stark in
Bach-y-Rita and Collins, 1971.] Control theory, including block diagrams
and circuit models, but little mathematics, is used by Ernest R. Hilgard in a
paper entitled “Pain Perception in Man.” [Hilgard, 1978.] Indeed, the col¬
lection in which Hilgard’s work is to be found begins with a series of five
papers grouped under the heading “1. Basic Features, A. Feature Filters and
Channels.” Here, we find all the paraphernalia of communication theory:
features, filters, channels, Fourier transformations (for space and time),
autocorrelation, cross-correlation, and so forth. Later on in this book, a
426 MURRAY EDEN

great deal of space is devoted to articles that require the notions of informa¬
tion and pattern recognition. [Held, Leibowitz, Teuber, 1978.] But since we
have already seen them in introductory texts, there is no use belaboring the
point.
Outside the domain of neurology or psychology, we find traces of the
ideas of cybernetics in many corners of biology. Carl F. Rothe contributes a
chapter to Selkurt’s Physiology, which he entitles “Regulations of Visceral
Function—B. Homeostasis and Negative Feedback Control.” [Rothe in Sel-
kurt, 1971.] He presents a block diagram for a negative feedback-control
system but includes no mathematics.8
Brian C. Goodwin’s book deals with a biological theory intended to pro¬
vide insight into the basic formal organization of the highly evolved and
elaborate forms of life. He casts much of his argument in terms of control
models and makes heavy use of the mathematics of control theory and
probability. [Goodwin, 1963.]
David Givol, an immunologist, begins a paper entitled “A Structural
Basis for Molecular Recognition: The Antibody Case” [in Cuatrecasas and
Greaves, 1976] with the sentence “Molecular recognition stands at the roots
of biology. . . . Recognition via combining sites is a property of enzymes,
antibodies, hormones, receptors, lectins and perhaps other classes of pro¬
teins.” (Givol, 1976, p. 3.)9
Earlier in this collection, Melvyn Greaves writes of “biological communi¬
cation”10 as well as “signals and the cell surface.” [Greaves in Cuatrecasas
and Greaves, 1976.]

8 For the nonbiologist, it may be worthwhile to explain the distinction between homeostasis and
feedback-control. The concept of homeostasis is due to Walter B. Cannon. [Cannon, 1929.]
Rothe defines it thus: “Homeostatic mechanisms operate to counteract changes in the internal
environment which are induced either by changes in the (external) environment or by activity of
the individual so as to maintain a constant and optimal internal environment.” Rothe goes on—
alas—to equate homeostasis with negative feedback-control. He is wrong. Feedback models in
the physiology of organisms are instances of homeostasis, but the great bulk of homeostatic and
adaptive models for the explanation of stability of composition and chemical activity in organ¬
isms rarely appeal to the notions of set-point, error function, and negative feedback. Of course,
flow diagrams are put down, and (generally nonlinear) partial differential equations are used
extensively, but the framework is that of physicochemical kinetics rather than control theory.
We may regard homeostasis as the biological equivalent of LeChatelier’s Principle augmented
by the hypothesis that concentrations of material substances and the energetics of biological
internal processes retain their constancy (in the common sense of that word) because the
variation of any one is tied to very many, if not all, of the others.
y Molecular recognition draws on thermodynamics in order to give meaning to the notion of
information. Molecular biologists measure affinity or association constants. The logarithm of
this quantity is proportional to the (Helmholtz) free energy of binding. Very rarely, the entropy
is evaluated, which would bring the concept into formal equivalence to the concepts of physical
information originated by Leo Szilard [1929] and John von Neumann [1932 and 1955],
1 ’The notion of specificity in the action of biologically active substances is at least seventy years
old. Paul Ehrlich formed the concept of a receptor theory from immunological studies on toxin-
CYBERNETICS 427

It is now widely appreciated that these processes [intercellular communication

and recognition] play a crucial role in virtually all biological systems and func¬
tions. These encompass fertilization, embryonic development, infectious in¬
teractions, the activity of the nervous system, the regulation of growth and
development by hormones and the immune response to foreign or “non-self’
antigens. ... In the last decade several major advances have led to a new level
of understanding of the molecular basis of cellular recognition .... These
studies have been paralleled by equally important insights into the general
structure and organization of cell membranes and the possible ways in which
signals arriving from the “outside” can be transduced across the cell surface
membrane to induce or regulate the cells’ programmed responses. . . . Many of
us suspect intuitively that not only are intercellular recognition phenomena
fundamental to all biological phenomena but that the codes and rules of the
game will be few, and therefore similar in widely divergent cellular systems.
(Cuatrecasas and Greaves, 1976, pp. vii-viii.)

CYBERNETICS IN APPLIED SCIENCE

So much for biology. The basic concepts of Wiener’s cybernetics have be¬
come embedded in virtually every branch of scholarship. The notion that
information can be handled as a mathematical quantity; that this quantity
mediates the transmission of energy and, therefore, can be used to control it;
that statistical uncertainty is itself a quantity related to (and degrading)
information; that information can be used to guide decision-making and
clarify our own descriptions of real-world phenomena; that information may
be transformed by computation; and that information and control can be
combined to create complex mechanisms that behave as if they were intelli¬
gent—these ideas, I maintain, are pervasive.
I feel it necessary to repeat that I am not commenting on the question of
Wiener’s priority in all his claims, nor is it relevant to my argument that the
domain of cybernetics has changed with the times and now tends to exclude
many of the ideas Wiener had included originally. Simply put, it seems to me
that the ideas Wiener set down in his book and which he considered to be
within cybernetics are now the common currency of science. It seems fruit¬
less to attempt to prove my assertion. Mathematics—pure or applied—is
there to be used whenever a scientist who has at least a nodding acquain¬
tance with a particular subset of mathematics decides it is relevant to his
problem. Moreover, because Wiener adjoined so much of mathematics to
communication and control theory, we would be hard put to identify some
area (not even differential topology or algebraic number theory) that could

antitoxin interactions. Even earlier Emile Fischer had attributed enzyme-substrate interactions
to a lock-and-key (schloss und schlusset) mechanism. [Ehrlich, 1956, vol. 2, p. 178.] 1 wonder
whether the introduction of the words recognition and communication helped. The fact remains
that they are now part of the biologic vocabulary. [Cited in Greaves, 1976, p. 15.]
428 MURRAY EDEN

be excluded definitively. Nonetheless, the tools Wiener proposed to use in

1948 are still the tools of choice in many applications, although they have
undoubtedly been refined during the last three decades.

COMPUTER SCIENCE AND ARTIFICIAL INTELLIGENCE

One area of technological development related to information and control,

machine computation, has dominated all others in terms of number of
participants and its influence on Western, especially American, society dur¬
ing the last twenty years. On the academic side, its practitioners prefer to
call their discipline computer science rather than machine computation.
Wiener included machine computation and emphasized its use as an analogy
to the brain within the purview of cybernetics. Ironically, the concepts of
cybernetics have played only a very minor role in the more recent growth of
machine computation. The great majority of people working with com¬
puters, attempting to improve or to study their behavior, are interested in
programming, computer architecture, computer languages, computational
methods (algorithms), communication networking, computer system design
and operations, and so forth.
It is only on the periphery of computer science that the special role
Wiener has played is explicitly acknowledged. Marvin Minsky, who, with
John McCarthy, coined the phrase artificial intelligence, discusses the place
of cybernetics in artificial intelligence in a recent paper. [Minsky, 1979.]
Whereas Minsky’s comment exhibits a strong bias in favor of more recent
work done under the banner of artificial intelligence rather than older work
done under another rubric, it is an accurate picture of one highly publicized
facet of current computer research indebted directly to the original concepts
of cybernetics.

The era of cybernetics was a premature anticipation of the richness of com¬

puter science. The cybernetic period seems to me to have been a search for
simple, powerful, general principles upon which to base a theory of intelli¬
gence. Among the ideas it explored were the following

Negative feedback The psychological concept of goal was identified with the
mechanism of setting up a generalized servomechanism to reduce the differ¬
ence between an input goal parameter and an observed system parameter. This
idea was exploited in various mathematical directions, but the secret of intelli¬
gence was not to be found in “optimal control” or similar knowledge-free
theories. Nonetheless, the difference-reduction concept, reformulated in terms
of a symbolic description of differences, finally became a key concept in
artificial intelligence in the General Problem Solver system of Newell and
Simon.

Pattern recognition The search for abstraction or invariance developed

slowly, I think, because of the error of identifying the thing with itself; that is,
CYBERNETICS 429

attempting to match image against image, rather than description against de¬
scription. The issues raised in this exploration—how a global or gestalt charac¬
teristic of a situation is discerned from an ensemble of local features—slowly
evolved into a very rich collection of theories of description and representa¬
tion, so rich, in fact, that the subject is still difficult to survey and criticize.
Stochastic learning The secret of creativity was sought in the area of con¬
trolled random search, both in models of learning and in models of problem
solving. The “stochastic learning models” of that era’s mathematical psychol¬
ogy did not lead anywhere, nor did experiments on “programmed evolution"
or “random neural networks.” A somewhat later approach based on “percep-
trons,” which were self-adjusting learning devices, seemed more promising but
subsequently died out. (Minsky, 1979, pp. 401-402.)

Nowadays, workers in artificial intelligence attempt to characterize tasks

in terms of meaning and semantics. Their insights are drawn from linguistics
and cognitive psychology rather than neurophysiology and psychophysics.
Thus, their work is largely outside the area of cybernetics as it is ordinarily
construed. They rarely use the sorts of applied mathematics Wiener thought
to be important in cybernetics, and certainly their concepts of goals and
directions are far from his notion of negative feedback or purposive control.

CYBERNETICS AS A SCIENTIFIC DISCIPLINE

Having concluded that the cybernetic hypotheses are part of the subcon¬
scious culture of science and technology, we are led also to the conclusion
that cybernetics does not constitute a well-defined scholarly discipline nor
does it connote a group of professionals with common technical problems
and interests. There remain a few relatively minor journals whose names
concatenate several terms; thus, Journal of Cybernetics and Information
Science; Biological Cybernetics, Biofeedback and Self-Regulation (formerly
published as Kybernetik); Kybernetes: An International Journal of Cyber¬
netics and General Systems; and Transactions of the IEEE Systems, Man
and Cybernetics Society. Occasionally symposia—also with concatenated
titles—are held and their papers published; thus, Progress in Cybernetics
and Systems Research. [Trappl and Hanika, 1975.] There is an International
Association of Cybernetics.11

"There are some differences between terminology in the United States and Europe. The editors
of the symposium just cited are president and vice president, respectively, of the Austrian
Society for Cybernetic Studies. Professor Robert Trappl, the president, is also professor of
biocybemetics and bioinformatics at the University of Vienna Medical School. The Technical
University of Berlin has a cybernetic department (Fachbereich Kybernetik). I also note re¬
search groups on sociocybemetics in the computer science laboratory of Uppsala University
and a Department of Cybernetics at Brunei University, England.
430 MURRAY EDEN

A search of U.S. college catalogs failed to turn up courses in cybernetics.

I have referred earlier to the absence of mathematically oriented texts. Two
institutions, San Jose State University and the University of California at
Los Angeles, grant degrees in cybernetic systems. By contrast, thirty-four
institutions of higher learning offer degrees in systems and information sci¬
ence, systems engineering, systems science, or systems technology. More
than 300 universities offer degrees in computers, computer science, or
equivalent titles; there are as yet no degrees available in artificial intelli¬
gence, pattern recognition, and the like.12 Regular conferences are scheduled
by the Institute of Electrical and Electronics Engineers (IEEE) Systems,
Man, and Cybernetics Society. At a recent meeting (August 24-27, 1981),
the list of topics included pattern recognition, artificial intelligence, man/
machine systems, biomedical systems and biocybernetics, system design
methodology, computer-aided design, and robotics. Each of these topics
was explicitly referred to by Wiener in his second edition of Cybernetics.13
In this sense, the interest of the field has remained stable. But virtually every
one of these subjects has developed in large measure independently of the
others, with its own journals, societies, and in some instances separate
academic departments.

CYBERNETICS AND SYSTEMS THEORY

The Russian mathematical scientist Valentin F. Turchin has cast the ideas of
cybernetics into a more formal mathematical way of looking at things than
did Wiener. [Turchin, 1977.] For Turchin, cybernetics is based above all on
the concept of the system, consisting of other objects called subsystems; so
cybernetics considers the relations between systems and subsystems. A
system or subsystem is characterized by its state, a concept that relies on
our intuition of time; thus, cybernetics studies the organization of systems in
space (real and abstract space) and time. Organization in time when it is
purposive is called control. Finally, cybernetics tests the concept of hierar¬
chy; the description at one hierarchical level will be different from that at
another level. Hierarchical structure or subsystems are intended to recog¬
nize or classify concepts derived from lower levels. Hierarchies also estab¬
lish the relations of goals and plans for the system by which recognition and
the consequent control decisions are made.
When cybernetics is cast in this framework, it is easy to understand the
growth in the use of the term systems analysis or systems theory as an

1'The College Blue Book, 18th Edition—Degrees Offered by College and Subject (New York:
Macmillan, 1981).
13 Wiener devotes a significant portion of the pages of Cybernetics to mathematical discussions.
We have the impression that the mathematics was introduced for illustration rather than for
teaching the mathematics or its application.
CYBERNETICS 431

equivalent to cybernetics. Many times the words are conjoined; for example,
the largest society dealing with cybernetic issues per se is the IEEE Sys¬
tems, Man and Cybernetics Society.
Systems theory is a domain of scholarship that exists in uneasy relation to
that of cybernetics. The originator of the term general system theory and one
of its major expositors, Ludwig von Bertalanffy, subsumes cybernetics
within general systems theory. [Bertalanffy, 1968.] As with Wiener, Ber¬
talanffy has no succinct definition for his newly coined phrase. It would
appear that all problems not solvable by classical science are problems of
general systems theory:

The methodological problem of systems theory, therefore, is to provide for

problems which, compared with the analytical-summative ones of classical
science, are of a more general nature.

As has been said, there are various approaches to deal with such problems. We
intentionally use the somewhat loose expression “approaches” because they
are logically inhomogeneous, represent different conceptual models, mathe¬
matical techniques, general points of view, etc.; they are, however, in accord
in being “systems theories.” (Ibid., p. 19.)

Later on, Bertalanffy particularizes slightly:

Not only are general aspects and viewpoints alike in different sciences; fre¬
quently we find formally identical or isomorphic laws in different fields. In
many cases, isomorphic laws hold for certain classes or subclasses of “sys¬
tems,” irrespective of the nature of the entities involved. There appear to exist
general system laws which apply to any system of a certain type, irrespective
of the particular properties of the system and of the elements involved.

These considerations lead to the postulate of a new scientific discipline which

we call general system theory. Its subject matter is formulation of principles
that are valid for “systems” in general, whatever the nature of their compo¬
nent elements and the relations or “forces” between them. (Ibid., p. 37.)

Despite the lack of a formal definition, Bertalanffy, again as with Wiener,

provides a definition by extension. His list of the systems theories com¬
prises: classical system theory, computerization and simulation, compart¬
ment theory, set theory, graph theory, net theory, cybernetics (restricted to
a theory of control systems—M. E.), information theory, theory of auto¬
mata, game theory, decision theory, and queuing theory. He mentions, but
does not discuss applied systems research, which includes systems engi¬
neering, operational research, linear and nonlinear programming. Later on,
he adds factor analysis, lumps network and graph theory into topology or
relational mathematics, and adds human engineering to applied systems re¬
search. (Ibid., p. 90.) Bertalanffy puts the origins of general systems theory
earlier and in places different from those of cybernetics. His first publication
432 MURRAY EDEN

with general system theory in the title was in proof when it was destroyed in
World War II.14
The systems theorist George Klir holds a somewhat more restricted view
of the general systems theory but includes cybernetics within systems
theory. We are indebted to Klir for explicitly distinguishing between cyber¬
netics as a science and cybernetics as a viewpoint. [Klir, 1970.] I could find
in his work no suggestion that an analogous division of systems theory
would be appropriate.
A currently active general systems theorist, Mihajlo Mesarovic, offers
another definition, one that stresses the mathematical nature of the theory:

[The] conceptual basis for systems theory might be summarized in the follow¬
ing observation: Study of any real-life phenomena reveals two aspects—in¬
formal, dealing with the meaning, interpretations, significance, objectives,
values and the like—and formal, dealing with the form (structure) in which the
relationship between the attributes appears. Systems theory is concerned with
the second, formal, aspect of the observations. Apparently, these formal rela¬
tionships are invariant with respect to the specific nature of the phenomena
under consideration.
Systems theory, then, is based on the following fundamental premises:
(a) A theory of any real life phenomena (biological or otherwise) is always
based on an image, termed a model.
(b) Without introducing any constraints whatsoever the formal, invariant,
aspects of the model can be represented as a mathematical relation. This
relation will be termed a system.

A system (i.e., a mathematical model of a real-life phenomenon) is usually

specified by means of a set of equations, e.g., difference equations, differential
equations, etc. To comprehend the essence of the foundations of systems
theory, it is important to appreciate, however, that a set of equations only
specifies the given system and that the distinction should be made between a
system and the set of equations that are used for its (constructive) specifi¬
cation. (Mesarovic, 1968b, pp. 60-61.)

There is an obvious contradiction between Mesarovic, who restricts sys¬

tems theory to the formal relationships among the observable attributes of a
real-life phenomenon, and Bertalanffy, who embraces logically different ap¬
proaches, mathematical techniques, and general points of view as being
systems theories. We face a procedural dilemma here that is equivalent to
the one that exists for us in deciding how to describe cybernetics. Is systems
theory what its generally acknowledged inventor says it is, or do we inter-

l4The citation for this work appears to be missing from General System Theory, although there
may be an inconsistency between the text (p. 14) and the bibliography. The text suggests the
first paper was to appear in Deutsche Zeitschrift fur Philosophic, while the bibliography, or¬
dered chronologically, cites “Zu einer allgemeinen Systemlehre,” Blatter fur Deutsche
Philosophie 3/4 (1945), extract in Biologia Generalis 19 (1949), 114-129. Are these one and the
same?
CYBERNETICS 433

pret it in terms of the (different) ways in which later workers describe its
domain? Other domains of scholarship do not ordinarily pose such prob¬
lems.

CYBERNETICS/SYSTEMS THEORY VIS-A-VIS

CONTROL/COMMUNICATION THEORY

There is a direct line of descent in the history of control theory from Harry
Nyquist through Wiener, followed by such control theorists as Richard Bell¬
man, Rudolph Kalman, and onward to the present day. It is not implausible
to suggest that James Clerk Maxwell would agree with his intellectual de¬
scendants as to the scope of a theory of governors that began with his work
of 1868. [Maxwell, 1868.]15
Very much the same argument can be made with regard to the history of
information or communication theory. Claude Shannon in The Mathemat¬
ical Theory of Communication asked and largely answered a question that
was undoubtedly close to Wiener’s problems with noise in stationary time-
series analysis: “How is an information source to be described mathemat¬
ically and how much information in bits per second is produced in a given
source?” (Shannon and Weaver, 1949, p. 10.) Shannon thought of his work
as an extension of the general theory of communication contained in the
work of Nyquist [1924] and Ralph Hartley [1928] twenty years earlier. In a
comment appended to Shannon’s seminal paper, Warren Weaver acknowl¬
edges the contribution of several precursors.

Dr. Shannon’s work roots back, as von Neumann has pointed out, to
Boltzmann’s observation, in some of his work on statistical physics (1894), that
entropy is related to “missing information,” inasmuch as it is related to the
number of alternatives which remain possible to a physical system after all the
macroscopically observable information concerning it has been recorded. L.
Szilard (Ztsch. f. Phys., Vol. 53, 1925) extended this idea to a general discus¬
sion of information in physics, and von Neumann {Math. Foundation of Quan¬
tum Mechanics, Berlin, 1932, chap. V) treated information in quantum me¬
chanics and particle physics. Dr. Shannon’s work connects more directly with
certain ideas developed some twenty years ago by H. Nyquist and R. V. L.
Hartley, both of the Bell Laboratories; and Dr. Shannon has himself empha¬
sized that communication theory owes a great debt to Professor Norbert
Wiener for much of its basic philosophy. Professor Wiener, on the other hand,
points out that Shannon’s early work on switching and mathematical logic
antedated his own interest in this field; and generously adds that Shannon
certainly deserves credit for independent development of such fundamental
aspects of the theory as the introduction of entropic ideas. Shannon has natu-

15 Maxwell writes: “I propose at present, without entering into any details of mechanism, to
direct the attention of engineers and mathematicians to the dynamical theory of governors.”
(Maxwell, 1868, p. 106.)
434 MURRAY EDEN

rally been specially concerned to push the applications to engineering com¬

munication, while Wiener has been more concerned with biological application
(central nervous system phenomena, etc.). (Shannon and Weaver, 1949, p. 95.)

For completeness, we note that Shannon also acknowledged the early use by
Ronald A. Fisher of an information measure different from his: the quantity
of information in a sample drawn from a density distribution with finite
expectation. Colin Cherry carried the history back much further, but it need
not concern us here. [Cherry, 1951.]
The appearance of Shannon’s work had an explosive effect, not only in
communications theory, but in sparking a multitude of attempts to apply his
theory to all sorts of processes that appeared to be related to the transmis¬
sion of objects with semantic content. How much of this efflorescence was
due to Shannon and how much to Wiener is impossible to assess; the appear¬
ances of Cybernetics and “The Mathematical Theory of Communication”
(as a paper) were contemporaneous [1948]. However, there are several im¬
portant differences. Shannon presented a developed mathematical theory;
he offered a transparently simple set of measures of information. He avoided
entirely any allusion to the many kinds of transactions that people are likely
to call informative. Rather, he asked how to measure information, and he set
the ground rules for models that could conceivably use his measures.16 Then,
Shannon, in a simple example, showed how to estimate the entropy of a
natural information source, the English language.
Information theory has developed much as any other branch of applied
mathematics. Many applications have been developed for communication
problems. Some of them led to newer, richer theory, and these were applied
in their turn. And so on to the present.
This is not the place to elaborate on general systems theory, but we must
distinguish between systems theory as a viewpoint and mathematical sys¬
tems theory. The latter is quite clearly to be included within mathematics
and continues to be an active area of mathematical development. It is in the
direct line of descent of govenors, servomechanisms, feedback, and optimal
control previously alluded to. Rudolf Kalman, Peter Falb, and Michael Ar-
bib in the introduction to their text Topics in Mathematical System Theory
lay emphasis on “the concepts of states, control, optimization and realiza¬
tion.” [Kalman, Falb, and Arbib, 1969.] Suffice it to say that systems theo¬
rists and cyberneticians occupy the same turf. The choice of a name for a
new domain of scholarship is a sociological rather than a scientific question.
Scholars of the West tend to favor systems theory, the Eastern bloc appears
to prefer cybernetics; there are some who will use both, and others who use
neither, favoring instead more restricted labels to classify their work.

l6This is by no means a trivial exercise. The probability distribution on an infinite set of possible
messages must be properly defined or else the apparatus collapses. Wiener provided no handy
mathematics; although at times he urges caution, he did not provide the apparatus for imposing
restraint.
CYBERNETICS 435

It is also worth noting that much as with cybernetics, systems theory is

regarded by some workers—mathematicians in particular—as a field with
many naive, ill-equipped, self-serving, and even fraudulent practitioners.
The philosopher and student of mathematics, David Berlinski, furnishes a
detailed critique of systems analysis from this point of view in an extrava¬
gantly polemical but still worthwhile book. [Berlinski, 1976.] He holds no
higher opinion of cybernetics, though his blows at the latter are en passant.
He describes his motivation thus: “It is the use of mathematical methods for
largely ceremonial reasons that I deplore and denounce as pernicious.”
(Ibid., p. ix.)

CYBERNETICS/SYSTEMS THEORY VIS-A-VIS BIOLOGY AND OTHER

THAN NATURAL SCIENCES

Biology, as was indicated earlier, has adopted the insights of cybernetics and
many of the terms used in control theory, information theory, and com¬
puters. While I have attributed it to Wiener’s influence, it may well have
been due to the rise of systems theory. That is to say, biologists have
presented their ideas in fora of both kinds. The Yale physiologist, Talbot
Waterman, has put forward his view of the importance of systems analysis at
a symposium on Systems Theory and Biology.

Because of this multivariable and highly interconnected organization, living

things require for their effective study some overall strategy like systems analy¬
sis. Such indeed is the fundamental thesis of this book. . . . my intention here
is to indicate from an experimental biologist’s point of view the relevance of
the systems approach to understanding a living organism ....
Organisms are self-regulating, adaptive systems capable of autoduplication. As
such they must acquire energy from the environment and utilize it to do their
biological work. To this end living things act thermodynamically as irreversible
chemical machines. But to be self-regulating and adaptive, control and infor¬
mation are essential. Cybernetic mechanisms therefore must provide at once
the basis for the organism’s steady state and dynamic characteristics as well as
the adaptedness required for its survival.
As a consequence we have two major components to analyze for understand¬
ing biological systems. One of these comprises all the elements relating to the
acquisition, transfer and utilization of energy. These permit the organism to
grow, to move and in general to do work. The other consists of everything
functioning in the detection, processing, retention, and utilization of informa¬
tion. These permit the control of what the organism does and how fast. Essen¬
tial for the maintenance of life, both components are closely interrelated in
jointly sustaining the steady state. (T. H. Waterman, 1968, p. 13.)

A few paragraphs later, Waterman turns his attention to the cybernetic

component. It is not clear to me that he distinguishes between cybernetics
and systems analysis except perhaps by intension:
436 MURRAY EDEN

... it is important to realize that the complete biological system comprises

genetic (including, probably, developmental), synecological and evolutionary
regulators as well. While control in all four (or five) of these realms shows
certain fundamental similarities, physiological regulation has both the shortest
time scale and usually its own characteristic machinery. Such control includes
both the maintenance of the physiological steady state (homeostasis in Walter
Cannon’s sense of the word) . . . and the production of dynamic servocontrol
exemplified by pursuit of prey or by adaptive synchronization of feeding and
hibernation with the cycle of the seasons.

But homeostasis itself is truly dynamic in nature despite its proper distinction
from the reference-input-following servo-type control. This is so because the
physiological steady state is neither just a passive resistance to change nor a
mere compliance to pattern imposed mainly from outside. Rather it results
primarily from compensatory adjustments actively programmed within the or¬
ganism in response to the total relevant information it has available. Thus in
spite of an exquisite sensitivity to changes of many kinds, organisms maintain
their stable state with remarkable thoroughness and precision. That they sur¬
vive attests also to the reliability of their overall regulation despite failures or
irregularities in detail. . . .

Systems Analysis. To resolve such an apparent paradox, systems analysis

provides a powerful methodology. Not only in biology but in its engineering
and other applications, systems analysis may be defined quite generally as the
application of organized analytical and modeling techniques appropriate to
explaining complex multivariable systems many of whose functional compo¬
nents may be initially quite imperfectly measured or even largely unidentified.
Depending on the available data and the purpose of the analysis any, or more
usually several, of a wide range of specific techniques may be employed rang¬
ing from the use of information theory and cybernetics to computer simulation
or multivariable statistical analysis.

While this may seem a rather loosely defined recipe for advancing science, it
does in fact provide a reasonably effective and increasingly appreciated array
of interrelated disciplines appropriate in particular ways to the formidable
problems in hand. (T. H. Waterman, 1968, pp. 3-4.)

One final fact completes this attempt to characterize the relation between
cybernetics and systems theory; that is, the interest they have engendered in
fields of study outside the natural sciences. Because these fields are not in
my area of competence, I hesitate to venture a judgment as to the influence
either cybernetics or systems theory has had. Nonetheless, there is little
question that a number of distinguished social scientists, economists, and
students of business administration have looked to these two fields for a
source of ideas and tools relevant to their particular professional needs. For
example, volume two of Progress in Cybernetics and Systems Research is
comprised of papers in Socio-Economic Systems, Cognition and Learning,
Systems Education, Organization and Management. [Trappl and Hanika,
1975.] A Survey of Cybernetics, published in 1969 and dedicated to the
CYBERNETICS 437

memory of Norbert Wiener, devotes the bulk of its pages to papers on

Cybernetics and Artifacts (Information, Learning, and Teaching Systems;
Models of Development; Artificial Intelligence Control), Cybernetics and
Industry, Cybernetics and Society. [Rose, ed., 1969.] Stafford Beer provides
the keynote introduction to the first collection just cited and the epilogue to
the second. Beer, a pioneer in applying cybernetics to management and a
prolific writer in the field, assigns an importance to applied cybernetics that
transcends its role in explaining problems of communication and decision¬
making.

The reason for my defection [from the First Symposium on Cybernetics and
Systems Research] was in fact an urgent recall to Santiago from the late Presi¬
dent Salvador Allende of Chile. We had embarked six months earlier upon a
program so ambitious as to have had at least a chance of revolutionizing the
form of government on a cybernetic basis that would match the revolutionary
political intentions of that democracy. This endeavor took precedence with me
for two years, and I emerged from the experience very much changed. I
changed in my awareness of myself, of my fellow men and of political realities;
but these are not the topics that I shall discuss today. I changed also as a
technologist, in terms of confidence. For I now know that it is possible to do
what I have advocated for so many years—things which many used to say, and
some still do say, are impossible.
But the changes that bear upon the nature of this symposium have to do with
cybernetic insights themselves. There is of course no way of changing the laws
by which large systems operate; but there can be a change in one’s perception,
and a change in the depth of understanding of principles we have known about
all the time. It is of these matters that I speak today, because I know more
clearly now what I am trying to say, and because I also know more about the
direct practical relevance of these things to society at large. (Beer, 1975, p. 3.)

Wiener, writing twenty-eight years earlier, discussed the responsibilities

of cyberneticians in the sphere of social activism. He shares much of Beer’s
view but paints his landscape in more sombre colors.

Those of us who have contributed to the new science of cybernetics thus stand
in a moral position which is, to say the least, not very comfortable. We have
contributed to the initiation of a new science which, as I have said, embraces
technical developments with great possibilities for good and for evil. We can
only hand it over into the world that exists about us, and this is the world of
Belsen and Hiroshima. We do not even have the choice of suppressing these
new technical developments. They belong to the age, and the most any of us
can do by suppression is to put the development of the subject into the hands of
the most irresponsible and most venal of our engineers. The best we can do is
to see that a large public understands the trend and the bearing of the present
work, and to confine our personal efforts to those fields, such as physiology
and psychology, most remote from war and exploitation. As we have seen,
there are those who hope that the good of a better understanding of man and
438 MURRAY EDEN

society which is offered by this new field of work may anticipate and outweigh
the incidental contribution we are making to the concentration of power (which
is always concentrated, by its very conditions of existence, in the hands of the
most unscrupulous). I write in 1947, and I am compelled to say that it is a very
slight hope. (Wiener, 1948 and 1961, pp. 28-29.)

CONCLUDING REMARKS

To be sure, the analogies among such mechanical functions as information

processing, computation, control, estimation, and the behavior of function¬
ing organisms have been pushed quite far. Where experiments have per¬
mitted mathematical tests of mechanistic or probabilistic models for describ¬
ing physiological or sensory-motor behavior, they have been well accepted,
so much so that the terminology of computers and statistical signal process¬
ing has become part of the standard vocabulary of psychology and physiol¬
ogy. It is a tribute to Wiener that there are still a significant number of
scientists who wish to preserve the word he used to define an intersection of
many fields.
At the present time, it appears that cybernetics as a science in the ter¬
minology of Klir never really existed under that name, though many of the
mathematical specialties co-opted by Wiener into cybernetics are flourish¬
ing. Cybernetics as a viewpoint has entered into the unconscious of workers
in almost every branch of applied science and the sciences of man. It plays a
role, though a minor role, in the reductionist domains of science and, hence,
in virtually all of physics and chemistry, and much of biology.
Perhaps an explanation for the meteoric trajectory of cybernetics was
prefigured in the very origin of Wiener’s reason for inventing it.

If the difficulty of a physiological problem is mathematical in essence, ten

physiologists ignorant of mathematics will get precisely as far as one physiolo¬
gist ignorant of mathematics, and no further. If a physiologist who knows no
mathematics works together with a mathematician who knows no physiology,
the one will be unable to state his problem in terms that the other can manipu¬
late, and the second will be unable to put the answers in any form that the first
can understand. (Wiener, 1948 and 1961, pp. 2-3.)

He softened this statement by adding:

Dr. Rosenblueth has always insisted that a proper exploration of these blank
spaces on the map of science could only be made by a team of scientists, each a
specialist in his own field but each possessing a thoroughly sound and trained
acquaintance with the fields of his neighbors; all in the habit of working to¬
gether, of knowing one another’s intellectual customs, and of recognizing the
significance of a colleague’s new suggestion before it has taken on a full formal
expression. The mathematician need not have the skill to conduct a physiologi-
CYBERNETICS 439

cal experiment, but he must have the skill to understand one, to criticize one,
and to suggest one. The physiologist need not be able to prove a certain
mathematical theorem, but he must be able to grasp its physiological
significance and to tell the mathematician for what he should look. (Ibid., p. 3.)

Yet, the history of modern mathematics, applied science, and technology

suggests that the first statement is still a more nearly true state of affairs; the
second describes, at best, a transitional state. Mathematicians, scientists,
engineers, and scholars continue to segregate into narrower and narrower
specialties. The intersection of their interests, where it occurs, may well
produce another specialty narrower still. When a physiologist is ignorant of
the mathematics, his or her best course is to learn it. Someone may first be
required to prepare the way by particularizing the mathematics for the physi¬
ological problem. This is a common precondition for much of applied mathe¬
matics. For example, the first courses taught to beginning electrical en¬
gineers are derived from the particularization of solutions of certain linear
differential equations, the elements of graph theory, and a specialization of
the propositional calculus. When the concepts for the physiological specialty
are embedded in a mathematical framework shared by the community of
specialists, there is no longer a need to acknowledge the specialist’s indebt¬
edness to the originator of the concept. After all, we would be quite sur¬
prised and not a little suspicious were an astrophysicist, say, to cite Newton
or Leibniz for his or her use of the calculus.
Although many scholars have faith in the utility of the exchange of ideas
among practitioners from different fields, the intersection of different disci¬
plines remains an intersection and not a field of study itself, although with
time it may be transformed into yet another specialty. The notions of cyber¬
netics have permeated many disciplines—computer science, information
theory, control theory, pattern recognition, neurophysiology, psychophys¬
ics, perceptual psychology, robotics, and the like. Having been integrated
into them, cybernetics has performed the function for which it was pro¬
posed.
 CYBERNETICS
Past and Present, East and West

Peter Elias

Eden gives an excellent commentary on the field of cybernetics, as intro¬

duced by Norbert Wiener in his book Cybernetics. [Wiener, 1948 and 1961.]
1 would like to add some terminological, historical, and geographical com¬
ments.
Cybernetics, or Control and Communication in the Animal and the Ma¬
chine was published first by Hermann et Cie. in Paris in 1948. The internal
evidence suggests that it was set in type and proofread (if at all) by people
knowing neither English nor mathematics. The American edition, copub¬
lished by the Technology Press, John Wiley, and Hermann et Cie., was
photo-offset from the original plates without correction: The numerous ty¬
pographical errors were dealt with for the first time only in the second
edition, published by the MIT Press in 1961.
In the preface and introduction to the second edition, Wiener gives some
of the history of the field of cybernetics and mentions a series of conferences
starting in 1944 and later sponsored by the Macy Foundation. The founda¬
tion sponsored ten such conferences in all. Five of these occurred after the
publication of Cybernetics, and their proceedings were published by the
foundation. [Josiah Macy Jr. Foundation, 1949-1953.]
1948 was also the year in which Shannon published his paper '‘A Mathe¬
matical Theory of Communication.” [Shannon, 1948.] Wiener mentions
Shannon’s definition of information in his book. Shannon mentions Wiener’s
work on the prediction and filtering of noisy signals, which had appeared
during World War II as a report from the Office of Scientific Research and
Development entitled The Extrapolation, Interpolation, and Smoothing of
Stationary Time Series with Engineering Applications, and which later ap¬
peared in book form under the same title [Wiener, 1949], but Shannon does
not mention Cybernetics in his paper. In the United States, Shannon s
theory and its later developments were not absorbed in cybernetics but
acquired their own label, information theory. This label is still most com¬
monly used in the same narrow way. However, the label is also the present
title of a group within the Institute of Electrical and Electronics Engineers
441
 442
PETER ELIAS

(IEEE) whose Transactions are broader: They are devoted to theoretical

work relevant to communications engineering, which includes, for example,
not only extensions of Shannon s theory but also the filtering and detection
of signals in noise and, in addition, algebraic coding theory—a field of dis¬
crete mathematics that is closely related to Shannon’s work in both its
origins and its applications, but has attained mathematical independence.
The same complex of ideas that captured the attention of Wiener and
others in the United States was of great interest in England immediately
after World War II. Cybernetics had a British audience, especially in
neurophysiology. In fact, the earliest textbook of cybernetics that I have
found is not the Glushkov volume of 1966 cited by Eden but a British volume
published ten years earlier, complete with exercises for students [Ashby,
1956] by an author who had earlier written a book dealing enthusiastically
with a cybernetic, homeostatic model of the human brain [Ashby, 1952],
Four symposia on information theory were held in London in 1950, 1952,
1955, and 1960; the first two organized by Willis Jackson and the last two by
Colin Cherry (both professors of electrical engineering at Imperial College);
proceedings were published. [Jackson, 1950, 1953; Cherry, 1956, 1961.]
There were also two symposia in the United States, in 1954 and 1956, orga¬
nized by the Information Theory Group of the IEEE and held at the Massa¬
chusetts Institute of Technology. [IEEE Transactions, 1954; 1956.] These
symposia, and especially the ones in London, used a definition of informa¬
tion theory much broader than the narrow American usage I previously
cited, which is still current in the United States. The papers presented dealt
with a range of topics and fields of application as broad as, and quite similar
to, those discussed by Wiener in Cybernetics. There was perhaps more
emphasis on communication among people, and more interest by psycholo¬
gists and linguists, in the London meetings than in Wiener’s book, and less
focus on the neuromuscular feedback loops in which Wiener and some of his
early associates were particularly interested. However, a similar shift to¬
ward human studies and more complex problem domains is visible, for
example, in the transactions of the tenth Macy meeting in 1955. The content
of the London meetings together with Colin Cherry’s book On Human Com¬
munication [1957 and 1966] may be taken to define what information theory
meant in England during the period 1950-1960, and to some extent even
later, while cybernetics was the correspondingly broad term widely accepted
in the United States in the early 1950s.
In the United States, there is no present designation that has widespread
professional acceptance and denotes approximately the particular mix of
topics and techniques denoted by cybernetics here and by information
theory in England in the early 1950s. The psychologists, linguists, neurophy¬
siologists, and sociologists interested in communication within and between
organisms, the biologists interested in genetic information, and others who
might have called themselves cyberneticists in the early 1950s had mostly
returned by the 1960s to their original disciplines, taking with them and
 443
CYBERNETICS: PAST AND PRESENT, EAST AND WEST

integrating into those disciplines such concepts and techniques as had

proved fruitful for their fields. In the Soviet Union, however, cybernetics
became an important label just as its use was fading in the West.
There seems to have been no native development analogous to cybernet¬
ics in the Soviet Union before or during World War II. The necessary mathe¬
matical background, in mechanics for control theory and in probability and
random-process theory for communications and stochastic control, was
available because of the historic strength of Russian mathematics and the
interest in applications of many of the best mathematicians, pre- and post¬
revolutionary. Kolmogorov, in fact, had done work related to Wiener s
wartime work on filtering and prediction of random processes, and others
also worked on random process problems. Kotelnikov, a talented communi¬
cations engineer, submitted a doctoral thesis in 1947, which independently
developed some of the new ideas about modulation that also appeared in the
West after the war. [Kotelnikov, 1959.] But the high-quality work was al¬
most all theoretical and mathematical in character and had little to do with
the biological or behavioral sciences. The engineering development of
sophisticated control, computer, and communications systems was much
slower in the Soviet Union, even after the war. Most Russian engineers were
educated in narrow engineering specialties and evidently did not experience
the kind of interaction with behavioral scientists that occurred in the United
States and Britain during wartime research and development. There were
well-entrenched orthodoxies in the fields of linguistics, “higher nervous
activity,” psychology, and economics that were not receptive to new ideas,
and researchers in these fields did not have access to the new instrumenta¬
tion and computer resources that helped generate enthusiasm for cybernetic
ideas in the West.
The first introduction of ideas from the neighborhood of cybernetics to the
Soviet scientific community seems to have occurred from 1953 to 1956.
Shannon’s information theory was presented to the Soviet mathematical
community by Aleksandr Khinchin, a distinguished mathematician, in two
articles [1953; 1956] available in English. [Khinchin, 1957.] Khinchin empha¬
sized the mathematical significance of the theory and not its potential for
application to the social sciences and linguistics and gave it high praise.
Khinchin was not alone: Kolmogorov, one of the world’s great mathemat¬
icians, also showed early interest and has made a large number of fundamen¬
tal contributions to questions arising from information theory since. The
following passage from the 1956 article by Khinchin shows the enthusiasm of
Soviet mathematicians for Shannon’s work:

Information theory is one of the youngest branches of applied probability

theory; it is not yet ten years old. The date of its birth can, with certainty, be
considered to be the appearance in 1947-1948 of the by now classical work of
Claude Shannon. Rarely does it happen in mathematics that a new discipline
achieves the character of a mature and developed scientific theory m the first
 444
PETER ELIAS

investigation devoted to it. Such in its time was the case with the theory of
integral equations, after the fundamental work of Fredholm; so it was with
information theory after the work of Shannon. [Khinchin, 1957, p. 30.]

At the same time, however, the 1956 philosophical encyclopedia of the

Soviet Union defined cybernetics as a “reactionary pseudo-science, that
arose in the United States after World War II and that has attained wide
dissemination in other capitalistic countries . It was only gradually after
Stalin s death that elements of Soviet society succeeded in taking steps to
fiee more of science from ideological orthodoxies and constraints. Cyber¬
netics became in a short time a highly positive concept in the Soviet Union
and received a great impetus when Khrushchev at the 22nd Party Congress
in 1961 said: “It is imperative to organize wider application of cybernetics,
electronic computing, and control installations in production, research
work, drafting and designing, planning, accounting, statistics and manage¬
ment.”
The picture that Khrushchev and others presented was that cybernetics
was the science of control and a socialist society could exploit such a science
in organizing and planning the entire economic and social systems in ways
ideologically impossible in capitalist countries. The result was the appear¬
ance of a flood of papers in a number of fields in the natural sciences, the
social sciences, and engineering, which embodied new and often Western
scientific approaches and did so under the banner of cybernetics. For a
number of years, papers that were abandoning old orthoxodies cited
Khrushchev and labeled themselves as cybernetic.
In the West, mathematical models for complex problem domains had
been introduced in a number of fields under a variety of different names.
Mathematical economics, operations research, input/output analysis, Monte
Carlo techniques, information theory, game theory, system engineering, fac¬
tor analysis, and other statistical techniques in the behavioral sciences all
had origins prior to, and independent of, cybernetics in the United States.
Mathematization and computer modeling in both engineering and the biolog¬
ical and behavioral sciences developed under all of those names. Even at its
peak, cybernetics as a label was used in the United States by only a small
subset of all of the users of computers and mathematical models. In the
Soviet Union, however, such models had been much less developed and
were not in significant use in the behavioral and social sciences. When the
floodgates opened, all of these mathematical ideas useful in the analysis and
computer modeling of complex systems—and indeed the very notion of
doing such modeling—became identified with cybernetics.
 CYBERNETICS AND
SYSTEMS THEORY
A Search for Identity

Richard Mattessich

Murray Eden brings a wealth of historical material to bear on the task of

clarifying relations between cybernetics and system theory as well as other
connected areas. To disentangle those relations is an arduous task, since
cybernetics, general system theory, operations research, systems analysis,
computer science, communications theory, information science, artificial
intelligence, automata theory, and so forth, all overlap with each other to a
greater or lesser extent. This network of interrelations might be less complex
were it not for the human urge to devise impressive labels and classification
schemes and to cling tenaciously to one’s own home-made categories. But
this itself could be the consequence of our inability to come to grips with the
universe as an indivisible entity. And, knowing no better, scholars pro¬
ceeded, each in his or her own fashion, to chop reality into fairly arbitrary
conceptual bits and pieces. However, the history of science assures us that
in time, every paradigmatic storm settles down and a conceptual scheme
emerges that appears fairly satisfactory even if not optimal. As some thirty
years or more have passed since many of these areas have come into being,
we should expect that most of the consolidation has taken place. This,
indeed, seems to be the case, but perhaps not so definitely as to preclude
some supplementation or moderate divergence of opinions.
As can be concluded from Eden’s presentation, the term cybernetics has a
narrow and a broad meaning. In the narrow and original sense, it represents
“the science of control and communication in the animal and the machine.
[Wiener, 1948 and 1961.] It is in this sense that it is usually understood in the
Western Hemisphere, while in East European countries, as Eden points out,
it is used in a much broader sense, which comes close to, or is identical with,
our understanding of system theory. But the relation between cybernetics

Additional support from the Social Sciences and Humanities Research Council of Canada is
gratefully acknowledged.
445
 446
RICHARD MATTESSICH

(in the narrow sense) and systems theory becomes clear only after grasping
the reason why the former is only a part of the latter. This can best be
demonstrated by discussing the major ideas and components that modern
system theory consists of. And we may thus distinguish the following basic
ideas and notions of the systems approach:

1. Organization and emergent properties.

2. Structure, hierarchy, and evolution.
3. Function and goal orientation.
4. Information, control, and feedback.
5. Environment and its influence.
6. System laws and mathematical homologies.

ORGANIZATION AND EMERGENT PROPERTIES

A major concern of Aleksandr Bogdanov [1913 and 1922; 1980] and Ludwig
von Bertalanffy [1928 and 1933; 1950; 1968], the founders of system theory,
was the reductionistic and atomistic attitude of the physical sciences and the
related neopositivistic philosophy. Due to this attitude and to the fact that
the biological and social sciences were about to borrow a good deal of their
methodological inventory from the physical sciences, the phenomenon of
emergent or holistic properties, especially those of living or social systems,
was unduly neglected. But not even in physics and chemistry could those
newly emergent properties be explained in terms of mere aggregation.1 This
initiated the quest for principles of organization within the framework of
which the unique status of emergent properties could be given due recogni¬
tion and which, in time, might even acquire explanatory powers. It is for this
reason that Bogdanov’s book [1913] was called Tektologia and bears the
subtitle The Universal Science of Organization and why Bertalanffy [1968,
p. 187] speaks of an organismic revolution, the core of which is supposed to
be the notion of system. Closely connected with this is James G. Miller’s
[1978] hint that Alfred N. Whitehead, with his “philosophy of organism,’’
might be the founder or precursor of system philosophy, and Miller’s asser¬
tion that the “key concepts later accepted as basic to systems science occur
in his [Whitehead’s] writings.” (J. G. Miller, 1978, p. xiii.)2

For example, such holistic properties as the relatively harmless and useful properties of water
or cooking salt that emerged from the combination of such explosive substances as hydrogen
and oxygen or of no less dangerous substances as sodium and chloride, respectively
Personally, I am inclined to side with Wiener [1950, 1954, and 1967, pp. 27-28] in regarding
Leibniz as the true precursor of the system approach. His monadology comes fairly close to a
kind of system theory, although we believe that a monade is not to be identified with a system,
but rather with the reflective (i.e., mental or quasimental) aspect of a system.
CYBERNETICS AND SYSTEMS THEORY

Of particular interest to us is the fact that Norbert Wiener either in his

books on cybernetics [1948 and 1961; 1950, 1954, and 1967] or in his later
work [1949, 1964]—is not concerned with the problem of emergent proper¬
ties and those organizational aspects that make this concern inevitable.
Therefore, it seems that his cybernetics lacks the broad quest for principles
of organization beyond those referring to control, homeostasis, and feed¬
back information. It should be admitted, however, that Wiener’s book
Cybernetics, especially its last chapter, is concerned with several aspects of
organization in the general sense.

SYSTEM STRUCTURE, HIERARCHY, AND EVOLUTION

Much of the literature of modern system theory is concerned with a concept

very closely related to that of organization, the notion of structure or form as
the relation between elements. Some system theorists go even so far as to
regard a system as “an abstract relation on a set of objects’’ [see Mortaza-
vian in this volume], thereby identifying and possibly confusing the structure
of a system with the system itself [see footnote 7 and my comment on
Mortazavian’s paper].3 Of special concern to system theorists are the hierar¬
chical relations encountered in most systems and the fact that every system
(except the cosmos seen as “the system of all systems ) is embedded in an
environment that also appears to be subjected to a further hierarchy. Thus,
the concern with many system properties (such as open system, closed
system, etc.) is characteristic for this approach. Our present world picture,
formed by relatively recent insights into atomic and subatomic physics,
astrophysics, and astronomy as well as the biophysical and biomolecular
extension of evolution theory, conforms with such a hierarchy of open sys¬
tems, in spite of the perennial uncertainty about the ultimate building stones
of our universe. Here, too, Wiener’s cybernetics does not seem to pay much
heed to hierarchical relations (and many of their properties) that span the
universe—whether we go from quarks to galactic clusters or through the
hierarchic structure of a cell or larger organism. Might the reason for this
neglect be the fact that Wiener, like many of us, realized only in the 1950s
and 1960s that stars, galaxies, and the universe itself are subject to evolu¬
tionary changes—changes that ensure, on one side, the existence of struc¬
tures and, on the other side, the impermanence of those structures? Only
after such a realization might the notions of structure and hierarchy assume
their full and pervasive significance.

3 A more correct definition is presented by Oskar Lange: “Generally speaking, ‘system is taken
to mean a set of elements together with the set of relations between the elements. The set ot
such relations (and their isomorphic transformations) is called the structure of the system.
(Lange, 1962 and 1965, p. 17.)
 448
RICHARD MATTESSICH

FUNCTION AND GOAL ORIENTATION

One of the most crucial aspects found in system theory, but lacking among
the concepts of traditional science, is the notion of goal orientation or purpo¬
siveness.4 Physics and chemistry can get along without it fairly well (al¬
though the concept of entropy may weaken this statement), but there is
growing evidence that the biological sciences and certainly the social sci¬
ences need the idea of goal orientation and the related notions of preference,
value, efficiency, and so forth, most urgently. And it is no coincidence that
the biochemical theory of the Nobel laureate Manfred Eigen [1971] intro¬
duces a value function into biology that has at least some affinity with the
value function in economics.
Most of the approaches dealing with systems strongly emphasize the
notions of purpose, goal orientation, efficiency and effectiveness of goal
attainment, and so forth. But surprisingly enough (as Eden points out),
Wiener does not deal with such notions in his cybernetics, in spite of the fact
that he considered them in an earlier, coauthored paper. It is revealing for
Wiener’s thought process that the subject index of his work on cybernetics
and society [1950, 1954, and 1967] contains only a single entry under pur¬
pose, and this pertinent passage must be taken as a rejection rather than an
acknowledgment of this very notion: “Here I want to interject the semantic
point that such words as life, purpose, and soul are grossly inadequate to
precise scientific thinking.” (Wiener 1950, 1954, and 1967, p. 45; emphasis
added.) Such a radical attitude with regard to the notion of purpose is not
typical for a general-systems theoretician, mainly because the applied and
social sciences, and to some extent even the biological sciences (see Monod
[1970 and 1971] and Jacob [1971] and their project telenomique possessed by
every organism; see also Eigen [1971]) are in need of various notions of goal
and goal attainment. But some proponents of the systems approach caution
us by asserting that “biology does not have a single law statement, let alone
theory, containing the concepts of plan (or design) and purpose.” (Bunge
1979a, p. 120.) 6 ’

INFORMATION, CONTROL, AND FEEDBACK

The fact that cybernetics (in the narrow sense) is more or less limited to this
area of control information, forms the strongest argument in favor of regard¬
ing cybernetics as a subset of systems research. The insights that (1) a
control state is dependent on the pertinent information-flow and (2) the
scientific laws governing control are universal, hence, independent of the
traditional dichotomy between living and nonliving entities, seem to form

And the hierarchy connected with this notion: state-maintaining, goal-seeking multigoal-
seeking, purposive, and purposeful. [Ackoff, 1971, p. 665.]
CYBERNETICS AND SYSTEMS THEORY 449

the essence of Wiener’s cybernetics and undoubtedly are necessary condi¬

tions, though by no means sufficient conditions, of systems thinking. The
application of the theory of information, homeostasis, and feedback to sys¬
tem theory constitutes a crucial achievement of applied mathematics, but
this application is the natural consequence of a much broader set of empir¬
ical and philosophic ideas.

ENVIRONMENT AND ITS INFLUENCE

From the very beginning of system theory, the notion of environment has
played an essential role. Apart from the fact that it represents the logical
counterpart of the system notion, it performs important analytical and prac¬
tical services in bounding a system and determining whether certain proper¬
ties or phenomena (e.g., norms) are part of the system influencing the envi¬
ronment, or whether they impinge from outside on the system. [Mattessich,
1974a; 1978, chap. 2.] Distinctions—as those between internal versus exter¬
nal structure and relations; subsystem, system, and meta- or supersystem,
and so forth—are consequences of this basic but perhaps artificial dichotomy
between system and environment. There is little evidence that Wiener made
any conscious use of this fundamental notion. And yet, the great analytical
and pedagogic value of the system/environment dichotomy has become one
of the hallmarks of the systems approach. We encounter it everywhere,
whether we listen to a television program in which Jonathan Miller [1978]
demonstrates the analogy between the artificial inner environment of a
stratospheric airplane and the human bloodstream or whether we read a
National Aeronautics and Space Administration report on the no less
artificial environment afforded an astronaut by a space suit.

SYSTEM LAWS AND MATHEMATICAL HOMOLOGIES

An essential feature of systems theory is, or at least ought to be, the search
for general laws underlying all systems and laws underlying more specific
but still fairly general system-types. If at least a few such nontrivial laws can
be found, the systems approach may, indeed, become a theory, although
hardly an entire scientific discipline or a superscience. But if such laws
prove to be elusive or fairly insignificant or too specific, then the systems
approach will remain a mere methodology, the ideas and tools of which will
be absorbed in time by the methodology of general science or some specific
disciplines. At this moment, it seems that the latter alternative is more likely
to occur. Although the systems approach belongs to empirical science [see
my comment on Mortazavian’s paper in this volume], the prospect of for¬
mulating fairly general and yet powerful empirical system hypotheses is not
very bright; at best, we have mathematical homologies, which, however, do
 450 RICHARD MATTESSICH

not make the systems approach an analytical discipline like logic or pure
mathematics. Such homologies may best be illustrated by the isomorphism
between a mechanical and an electrical oscillator. [Compare Rapoport,
19686, p. 455.] From his introductory physics lessons, the reader may re¬
member that the force f(t) impacting a mechanical oscillator is

At) = m -^4- + r + kx,

dtz dt

where x is the spatial displacement of mass m, while r and k are coefficients

of friction and elasticity, respectively (t stands for time and d and d2 are the
differential operators). Surprisingly, the impressed electromotoric force E(t)
of an electric oscillator (as used, for example, in radio and television) is
expressed by the following relation isomorphic to the one previously stated:

E(t) = L^L + rJ3- + Cq,

dt2 dt

where L stands for inductance, R for resistance, C for capacitance (and t and
d or d~ maintaining the previous meaning). This reveals significant structural
homologies between mechanical and electromotoric force, mass and induct¬
ance, mechanical friction and electrical resistance, elasticity and electrical
capacity. Perhaps the best evidence for the nonaccidental and empirical
nature of these formal or mathematical homologies lies in the fact that the
amount of heat produced by overcoming the mechanical friction is identical
to that produced by overcoming the electrical resistance. Cybernetics and
information theory did, indeed, produce such homologies, but again, they
are fairly limited to feedback and homeostatic mechanisms, on one side, or
analogies of entropy and negentropy, on the other side. While the present
practice of system theory—especially applied system theory as practiced,
for example, by the International Institute for Applied Systems Analysis
(IIASA) in Laxenburg, Austria—operates on a much broader empirical basis
but with homologies and system models of a lower level of generalization.

CONCLUSION

These few considerations will, I hope, help to sharpen our distinction be¬
tween cybernetics and the more comprehensive systems approach. But
Eden is quite correct in pointing out that in both areas “we face a procedural
dilemma that “other domains of scholarship do not ordinarily pose.” To
regard both of them as viewpoints and methodological tools instead of scien¬
tific disciplines might be the ultimate resolution. As regards the contradic¬
tion (to use Eden’s expression) between Mesarovic—who leans toward a
purely analytical presentation of system theory (mathematical system
theory) and others, like Bertalanffy, who have a much broader vision of
CYBERNETICS AND SYSTEMS THEORY 451

the systems approach, I may refer the reader to my comments on Mortaza-

vian’s contribution in this volume. There, it will become obvious why other
domains do not ordinarily pose such a problem of contradiction. In physics,
for example, there could be no doubt that such theoretical physicists and
model-builders as John A. Wheeler, Murray Gell-Mann, and Stephen Hawk¬
ing are serving an empirical and not an analytical science, while some sys¬
tem theorists and model-builders, such as Mortazavian, seem to regard
themselves as being in the service of an analytical discipline. The contusion
and apparent contradiction is due to the failure of some mathematical-
system theorists to stress two important and related facts, that (1) applied
mathematics is the application of mathematics to factual sciences and (2)
propositions represented by more or less unrealistic system models are still
empirical propositions independent of their degrees of reliability and real¬
ity.5 Thus, I wholeheartedly agree with Klir, Eden, and others that we must
distinguish between “systems theory as a viewpoint” and “mathematical
systems theory.” But since there is a plurality of system theories, even of
mathematical system theories, there is little justification for using the singu¬
lar for the entire area or subarea. Comprehending it as a viewpoint, we might
talk about a systems approach, which implies its philosophic and especially
its methodological aspects. [Mattessich, 1978.]6 We might also construct, as
Bunge [1979a] does, a systems framework out of which various system
theories may emerge through further interpretation. And for the purpose of
empirical research as well as training scientists, the systems approach can¬
not be limited to the construction of mathematical models, but must include
studying concrete systems and fitting abstract models to those concrete
systems.
To conceive of an abstract system as a set of objects with one or more
relations between at least two of those objects7 seems at first glance to be a
most simple or trivial methodological device. But the same holds true for
Descartes’s idea of representing algebraic functions by the geometric means
of a coordinate system or Einstein’s idea of transforming the physical phe¬
nomenon of the curvature of light (under the influence of mass) into the

5The renowned mathematician Rene Thom offers an interesting discussion about the distinction
between the concrete (morphological) and the abstract (mathematical) notion of a system and
points out that “from a mathematical viewpoint, the systemic approach leads to a rather
sketchy theory.” (Thom, 1980, p. 3.) , ,
6This point of view comes close to that of Rapoport, who suggests that “general systems theory
is best described not as a theory in the sense that this word is used in science but, rather as a
program or a direction in the contemporary philosophy of science. " (Rapoport, 1968b, p. 4,2.)
7There is a crucial difference between this system definition and that of Mortazavian, admit¬
tedly based on that of Mesarovic [1964a], Apart from the fact that we distinguish between a
concrete and an abstract system, our definition of the latter is first of all a special kind of set of
objects and not merely a set of relations between those objects. As can already be seen from
footnote 3,1 respect the importance of this set of relations, but it represents the structure of the
system, not the system itself.
452 RICHARD MATTESSICH

geometric phenomenon of curvature of space or Godel’s idea of mapping all

components of an axiomatic system into the system of natural numbers
(Godel numbers). The proof of the pudding does not lie in the degree of
complexity of a methodological device, but in the consequences emerging
from this device. Undoubtedly, the ideas of Descartes, Einstein, and Godel
have proved immensely fertile. But is this also true for the systems ap¬
proach? Although the latter has gained much popularity, something still
seems to be missing. This is not to deny that conceiving of reality as a
hierarchy of interacting systems with emergent properties has certain meth¬
odological advantages; but we should ask two questions: How vast is the
potential of this idea? And is there still a crucial step to be made in order to
exploit the full power of this methodological device?
I have previously expressed the need for a formal or mathematical super¬
structure of system theory [Mattessich, 1978, pp. 18-24], and perhaps the
solution lies in a comprehensive attempt to axiomatize system theory. The
formalizations by mathematicians such as Kalman, Falb, and Arbib [1969],
Mesarovic, Macko, and Takahara [1970], and above all, Mesarovic and
Takahara [1975], are serious and meritorious attempts in this direction. The
most comprehensive attempt is Bunge’s [1979a] ontological framework of
systemics, which I discussed in some detail elsewhere. [Mattessich, 1982.]
Although at this stage it is difficult to assess the future potential of the
system notion, the systems approach certainly is worth further exploration.8

KIn connection with Eden’s paper and my present comment, it is interesting to note that the
British Journal Kybernetes (Thales Publications Ltd.) initiated the (1977) Norbert Wiener Com¬
petition under the closely related title. “The Unity between Cybernetics and General Systems
Theory," published in its vol. 8 [1979] with pertinent contributions by Counelis (third prize),
Gergely, Majumder (first prize), Mayne, Stanley-Jones, and Zeleny (second prize). I am grateful
to George Gorelik, University of British Columbia, for drawing my attention to this particular
publication.
CYBERNETICS IN THE INFORMATION
DISCIPLINES

Manfred Kochen

Murray Eden concludes his essay on the history of cybernetics with the
implication that the only function of cybernetics was to diffuse the principles
of communication and control engineering into the culture of science and
technology and to integrate itself into it. That has in fact been its primary
effect. But it had more than one intended function. Another was to identify
analogous or identical concepts in diverse fields and analyze those that are
essentially mathematical by appropriate mathematics. It has not yet per¬
formed this function with equal success. The prospect of that happening in
the next decade or two is the first point on which I would like to comment.
The second point concerns Dr. Rosenblueth’s insistence on the need for
teams of scientists to establish fruitful connections among principles of com¬
munication and control engineering and living systems, as reiterated by
Wiener in Cybernetics and quoted by Eden. The need for teamwork is more
urgent today than ever before, not only for interdisciplinary scientific re¬
search but for many survival-related activities. Despite the trend toward
segregation into increasingly narrow specialties, as indicated by Eden, there
are now techniques and signs of hope that this trend could be reversed.
My third point is stimulated by the phrase “the map of science” in the
same quotation of Rosenblueth by Wiener. Cybernetics did not become a
distinctive discipline or specialty as pointed out by Eden, though he hints
that like other intersections of different disciplines, in time it might become
yet another specialty. Under what conditions do new scientific specialties
come into being, survive, wax or wane, merge with others, and die out?
What is the likely role of cybernetics in the emerging information disci¬
plines? A catalyst? One of them? A substrate?
It is quite possible that a new discipline of scientific planning to steer and
manage complex human systems will emerge to perform one of the major
intended functions of cybernetics: interdisciplinary teamwork in both scien¬
tific research and social development on a set of concepts related to syn¬
ergetics. [Haken, 1980, p. 121.]
453
454 MANFRED KOCHEN

STIMULATING NEW MATHEMATICS AND COMPUTATIONAL

ALGORITHMS

The attempt to link principles of communication and control to living sys¬

tems by searching for essentially mathematical concepts common to the
study of man-made and living systems can lead to new mathematics. One of
the first such contributions came from Alan Turing’s model of two interact¬
ing chemical species. [Turing, 1952.] This was a precursor to the theory of
dissipative systems and more generally to the study of cooperative phe¬
nomena. [Nicolis and Prigogine, 1977; Eigen and Schuster, 1977, 1978;
Thom, 1975.] It appears to have stimulated or at least reinforced the emer¬
gence of the mathematical specialty known as bifurcation theory. Though it
had mathematical roots 200 years ago in Euler’s time and was greatly ad¬
vanced by Wiener’s coauthor, Hopf, it has undergone vigorous development
only recently. [Rollo May, 1973; Hopf, 1943; Sattinger, 1980; Stakgold,
1971; Arnol’d, 1972; Mittelmann and Weber, 1980.] It is the study of equa¬
tions whose solutions bifurcate or branch out when a parameter in the equa¬
tion is changed. Renewed interest in bifurcation theory was stimulated by
results about cascading bifurcations and chaotic regimes.
It is not quite true, as asserted in the quotation of Minsky in Eden’s paper,
that the approach based on “perceptrons,” or self-adjusting learning de¬
vices, has subsequently died out: A major effort on related lines of inquiry is
underway at Rockefeller University. [Edelman, 1981; Edelman and Mount-
castle, 1978.] It is based on the Darwinian notion that there must be
sufficient preexisting diversity for self-selection of neuronal groups that re¬
spond to external patterns and whose success is reinforced by an ampli¬
fication process. A second-order repertoire that operates selectionistically
on the neuronal groups of the first-order repertoire is also postulated.
Though this model is being analyzed with the help of a complex computer
simulation, there appears to be a rather profound theorem underlying this,
and this may stimulate the development of a mathematics of patterns of
considerably greater sophistication than that which already exists.1
The mathematics underlying the study of pattern recognition appears to
be in ferment. Not only are there promising signs of better understanding of
the problem from novel psychological, engineering, and physiological per¬
spectives [Uhr, 1982; Rosenfeld and Kak, 1976; S. R. Sternberg, 1980;
Ermentrout and Cowan, 1980; Blaivas, 1977] but there are signs of new
mathematics as well [Grenander, 1976; 1978].
The reliability of the human nervous system is remarkable. [Asratyan and
Simonov, 1975.] It has been known for some time that large parts of a rat’s
brain can be extirpated without major behavioral deficits, suggesting that the
brain acquires its reliability at the cost of redundancy. This is also what

1 Marvin Minsky and Seymour Papert, Perceptrons [1969]. The invention of the perceptron was
reported by Frank Rosenblatt [1958] and developed more fully in Rosenblatt, Principles of
Neurodynamics [1962],
CYBERNETICS IN THE INFORMATION DISCIPLINES 455

engineers have been doing in designing reliable computer systems. John von
Neumann was the first to see the essentially mathematical problem common
to both engineered and living systems, and this has led to a line of research,
the full potential of which has yet to be developed, [von Neumann, 1956;
Moore and Shannon, 1956; Kochen, 1959; S. Winograd and Cowan, 1963.]
Though the stimulus to several other promising areas of new mathematics
could be traced to Wiener and cybernetics, only two will be mentioned. The
first is the use of probabilistic and computational methods in even as pure a
branch of mathematics as number theory.
To test whether a number n, less than, say, 2400 —593, is prime, select a
random sample of 50 numbers a\, , a50, each less than n, and test if a, is a
witness to the nonprimality of n by checking that n does not divide a1~ 1 for
each i. If all a, fail this test, call n prime; the probability of error is less than
(1/4)50 or (1/2)100. The computational complexity of such an algorithm is far
less than that of a nonprobabilistic one. Somewhat similarly, the four-color
conjecture has been settled with the help of a computer analysis of cases.
While such methods are not universally accepted in the mathematical com¬
munity, they are increasingly used by its leaders and may contribute to the
way we think in mathematics. The idea of thinking stochastically is as pro¬
found as the Newtonian idea of thinking continuously or the Poincare idea of
thinking qualitatively. One of the roots of thinking stochastically is in a
central tenet of Wiener’s formulation of cybernetics: Our capacity to control
is inherently incomplete and necessitates a statistical approach.
The same tenet may also have been a precursor to “fuzzy-set theory.”
Control engineers, interested in designing a system that could park a car, for
example, did not believe that the classical mathematical concepts used in
designing control systems were appropriate. [Zadeh, 1965; Lowen, 1981;
Goguen, 1967; Smets, 1982.] A person performs the task in what seems to be
a totally different way. He or she does not measure the variables fed back to
the visual system with any greater precision than he or she specifies control
variables with the motor system. Nor did probabilistic concepts appear to be
fruitful. Instead, a new kind of mathematics was spawned that started from a
generalization of the characteristic function of a set, replacing its two-valued
range by the interval [0, 1]. Thus, the characteristic function of the set of
large numbers for a particular judge was interpreted as the degree of mem¬
bership in the class (fuzzy set) of large numbers assigned to any number x by
that judge. For x - 1 to 10, it might be 0, for x = 11 to 50, 0.1; for x = 51 to
100, 0.2, and so forth; and for x > 1000, it might be 1. Fuzzy-set theory has
stimulated contributions of a purely mathematical nature, though its future
standing among the mathematical specialties cannot yet be forecast.

THE PROSPECTS FOR TEAMWORK

It would appear that contemporary young scientists are more broadly and
deeply trained than their predecessors. Nearly every good science student
456 MANFRED KOCHEN

encounters in undergraduate education more basic modern mathematics,

physics, chemistry, and biology than was known when the previous genera¬
tion attended school. This is the result of the doubling of knowledge every
decade and the availability of ever better aids to learning. Thus, few phy¬
siologists today ought to be ignorant of mathematics and computation.
Yet, the very growth of knowledge has forced each contributor to confine
attention to an increasingly narrow area of specialization in which he or she
can keep up with all that is important to that area. In staying within the
bounds of cognitive capacity, he or she has to decide not to keep up with
what falls outside that specialty. This lack of keeping up may lead to loss of
fluency, competence, confidence. Thus, a physiologist whose specialty is
not mathematical physiology may be considered to be ignorant of mathemat¬
ics.
Many of the tasks facing researchers as well as planners and decision¬
makers concerned with communications and control are increasing in
complexity. Rapid progress in identifying neurobiochemical constituents,
physiological functions, and anatomical structures of the nervous system
has vastly increased the number of phenomena to be explained. Hormones,
peptides, and opiates are made in the brain, and there are special receptors
for them. The nervous system shows a remarkable degree of genetic prede¬
termination and neuronal specificity.
The difficulties of controlling very large computer operating systems or
dealing with many large, diverse databases or with the many incompatible
parts of a large computer-communication system are being recognized. It is
rare that one person can master the overall structure of a complex modern
system.
The number of variables representing first-order effects has increased.
What were higher order variables a few decades ago (e.g., air pollution from
automobile exhaust) that could be neglected in an engineering approxima¬
tion are now of first order. Interaction among variables is greater. The stakes
are higher. There is more knowledge that must be sorted for relevance at the
same time that there are fewer directly relevant experiences to guide us in
coping with complex tasks.
We can cope by fragmenting the task or through teamwork. The trend has
been toward reductionist analysis or fragmentation. As noted previously,
each specialist is restricted to what he or she can cope with, and communi¬
cates at best with follow specialists. It is as if 49 blind men were grouped into
7 groups of specialists, each group thoroughly probing one part. Wiener and
Rosenblueth realized that cybernetics, as a program, could not flourish in
this way, and they insisted on a broader view of teamwork. The need has
now become urgent.
The hope is that teamwork may now also become feasible. Despite the
remarkable degree to which the scientific community responded with team¬
work on a grand scale to the challenges of World War II, Wiener was not
optimistic. There are perhaps fewer grounds for optimism today. But many
CYBERNETICS IN THE INFORMATION DISCIPLINES 457

people in all walks of life seem spontaneously to have become aware of a

threat far greater than that of World War II: the extinction of mankind by
nuclear holocaust. The corresponding challenge is disarmament followed by
the replacement of national sovereignty in the international system. [Schell,
1982.] Grass-roots responses of a mass nature, from scientists and nonscien¬
tists alike, appear to be organizing themselves and may lead once again to
cooperative teamwork toward insuring survival. The threat of irresponsible
or accidental military action is not the only challenge. The possibility that
information technologies will pervade all aspects of the economy and dra¬
matically reduce the total world’s need for human labor is another challenge
requiring cooperative and perhaps equally radical responses. The threat to
our biosphere and ecological balance is yet another major challenge that is
being temporarily put aside.
The new technologies themselves, for example computer conferencing
and computer networks, may help with team formation and maintenance.
Trends toward decentralization may influence corresponding trends toward
teamwork. [Kochen and Deutsch, 1980.] The possible leveling effects of
computers on the distribution of power may generate changes in our reward
system, shifting some of the emphasis from competition to cooperation.
These are hopes, not expectations. But they are all we have.

A NICHE IN THE MAP OF SCIENCE?

It is now possible to display a “map of science” in graphic form. By defining

the proximity of two published papers according to, for example, how many
other papers each cited both of them, it is possible to group the papers as
well as their authors into proximity clusters. It is curious that given the
“distance” between every pair in a large sample of papers, the application of
multidimensional scaling techniques results in a map for which just two
dimensions suffice. [Small and Griffith, 1974.] It is then possible to see how
this map changes from year to year; how some clusters break apart, move
toward or away from others or fuse with others; and so on. Each such
cluster can be interpreted as a specialty at some level.
If cybernetics were to re-emerge as yet another specialty, this method of
mapping science would detect it, albeit with a great deal of interpretation.
The American Society for Cybernetics has recently, under new leadership,
experienced something of a renaissance. But that can also be said for various
related areas discussed by Eden: systems theory, control theory, communi¬
cation theory, and so forth. Systems theory and communication theory seem
to have a greater variety of multiple meanings than does cybernetics. There
are at least three disparate versions of systems theory and as many totally
different connotations of communication theory. The information disciplines
can be said to encompass all these areas; yet, cybernetics stands out as one
of their major common ancestors.
.

'
 CYBERNETICS
The View from Brain Theory

Michael A. Arbib

CYBERNETICS AN UMBRELLA TERM

Murray Eden starts his review with the sentence: “Norbert Wiener, in his
book entitled Cybernetics, or Control and Communication in the Animal
and the Machine, did not define explicitly the word he believed he had
coined.” Surely, the book’s title is, at least implicitly, a definition. How¬
ever, Eden is correct to stress that Wiener’s book is intensely personal and
brings together a vast array of hitherto disparate topics. He places Wiener’s
book in a historical perspective that goes back to Hero of Alexandria and
reaches the present day via Descartes and Ampere, among others. I trace a
similar history in my companion article (“Cognitive Science: The View from
Brain Theory”), where I argue that the history of cybernetics is, in large
part, also the history of cognitive science. This is consistent with Eden’s
discussion of the permeation of psychology by cybernetic concepts and
counterbalances Pylyshyn’s overemphasis on mathematical logic and
artificial intelligence (AI) as driving forces in the development of cognitive
science.
In our department at the University of Massachusetts at Amherst, we use
the word cybernetics to refer to a conjoined study of brain theory and AI.
[Lesser, 1981-1982.] However, there are many experts in AI who regard
cybernetics as an outdated, outmoded term. They insist that there is no need
for an umbrella term that relates AI to brain theory. Eden notes that others
use the word cybernetics in the sense of any study of complex systems that
is aided by a computer—a broad usage encompassing even mathematical
economics. (This is particularly so in the Soviet Union, see [Arbib, 1966].)
Finally, he suggests that, while some workers still use the word cybernetics,

This paper refers to research supported in part by NIH grant NS 14971 and NSF grant MCS
8005112 at the University of Massachusetts. Portions of the paper are adapted from a paper
prepared for a symposium marking the 25th anniversary of the publication of Wiener’s Cyber¬
netics. [Arbib, 1975b.]

459
460 MICHAEL A. ARBIB

much of the original energy of the field has moved to a vast range of descen¬
dant disciplines. I think that this is true but still believe the umbrella term
serves well to remind workers in different disciplines of the range of com¬
mon concepts they share.

THE FATE OF WIENER'S CONCEPTS

In this section, I want to review briefly a number of topics in Cybernetics

and discuss their appearance from a thirty-five-year perspective. I shall
follow Eden in analyzing Wiener’s book rather than analyze the other roots
of cybernetics.

Statistical Mechanics

As Eden stresses, Wiener’s book reflects strongly the influence of his rich
experience in using statistical methods in control, communication, and fil¬
tering problems. In World War II, Wiener had studied problems of tracking,
interpolation, smoothing, and prediction, particularly in relation to anti¬
aircraft gunnery. The great insight of this study had been that we should not
ask, “Here is a trajectory; now what is the best way of continuing it?’’
without having some information about what ensemble that trajectory be¬
longs to. Having characterized an ensemble by the noise statistics of a sto¬
chastic process, we could make meaningful predictions, such as, “This is
the proposed continuation of the trajectory to which the possible continua¬
tions will, in some well-defined statistical sense, come as close as possible.’’
In his book, Wiener sought to unify these mathematical studies with his
discussions with psychologists and physiologists, to lay the foundations for a
new field, to be called cybernetics. Let me quote a passage from Wiener’s
book where he presents his views:

. . . the theory of the sensitive automata is a statistical one. ... To function

adequately, [the automaton] must give a satisfactory performance for a whole
class of inputs, and this means a statistically satisfactory performance for [the]
class of inputs . . . which it is statistically expected to receive. Thus its theory
belongs to the Gibbsian statistical mechanics .... (Wiener, 1961, pp. 43-44.)

Interestingly, many people working in cybernetics today (whatever their

subdiscipline) do not hold this particular view. While workers continue to
relate the study of brains, biological systems, or social systems with the
study of mathematical and engineering systems, statistical methods are not
of central importance. For example, the AI approach essentially ignores
statistical questions, focusing on whether a heuristic or knowledge represen¬
tation can help solve a problem most of the time. But this “most of the time’’
is not in any statistical sense, and there is no question of having an ensemble
CYBERNETICS: THE VIEW FROM BRAIN THEORY 461

and using the techniques of statistical mechanics to estimate “the statistics

of success.”
Turning to brain theory, we find that models today are not statistical. We
went through a period of glorious ignorance when, not knowing enough
about the brain, we looked with awe at the great complexity of the neural net
and said it must be random. Much work in the 1950s and 1960s sought
information about the average activity of a statistically connected neural
network, with hopes of explaining the alpha-rhythm from the statistics of
such a network. In recent years, though, I would say that brain models have
become more detailed and more deterministic as we have had much more
detailed neurophysiological and neuroanatomical information. Those mod¬
els that are statistical do not usually employ the techniques of Gibbsian
statistical mechanics, but instead emphasize learning algorithms within the
tradition of self-organization and pattern recognition. [Amari, 1974; Gross-
berg, 1976; Sutton and Barto, 1981; Widrow, 1964.]
I would say, then, that the Gibbsian statistical-mechanics approach is
dead in the classic core of cybernetics. However, as I will discuss in the
section on brain theory, cognate methods—the theory of competition and
cooperation in neural nets—are now gaining increasing importance.

Feedback and State Variables

Wiener knew from his engineering experience that if the gain is too high in a
feedback system, it will go into unstable oscillations. This led him to suspect
that there might be a form of human brain damage that would cause the
human’s limb to go into violent oscillations when the patient tried to pick
things up. Sure enough, that turned out to be the case—his clinician friends
showed him such an effect in patients with cerebellar lesions. This is perhaps
the root of the whole development of control-theoretic analysis of complex
biological systems. [Milsum, 1966.]
Wiener’s discussion of filtering, feedback, and stability does not use the
idea of a state variable; rather, he discussed how, given the infinite history of
the machine’s input, we could specify the infinite story of its output. He did
not ask such questions as “What is going on inside the box?” or “How can
we relate the internal dynamics to the input/output behavior?” It is Rudolf
Kalman, perhaps as much as anyone, who has made state-variable thinking
mandatory in system theory. [Kalman, Falb, and Arbib, 1969.]
As Eden notes, Wiener suggested that “the computing machine must
represent almost an ideal model of the problems arising in the nervous
system.” This might seem at cross-purposes with the previous emphasis on
statistical mechanics. However, Kalman and Richard Bucy provide a rap¬
prochement, for they showed how to replace the history-based statistical
extrapolation method, known as the Wiener-Hopf filter, by the state-based
recursive estimation technique, known as the Kalman-Bucy filter—which,
implemented in computer hardware, is an integral part of guidance systems
462 MICHAEL A. ARBIB

in today’s planes, missiles, and spacecraft. [Kalman and Bucy, 1961.] Inter¬
estingly, the extrapolation problem has recently resurfaced in brain theory,
with the look-ahead modules of Andras Pellionisz and Rodolfo Llinas,
though, unfortunately, they have not tried to apply these earlier successful
techniques from system theory. [Pellionisz and Llinas, 1979.]

Internal Models

When a system is carrying out some control task, its parameters may not be
quite matched to its job. This suggests the need for what Wiener calls infor¬
mative feedback and what most of us today would call either model-
referencing or feed-forward. Here, we augment the core feedback-system by
using an identification procedure to upgrade the parameters of the core
system at the same time that it is locked into a feedback loop for ongoing
control.
What is very interesting from our present perspective is that despite this
feed-forward idea, there is no mention in Wiener’s book of the work of the
psychologist Kenneth Craik. Craik wrote perhaps the first work in cyber¬
netic psychology—though, of course, not yet with that name. [Craik, 1943.]
He looked at the nature of explanation and stressed the idea of our building
in our heads a complex model of our world, not in the cardboard-and-glue
sense, but, nonetheless, a model that could modulate our input/output be¬
havior to match our experience of our world. This idea, not in n-dimensional
differential-equation format, but in a very general data-structure format, has
become crucial to much of the exciting work in artificial intelligence. We saw
Shakey the robot using visual input to build an internal model of the world
about which it will navigate [Nilsson and Raphael, 1967], while Terry
Winograd based his approach to linguistics on having the computer able to
refer to a model of its simple world of blocks on a table top when “disambig¬
uating” sentences that it is using. [Winograd, 1972a.] The last ten years of
AI have been dominated by a retreat from general methods and an increasing
emphasis on techniques for representing specialized domains of knowl¬
edge—the so-called “expert systems.” [R. Davis and Lenat, 1982; Short -
liffe, 1976.]
So it goes on. We could relate the work in neurophysiology since 1948 to
Wiener’s interests; we could follow Wiener’s notions on intermodality cod¬
ing; and we could discuss at length the social problems that have become an
inescapable reality since 1948. Nevertheless, here I must leave this retro¬
spective look at Wiener’s book. What I hope to have done is confirm Eden’s
demonstration that Wiener had a broad intellectual range and many of the
questions Wiener raised are still very much alive, though transformed and
enriched by the developments—both technical and conceptual—of the last
35 years.
CYBERNETICS: THE VIEW FROM BRAIN THEORY 463

BRAIN THEORY

A few years ago, I discussed two specific aspects of my research that may be
viewed as having descended from Wiener’s work (among others). [Arbib,
1975b.] First, I explored some contributions in modern system theory—the
descendant of Wiener’s control theory—by giving a nonmathematical pros¬
pectus for the highly mathematical study of “machines in a category” that
Ernest Manes and I conducted at the University of Massachusetts (reviewed
by [Arbib and Manes, 1980]) and, second, I turned to a discussion of brain
theory. Rather than repeat here the discussion of machines in a category, I
want simply to recall its conclusion. We concluded that our general insights
into the functioning of systems had come about only because we could use
category theory to build on the theories of sequential machines, linear sys¬
tems, and tree automata. We had a sufficiently rich set of particular theories
as examples, and we had a sufficiently general mathematical theory to pro¬
vide the language for their unification. Brain theory, unlike mathematical
systems theory, is only now building up a reasonable stock of good examples
of theories of limited applicability, which may serve as a basis for developing
powerful general techniques that can be brought to bear on a wide variety of
specific problems relating the physiology and anatomy of neural circuitry.
For example, early studies of the role of lateral inhibition in, for example,
contrast enhancement in sensory systems can now be related to models of
mode selection, where emphasis is on directing attention to one mode, fea¬
ture, or set of features of activity, rather than on preprocessing an array for
further computation. [Ratliff, 1965.] These models are in turn part of a broad
class of models dealing with competition and cooperation in neural nets.
One of the first models of decision-making in neural circuitry to explicitly
opt for cooperative computation was the S-RETIC model. [Kilmer, McCul¬
loch, and Blum, 1969.] It suggested how the reticular formation of the brain
stem might switch the organism’s gross state from one overall mode of
behavior in the organism to another. Based on the anatomical work of
Madge and Arnold Scheibel, Kilmer and McCulloch believed that a rea¬
sonable structural simplification of the system was a stack of poker chips.
[Scheibel and Scheibel, 1958.] (The Scheibels’ work may be cited as one of
the earliest contributions to modules of neural structure intermediate be¬
tween brain region and neuron. [Mountcastle, 1978.]) In the S-RETIC
model, each module in the stack receives a sample of the overall system
input and on the basis of that sampling, assigns weights that provide likeli¬
hood estimates for each of the different modes. Initially, these estimates are
based only on a local sample. However, the modules are coupled in such a
way that each module readjusts its weights on the basis of activity in other
modules to which it is connected. Kilmer and McCulloch suggested a con¬
nection scheme that would lead to eventual consensus, with a majority of the
modules assigning the greatest weight to a single mode—all without any
executive control.
464 MICHAEL A. ARBIB

We now turn to the topic of segmentation of visual input from two eyes on
the basis of depth cues. Bela Julesz designed random-dot stereograms in
which each eye receives a totally random pattern but in which different
regions in the two inputs are identical save for a shift in position, yielding a
different disparity on the two retinas. [Julesz, 1971.] Although such a pattern
for a naive subject can initially appear to be visual noise, disparity-matching
takes place eventually, and the subject perceives surfaces at different
depths. Neurophysiologists have found that cells in a cat’s visual cortex are
tuned for retinal disparity, and similar cells are posited in the human. What
presumably causes the initial perception of visual noise is that in addition to
the correct correlation of points in the two retinas, there are many spurious
correlations, and computation is required to reduce them.
Parvati Dev was one of several workers to propose that the cells of a
given disparity be imagined as forming a population arrayed in a spatial map
corresponding to the map of visual direction. [Dev, 1975.] Connections be¬
tween cells could then be arranged so that nearby cells of a given disparity
would be mutually excitatory, whereas cells nearby in visual direction but
different in disparity would have inhibitory interaction. In this way, the
activity of the array would be organized into a pattern where cells of only
one disparity-type would be highly active in each region of visual direction.
As a result, the visual input would eventually be segmented into a number of
distinct surfaces. (See David Marr’s posthumously published magnum opus.
Vision, for an influential approach to such problems. [Marr, 1982.])
In the stereopsis model, then, we have competition in the disparity dimen¬
sion and cooperation in the other dimensions. We note the similarity to S-
RETIC, where the cooperation dimension is the row of modules, and com¬
petition is between modes rather than disparities. Other workers have
shown that a number of interesting features of visual psychophysics can be
captured in a fairly simple neural network in which we have two interacting
populations of cells—excitatory and inhibitory—with plausible interconnec¬
tions. Such a model exhibits hysteresis, one of the most famous psychophys¬
ical manifestations of which is seen in fusion. Such hysteresis phenomena
are also central to the study of stereopsis. Informal observations of this kind
have laid the basis for rigorous mathematical analysis, which also subsumes
an early model of frog prey-selection. (We note that these neural mecha¬
nisms are similar to the relaxation methods for segmentation and region
labeling in machine vision. [Hanson and Riseman, 1978.]) A recent United
States-Japan seminar brought together applied mathematicians, brain theo¬
rists, and neuroscientists to present a body of experimental and theoretical
techniques in brain theory. The papers, collected by Amari and Arbib, give a
good picture of both successes in building general theory and also the many
strands of theory and experiment that still need to be tied together—or
trimmed away completely. [Amari and Arbib, 1982.]
Another area of brain theory of continuing interest is the control of move¬
ment, including Nikolai Bernstein’s theory of synergies (motor schemas)
CYBERNETICS: THE VIEW FROM BRAIN THEORY 465

[Bernstein, 1967; Gel’fand, Gurfinkel’, Tsetlin, and Shik, 1971], but it is

beyond the scope of this paper to review related studies of cerebellum and
motor learning and studies of the role of optic flow in controlling locomotion.
[See, for example, Arbib, 1981; Szentagothai and Arbib, 1974.]
Eden quotes Wiener’s observation that the work of McCulloch and Wal¬
ter Pitts linked neurophysiology and mathematical logic, using the all-or-
none property of neuron firing to model the neuron as a binary discrete-time
element. [McCulloch and Pitts, 1943.] This was a historically valuable obser¬
vation, for it linked the study of effective computability to the study of
neural nets and played a seminal role in the later development of automata
theory. [Shannon and McCarthy, 1956.] However, modern brain theory no
longer uses the binary model of the neuron, instead using continuous-time
models that either represent the variation in average firing rate of the neuron
or actually capture the time course of membrane potential. [Rail, 1964;
Rinzel, 1978.] It is only by containing such correlates of measurable brain
activity that brain models of the kind just discussed can really feed back to
biological experiments. Such models also require the cybernetician to know
a great deal of detailed anatomy and physiology as well as behavioral data.
As I say elsewhere in this volume, I believe that the study of visuomotor
coordination is one where the relation between brain theory and artificial
intelligence is a very healthy one, for workers in both areas can fruitfully
share ideas on how complex psychological functions may be distributed over
an array of subsystems. We might then hope to break down these subsys¬
tems into neural terms rather than attempting to pass immediately from the
overall organismic behavior to the activity of the individual neurons.
Cybernetics, at least in the areas of mathematical systems theory, brain
theory, and artificial intelligence is alive and well. But, as Eden observes,
whether or not scientists working in those areas care to call what they are
doing cybernetics is an entirely different question.
'
 CYBERNETICS
Closing the Loop

Murray Eden

In my paper, I have attempted to describe the place of cybernetics within the

current constellation of scholarly disciplines by attending to Norbert Wien¬
er’s published statements and to two histories: that of the usage of cybernet¬
ics and the scientific fields Wiener included within cybernetics. Each of the
commentors has added to the picture in ways largely complementary to mine
and to one another. The differences between us are largely interpretative
nuances.
Michael Arbib uses cybernetics “to refer to a conjoined study of brain
theory and artificial intelligence.” Indeed, in his comments on the lead paper
by Pylyshyn, Arbib suggests that cognitive science is Cybernetics Redux.
Having made this identification, Arbib discusses the influence of Wiener’s
ideas on studies of brain and behavior. He notes that some of the mathemat¬
ical tools Wiener considered to be central have been superseded; new con¬
cepts and tools have transformed modern attempts to relate mathematical
brain theory to observations by neurophysiologists, psychophysicists, and
psychologists. Whereas Arbib is correct in discerning a direct line of descent
from Wiener’s cybernetics to current studies in artificial intelligence, his
definition of cybernetics is too narrow, in that it appears to exclude com¬
munication among animals or humans. Wiener devotes the last chapter of
Cybernetics to “Information, Language and Society.” [Wiener, 1948 and
1961.] It is true that this chapter is, in large measure, a polemic against the
venality of the free market, and the stupidity of the State, but before Wiener
arrives at that point, he discusses the cybernetic aspects of intercommunica¬
tion among social animals. Arbib’s predilection leads him to the conclu¬
sion—a correct one given his definitions—“that the Gibbsian statistical-
mechanics approach is dead in the classic core of cybernetics.” But, as with
artificial intelligence, theories intended to describe information exchange
among humans or animals—for example, theories for explicating the social
behavior of animals or theories of pattern analysis—exploit the common
patrimony of cybernetics. Both are heavily statistical. Thus, we note that
Manfred Kochen reports the studies of pattern recognition to be sources of
467
468 MURRAY EDEN

better understanding and new mathematics. The work of the probability

theorist Ulf Grenander to which Kochen refers, accords statistics a central
role. [Grenander, 1976; 1978.]
Kochen observes that scientists who want to use cybernetics must begin
by looking for matches between constructs in their fields of study and the
concepts of cybernetics. The analogies will be fruitful if they lend them¬
selves to mathematical description and can be analyzed by appropriate
mathematics. Kochen’s crucial point is that when suitable mathematics does
not exist, it must be invented; hence, cybernetics can lead to new mathemat¬
ics. This is hardly a function unique to cybernetics. Scientists borrow good
concepts wherever they can find them; witness the older use of the notions
of mechanics and kinematics in fields as disparate as economics (e.g., elas¬
ticity) and population genetics (e.g., selection pressure). Nonetheless, the
point is well taken.
Kochen cites, as his earliest example, Alan Turing’s remarkable model
for generating morphological symmetries from a family of arbitrary but
chemically plausible reactions. Kochen follows this thread to the theory of
dissipative systems and bifurcation theory. He cites other progressions that
are, historically speaking, largely autonomous—theories of learning ma¬
chines, patterns, reliability, fuzzy sets, and number theory. Even if some of
the connections to cybernetics are tenuous, this is an impressive list.
Richard Mattessich undertakes to demonstrate that cybernetics (in the
narrow sense) is only a part of system theory. He argues that there are six
ideas basic to systems thinking, of which only one is fundamental to cyber¬
netics. Mattessich points out that Wiener was almost entirely unconcerned
with the problems we must face in explaining such biological or social phe¬
nomena as the emergence of new properties, evolutionary change, hierarchi¬
cal structures, the environment and its influences. He also states that Wiener
did not deal with notions of purpose or goal orientation in his cybernetics.
This latter statement is not precisely correct. Four years before Cybernetics
appeared, Wiener coauthored a paper entitled “Behavior, Purpose and Tele¬
ology.” [Rosenblueth, Wiener, and Bigelow, 1943.] Wiener later described
its genesis; he and Bigelow, an engineer, had asked Rosenblueth, a physi¬
cian and physiologist, the following question:

Is there any pathological condition in which the patient, in trying to perform

some voluntary act like picking up a pencil, overshoots the mark, and goes into
an uncontrollable oscillation? Dr. Rosenblueth immediately answered us that
there is such a well-known condition, that it is called purpose tremor, and that
it is often associated with injury to the cerebellum.

We thus found a most significant confirmation of our hypothesis concerning the

nature of at least some voluntary activity. . . . This seemed to us to mark a new
step in the study of that part of neurophysiology which concerns not solely the
elementary processes of nerves and synapses but the performance of the ner¬
vous system as an integrated whole.
 CYBERNETICS: CLOSING THE LOOP 469

The three of us felt that this new point of view merited a paper, which we wrote
up and published. (Wiener, 1948, p. 8.)

It would appear that even in this paper for philosophers of science,

Wiener was concerned, not to explain nor use the concept of purpose, but to
explain the manifestation of purpose by analogy to a control problem. In the
larger sense, Mattessich is right; Wiener rejected purpose as a word
“grossly inadequate to precise scientific thinking.” Indeed, Wiener was not
a general-systems theoretician.
I believe Mattessich is successful in his intention “to sharpen our distinc¬
tion between cybernetics and the more comprehensive systems approach.”
As Mattessich points out, the founders of system theory were dissatisfied
with reductionism as the scientific paradigm for explaining the phenomena of
biological and social science. Therefore, they turned to a different set of
ideas and components that, in Mattessich’s view, comprise modern systems
theory. Wiener approached science quite differently. Although Wiener in his
early years was a student of philosophy and logic, by the time he occupied
himself with the notions of cybernetics, he had become an applied mathe¬
matician. His interest was in quantitative science, the invention or use of
mathematical theories that modeled a science with empirical content. He
searched almost obsessively for analogies, in neurology for example, to test
with his mathematical paraphernalia. Undoubtedly, much of what he consid¬
ered needed to be treated at the phenomenalistic level, but I find no evidence
in his writings that he believed ultimate reduction to mechanism was un¬
likely. Quite the contrary, he looked to statistical mechanics for the real
answers. He may have been wrong in this particular view, but his approach
is still well within the mainstream of current approaches in neurophysiology,
psychophysics, and perceptual psychology.1
I suggest that Wiener’s attitude toward science is the key to understand¬
ing his lack of concern with the problems that are central for systems theor¬
ists. In a few instances, as with purpose, he regarded them as nonproblems,
because they were “inadequate to precise thinking.” In other instances, for
example, on issues of political or social science, he saw no prospect of
success, because of technical reasons: The time series are too short and
there are too many instabilities. He is quite explicit. He states:

I mention this matter because of the considerable, and I think false, hopes
which some of my friends have built for the social efficacy of whatever new

1 He is in good company. John von Neumann began his seminal paper on probabilistic logics
with the statement: “Our present treatment of error is unsatisfactory and ad hoc. It is the
author’s conviction, voiced over many years, that error should be treated by thermodynamical
methods, and be the subject of a thermodynamical theory, as information has been, by the work
of L. Szilard and C. E. Shannon. The present treatment falls far short of achieving this, but it
assembles, it is hoped, some of the building materials, which will have to enter into the final
structure.” (von Neumann, 1956, p. 43.)
470 MURRAY EDEN

ways of thinking this book may contain. They are certain that our control over
our material environment has far outgrown our control over our social environ¬
ment and our understanding thereof. Therefore, they consider that the main
task of the immediate future is to extend to the fields of anthropology, of
sociology, of economics, the methods of the natural sciences, in the hope of
achieving a like measure of success in the social fields. From believing this
necessary, they come to believe it possible. In this, I maintain, they show an
excessive optimism, and a misunderstanding of the nature of all scientific
achievement. (Ibid., p. 162.)

We note that in the passage quoted, Wiener implied that his were “the
methods of the natural sciences.”
Peter Elias has contributed a valuable addition to the brief history I have
presented. His comments on the Soviet response to the idea of cybernetics
are particularly welcome. His explanation of the difference between Soviet
and Western usage of the term cybernetics is both simple and plausible.
Elias points out that there were, almost from the beginning, different at¬
titudes toward cybernetics in the United States and Europe. Sensitivities
with respect to terminology were heightened even within the community of
European workers. Elias cites W. Ross Ashby as the first author of a text on
cybernetics. [Ashby, 1956.] Curiously enough, Ashby, in his earlier Design
for a Brain makes only one reference to cybernetics. [Ashby, 1952 and
I960.] Colin Cherry, another early English worker in the field, much pre¬
ferred the label communication theory. For him, a part of the theory was
“the theory of feedback (sometimes called cybernetics).” (Cherry, 1957, p.
21.) Later he writes:

This wider field [communication theory and physics] ... is referred to, at least
in Britain, as information theory, a term which is unfortunately used elsewhere
synonymously with communication theory. Again, the French sometimes refer
to communication theory as cybernetics. It is all very confusing! (Ibid., p. 216.)

As a further example of the terminological dissonance, I am quoting from a

foreword that Meyer-Eppler, a German communication scientist and pho¬
netician, wrote in his capacity as editor of a series on Kommunikation und
Kybernetik, in the first volume of that series, a book he had written himself:

Max Bense has contrasted the classical (Archimidean) world emphasizing en¬
ergy and work with the nonclassical (Pascalian) world emphasizing information
and communication. Although some aspects of the nonclassical world have
been known for a long time, it has become accessible to systematic research,
transcending all narrow overspecialization, only thanks to two essentially
mathematical theories, information theory on the one hand and theoretical
cybernetics on the other. ... It is the task of information theory to open up to
quantitative and structural comprehension either communication between per¬
son and person, as it manifests itself in exchanges of symbols, or communica¬
tion between man and the world, which essentially amounts to observation; by
CYBERNETICS: CLOSING THE LOOP 471

contrast, cybernetics as “science of relations” (N. Wiener) employs mathe¬

matical methods to study regular patterns of behavior of highly complex ener¬
gized systems, such as information-processing “machines,” living organisms,
and groups of organisms. (Meyer-Eppler, 1959, editor’s foreword. Emphasis in
the original. My own translation—M.Eden.)

The field of scholarship that the commentators and I are trying to charac¬
terize—I dare not try to assign it a name—appears to encourage terminolog¬
ical invention and yet greater confusion. For example, Kochen, disagreeing
with Minsky as quoted by me, notes that perceptron-related work is being
actively pursued. The work Kochen cites is quite close to work that Arbib
includes within artificial intelligence. Minsky’s definition is implied by the
title of his book (with Papert), Perceptrons—An Introduction to Computa¬
tional Geometry. [Minsky and Papert, 1969.] On the other hand, Rosenblatt,
who coined the term perceptron, identifies it with a class of “brain models.”
[Rosenblatt, 1962.]
Mattessich has got it right when he attributes this tangle to “the human
urge to devise impressive labels.” I share with all four commentators a
measure of regret that the taxonomy of the information sciences is still so
ambiguous. Nevertheless, insofar as progress in scientific understanding is
concerned, there is no great loss. Fortunately, whatever the global label,
each particular research report is ultimately verified by the tests of the
underlying domain of scholarship. To put it somewhat simplistically, if the
work is mathematical, we check the theorems; if it is a model, we check to
see whether it fits empirical observations. Perhaps we should be guided by
Shakespeare: “What’s in a name? That which we call a rose by any other
name would smell as sweet.”
■

*•
SECTION 7
 THIRTY YEARS OF
INFORMATION THEORY

Myron Tribus

This paper is concerned with the impact of Shannon’s 1948 paper “A Mathe¬
matical Theory of Communication,” which was focused on the field of com¬
munications. [Shannon, 1948.] I especially wish to comment on the influence
of Shannon’s paper in fields other than communications. I approach the task
with some hesitation: In 1961 Shannon, in a private conversation, made it
quite clear to me that he considered applications of his work to problems
outside communication theory to be suspect and he did not attach funda¬
mental significance to them.
Despite Shannon’s misgivings, as time goes on, his great contribution to
the literature has made its existence felt in an ever-increasing number of
fields. Shannon’s clarification of the concepts of uncertainty and information
and the ability to give a quantitative measure to these concepts has served as
a powerful stimulant on the imagination of others. Of course, there have
been numerous unjustified excursions in the name of information theory, but
my studies show many solid accomplishments clearly inspired by this fa¬
mous paper.
To test the spread of Shannon’s influence, I thumbed through the En¬
gineering Index and found that the first entry under the heading Information
Theory occurred in 1951, only three years after his paper appeared. In 1952,
there were eight entries; in 1953, there were thirteen entries, including three
symposia and a reference to a bibliography containing over 1000 references
to information theory. By 1958, ten years after Shannon’s paper, the En¬
gineering Index contained several pages of titles covering papers in the field
of information. Many of the entries were references to symposia that them¬
selves contained between fifteen and twenty papers on information theory. I
can report also that the Engineering Index listing under the title Information
Theory is incomplete. For example, in 1961, I prepared a paper on informa-

This is a slightly modified version of a paper first published as an article in Levine, Raphael D.,
and Tribus, Myron, eds., The Maximum Entropy Formalism (Cambridge, Mass.: MIT Press,
1979).

475
476 MYRON TRIBUS

tion theory and thermodynamics. It was listed by the Index under Thermody¬
namics but not under Information Theory. There is, therefore, no easy way
to track down all of Shannon’s influence. For example, under the heading
Entropy, the Engineering Index for 1961 merely says “See Thermodynam¬
ics." Once we know the field that was influenced, however, it is fairly easy
to document the influence.
It is rather difficult to say where the field of information theory ends and
other fields begin. Certainly, information theory overlaps significantly with
many other fields. Define S — entropy, K = constant, pt = probability of the
truth of assertion i. For the purposes of this paper, I shall take the view that
any field of inquiry that uses the function

(1)

(or its continuous analog) in one of the following three ways is using informa¬
tion theory: (1) as a criterion for the choice of probability distributions; (2) to
determine the degree of uncertainty about a proposition; and (3) as a mea¬
sure of the rate of information acquisition.
Claude Shannon was not the first to use the function - X,-p,- In ph for this
function (or its continuous analog) had been used as long ago as 1872 by
Boltzmann. [Boltzmann, 1872.] What Shannon did was to give a universal
meaning to the function - In p; and thereby make it possible to find other
applications. Warren Weaver, in his popularization of Shannon’s work,
foresaw the widespread influence it was bound to have. Weaver understood
that any paper that clarified our understanding of knowledge was certain to
affect all fields that deal with knowledge.
In the 1961 interview with Shannon to which I referred, I obtained an
anecdote worth recording here. I had asked Shannon what his personal
reaction had been when he realized he had identified a measure of uncer¬
tainty. Shannon said that he had been puzzled and wondered what to call his
function. Information seemed to him to be a good candidate as a name, but it
was already badly overworked. Shannon said he sought the advice of John
von Neumann, whose response was direct, “You should call it ‘entropy’ and
for two reasons: First, the function is already in use in thermodynamics
under that name; second, and more importantly, most people don’t know
what entropy really is, and if you use the word ‘entropy’ in an argument you
will win every time!’’

THERMODYNAMICS

A field in which Shannon’s interpretation of entropy has had a profound

effect is classical thermodynamics. [Tribus, 1958.] In this field, the entropy
function has a long and involved history. Clausius coined the word entropy
THIRTY YEARS OF INFORMATION THEORY 477

from the Greek language to mean transformation and defined it via the
equation

dS = -42-, (2)

where dQ - increment of energy added to a body as heat during a revers¬

ible process;
T = absolute temperature during the reversible heat addition;
dS = change in entropy.

The function as Clausius introduced it was, at the time, quite mysterious.

It was an extensive property, like mass, energy, volume, momentum, charge
and number of atoms of the chemical species. Unlike these quantities, how¬
ever, entropy did not obey a conservation law. Instead, it showed a ten¬
dency to increase spontaneously. At a time so profoundly influenced by the
success of the various conservation laws in physics, when determinism was
at its peak, a measure that always increased was indeed mysterious and
called for an explanation.
Among scientists and engineers, there are those who speak of discovering
the laws of nature, and then there are those who speak of inventing these
laws. The former are more numerous; they include the majority of teachers
and, therefore, students. The chief difference between the two types lies in
differing interpretations of what constitutes an explanation. The former
would require that all mysteries be explained in terms of things already
known or new axioms that square with the old. The latter are ready to
rewrite and reorganize all human knowledge in their quest for consistency.
To the latter, there are no explanations—merely consistent ways of tracing
ideas to their common logical sources.
I have dwelt on the word explanation because the arguments that have
raged in thermodynamics hinge greatly on this point. The information-theory
approach raises a question in the following way. Shannon’s measure gives,
by ordinary differentiation,

dS = - K ^ (In pi + 1) dpt. (3)

i
But since X,p, = 1, we have X,dfp,- = 0, and, therefore,

dS = - AT ^ In Pi dp^ (4)
i

On the other hand, from classical thermodynamics, we have

dS = -42L. (5)

How then are we to explain these two equations for dS?

478 MYRON TRIBUS

In 1953, Brillouin formally stated the case for a close connection between
these two entropies. [Brillouin, 1953.] He followed his early paper with
several others and with his famous book on Science and Information
Theory. [Brillouin, 1962.] In these publications, Brillouin showed there was
an intimate connection between entropy and information, but he did not
think it necessary to define one in terms of the other. Brillouin treated the
field of thermodynamics as correct and preexisting, and he treated Shan¬
non’s information theory as a correct theory. He then proceeded to demon¬
strate a consistency between the two entropies.
As early as 1911, Van der Waals had proposed that there ought to be a
connection between Bayes’s equation in probability theory and the second
law of thermodynamics. [Van der Waals, 1911.] As we now know, Shan¬
non’s measure can be derived through the use of Bayes’s equation. [Tribus,
Evans, and Crellin, 1964; Good, 1950.] In 1930, G. N. Lewis, in a discussion
of irreversibility, had written “Gain in entropy means loss of information—
nothing more.” [Lewis, 1930.] As I mentioned earlier, Boltzmann had used
the //-function as early as 1872.
In 1938, Slater based his book on the definition

(6)

choosing this expression for entropy over the more conventional one,

S = k\n W, (7)

which is used by most authors in statistical thermodynamics if, = fraction in

microstate i; W = number of ways to realize a given macrostate). [Slater,
1938.]
With these historical antecedents, Brillouin evidently felt no need to take
one concept as more primitive and felt no need to explain the others in terms
of it. But he pointed the way.
It was E. T. Jaynes who first saw the complete answer and presented it in
a sequence of brilliant and historical papers. [Jaynes, 1957a, b.] As shown in
Equation (1), the entropy is defined by the probability. Jaynes turned it
around: He proposed that Shannon’s measure of uncertainty (entropy) be
used to define values for probabilities. Prior to Jaynes’s contribution, work¬
ers in statistical mechanics were forced to rely on classical thermodynamics
for their ideas about entropy and develop statistical mechanics as “an
analog.” [Gibbs, 1948; Denbigh, 1955.] There had been no clear and un¬
equivocal reason to define entropy in statistical mechanics other than the
after-the-fact conclusion that since statistical mechanics works, it must be
all right. Jaynes showed that by starting with the extremum of S [via Equa¬
tion (1)] statistical mechanics was derived in a few lines. In 1961, after
careful study of Jaynes’s paper, I demonstrated that all of the laws of classi¬
cal thermodynamics, and, in particular, the concepts of heat and tempera¬
ture, could also be defined from Shannon’s entropy using Jaynes’s principle
THIRTY YEARS OF INFORMATION THEORY 479

of maximum entropy. [Tribus, 1961a, b..] The debates engendered by this

approach have been extensive and, on occasion, bitter. I have concluded
from them that thermodynamics is as much a branch of theology as it is a
branch of science! It is very comforting to be able to show that although
Gibbs did not explain how or why he decided in 1901 to set the logarithm of
the probability of a system linearly proportional to the energy and numbers
of particles, thereby producing his “statistical analogs,” he did say that the
expected value of In p is a minimum; that is, - Jp In p dv is a maximum.
[Gibbs, 1902; 1948.] So we may conclude that the Gibbs and Jaynes formal¬
isms are essentially the same.
Using the information measure as a primitive idea, my colleague Robert
Evans has been able to demonstrate that thermodynamic information is
proportional to what has been called availability in thermodynamics. [R. B.
Evans, 1966.] Actually, Evans has shown that a new function, which he calls
“essergy” (and which is related to the function used in Germany under the
name Exergie), is at once a measure of information and a measure of the
potential work of systems. We have used the essergy function in the optimi¬
zation of such thermal systems as sea-water-demineralization plants and
steam-turbine generators. [El Sayed, 1968; El Sayed and Aplenc, 1968; El
Sayed and Evans, 1968.] The original impetus for these developments came
from Shannon’s work.

STATISTICS

Jaynes has converted Shannon’s measure to a powerful instrument for

generating statistical hypotheses, and he has applied it as a tool in statistical
inference. The mathematical aspects of Jaynes’s use of Shannon’s measure
are straightforward. If the given information is in the form

(8)

r = 1, 2, ... m (9)

the minimally prejudiced (i.e., least presumptive) probability distribution is

the set of p, that obeys the (m + 1) Equations (8) and (9) and maximizes

(10)

The resulting distribution is

Pi = exp [-\0 - XigiU/) - k2g2(*/) • • • (11)

where the X’s are Lagrangian multipliers. The entropy is
m

(12)
r= 1
480 MYRON TRIBUS

And the potential function, X0, is

X0 = In V exp [-^ \rg,(Xi)]. (13)

The Lagrange multipliers are related to the given data by the equation

dXq
<Sr> • (14)
dXr

The higher moments are given by

d2X0 d2X0
var (gr) covar (grgs). (15)
dX dXrdXs

For other relations, see Jaynes or Tribus. [Jaynes, 19576; Tribus, 1969.]
Jaynes’s principle shows that if Shannon’s measure is taken to be the mea¬
sure of uncertainty, not just a measure, the formal results of statistical-
mechanical reasoning can be carried over to other fields.
Jaynes proposed, therefore, that in problems of statistical inference
(which means problems for which the given data are inadequate for a deter¬
ministic prediction of what will [or did] happen), the probabilities should be
assigned so as to maximize

S = -K^ Pi \nPi (16)

subject to

X P> = 1 0?)
i
and any other given information. Jaynes thus essentially defines probabili¬
ties via Shannon’s measure. From this basis, we can see that any field that
uses probability theory is a candidate to be influenced by Shannon’s work.
I must confess that in 1957 I felt there was no explanation to be had and
complained publicly that Shannon had confused things by calling two differ¬
ent ideas by the same name! I was forced to retract that ill-considered
judgment in a footnote to my paper “Thermodynamics: A Survey of the
Field” and wish now that I had retracted the whole paper! [Tribus, 1958.]

THERMODYNAMICS (AGAIN)

An explanation of the two entropies can be had only if we clarify our ideas
about what we mean by (1) heat dQ, (2) temperature T, and (3) probability p,.
What Shannon’s work did, as interpreted by Jaynes, was to make it
possible to take the information-based entropy as the primitive concept and
explain dQ, T, and p, in terms of S.
Jaynes’s initial publication demonstrated that this approach permitted a
derivation of statistical mechanics. He later went on to give illustrations of
THIRTY YEARS OF INFORMATION THEORY 481

the use of this principle in decision theory [Jaynes, 1963], communication

theory [Jaynes, 1957a, b], and transport theory [Jaynes and Scalapino,
1963].

RELIABILITY

It was also straightforward to use the same approach in reliability engineer¬

ing. [Tribus, 1962.] In this case, the maximum-entropy principle was used
with Bayes’s equation to show how to form an initial estimate of reliability
and then incorporate field-test and laboratory data as they become available.

STATISTICS (AGAIN)

Shannon’s measure occurs in statistics, but unless we are attuned to its

meaning, the significance of Shannon’s function in any one case is apt to go
unnoticed. For example, in hypothesis-testing, I demonstrated that the em¬
pirical fitting of experimental data to a probability distribution is expedited
and rendered more accurate via maximum-entropy methods. [Tribus, 1969,
chap. 7.]
I have also demonstrated that in the analysis of the contingency tables the
first approximation to an exact Bayesian test for statistical dependence be¬
tween attributes A, and Bj is given by the measure

AS* = Z/ff1"*. <18>

i j ‘ J

which simply tells us to compare the sum of entropies in the margins of the
contingency table with the entropy of the center and see where the informa¬
tion resides. It is especially satisfying to be able to report that the first-order
approximation to the right-hand side of Equation (18) is the familiar chi-
square statistic. [Tribus, 1969, chap. 6.]

PSYCHOLOGY

Once the measure of uncertainty had been established, psychologists were

quick to recognize its utility. In 1949, only a year after Shannon’s article
appeared, Miller and Frick published an article illustrating the relevance of
the theory to psychology. [G. A. Miller and Frick, 1949.]
The first class of psychological problems to attract attention quite natu¬
rally centered on the human being as a communicator or receiver of informa¬
tion. Shannon had already given some impetus to the field by his analysis of
the information content of symbols in the English language. Shannon s
clever techniques provided brand-new tools for psychologists already at¬
tracted to engineering literature by Wiener’s provocative publications on
482 MYRON TRIBUS

cybernetics, which had just appeared. Indeed, the appearance of popular

treatments of feedback theory, decision theory, game theory, and informa¬
tion theory has had a profound effect on all of science. The impact on society
of the computer and these theories was not only in the hardware and result¬
ing increase in human ability to control and compute. They also had an effect
on the way people thought about things. We all appreciate how looking back
at the earth from space influenced our ability to think of “spaceship earth.”
But the introduction of the concepts of feedback and signal-to-noise ratio
also influenced the way we thought about what we were doing. Physicians
began to understand that the degree of feedback could influence our nervous
systems. Psychologists in education began to see examinations as a form of
feedback. The impact of the computer goes beyond its ability to do sums. No
one who has ever written a computer program, especially one with branches
in it, can fail to appreciate the idea of an algorithm. Such a person is likely to
begin to apply it to, say, social decision-processes or political situations. A
decision-tree is not only a way of portraying a method to describe uncer¬
tainty. It can also become normative and influence how people think about
uncertainty and public risk-taking. Taken all together, these theories
ushered in the age of the knowledge worker and the information revolution.
By 1954, William McGill was using differences in entropy between the
margins and the center of a contingency table to measure a subject’s re¬
sponse to stimuli. [McGill, 1954.] This work was unknown to me until I
began to prepare for this assignment. It is interesting to note that we can
derive Shannon’s measure from a Bayesian analysis of contingency tables
and then show the chi-square measure as its first approximation. [Tribus,
1969.] Fred Attneave uses the chi-square measure to see how significant the
entropy difference is! Attneave lists 87 primary references to information
theory in the field of psychology. [Attneave, 1959.] The range of topics
considered is most impressive. Included are such titles as “Information
Theory and Immediate Recall,” “Information in Absolute Judgments,” “In¬
formational Aspects of Visual Perception,” “Uncertainty and Conflict,”
“Model for Learning,” “Information from Dot and Matrix Patterns,” “Re¬
lation Between Error Variance and Information Transmitted in a Simple
Pointing Task,” and “Information in Auditory Displays.”
Irwin Pollack investigated the information transmission when nine differ¬
ent musical tones were used and found 2.67 bits with and 2.19 bits without
objective standard tones. [Pollack, 1953.] Attneave reports that Rogers
tested a concert master of a symphony orchestra having absolute pitch and
found he could transmit 5.5 bits!

NEUROLOGY

Measuring the ability of people and animals to transmit and receive informa¬
tion through various sensory channels is an obvious way of applying infor-
THIRTY YEARS OF INFORMATION THEORY 483

mation theory, and Shannon’s work could be taken over almost unchanged
for this purpose. George Miller has pointed out, however, that the concep¬
tual foundations of information theory have as much importance as the
numerical measure. They provide a method for organization and patterning.
Miller points out that concepts of gestalt psychology may be put to test via
information-theoretic concepts. [Miller, 1953.] The view that psychoneural
activity is the economical encoding of experience may also be attributed
directly to Shannon’s influence.
It may properly be said that the penetration of field X by field Y is com¬
plete when the concepts of field Y are used to prepare a textbook for stu¬
dents in field X. In this sense, the appearance of Wendell Garner’s book in
1962, explaining psychological concepts via information theory, was a mile¬
stone. [Garner, 1962.]

SOCIAL SCIENCE

An economist, Alfred Kuhn, has used concepts of uncertainty, feedback,

and decision analysis to produce a unified treatment of learning, motivation,
language, culture, personality, personal transactions, organization, social
systems, economics, public policy, and the political process. It is Kuhn’s
thesis that the concepts of engineering are to be subsumed as special cases in
the concepts of these social sciences. [A. Kuhn, 1963.]

OTHER FIELDS

New applications of information theory in turbulence theory are emerging

from the work of Silver and his colleague Tyldesley in Glasgow. [Tyldesley,
1962.]
John P. Dowds has developed his own method of applying the entropy
concept to an examination of data from oil fields. Based on such an analysis,
he purchased the rights to an abandoned oil well in Oklahoma, and, today,
this well, which he christened Rock Entropy #1, is a producer sufficiently
generous to support Mr. Dowds’s further researches into the uses of en¬
tropy. [Dowds, 1964.]
My colleague Alvin Converse has also applied the entropy concept in a
search technique, using entropy as a measure of the information obtained
from each sample point in the search for the maximum of a function in a
bounded interval. [Converse, 1967.]
The maximum-entropy principle has been used in land-use planning and
predicting travel between different communities. [A. G. Wilson, 1969.]
There seems to be a growing use of information theory in medicine. I have
searched, via machine techniques, for articles referring to Shannon’s 1948
article and Jaynes’s 1957 article and turned up papers on rheumatism, bacte-
484 MYRON TRIBUS

rial populations in the Beaufort sea [Kaneko, Atlas, and Krichevsky, 1977],
information-processing of schizophrenics [Wijesinghe, 1977; Neufeld, 1977],
molecular biology [Berger, 1977], diagnostic value of clinical tests, and the
nature of living systems [J. G. Miller, 1976].
The ideas have also spread over the globe. J. O. Sonuga in Lagos,
Nigeria, uses the entropy principle to analyze runoff. [Sonuga, 1976.] In the
Netherlands, Van Marlen and Dijkstra use information theory to select
peaks for retrieving mass spectra. [Van Marlen and Dijkstra, 1976.] At the
Free University of Berlin, Eckhorn and Popel apply information theory to
the visual system of a cat. [Eckhorn and Popel, 1974.] In Canada, Reilly and
Blau use entropy in building mathematical models of chemical reacting sys¬
tems [Reilly and Blau, 1974], as does Levine in Israel [Procaccia, Shimoni,
and Levine, 1976]. And I suppose we can say the subject has really arrived
when it is referenced by the U.S. Congress Office of Technology Assess¬
ment! [Coates, 1976.]
The future we can perceive but dimly. But surely Shannon’s paper de¬
serves to be read by future generations with the same sense of intellectual
adventure that we associate with the great works of the past.
 THE WIDER SCOPE OF
INFORMATION THEORY

Donald M. MacKay

Tributes to the genius of Claude Shannon are well merited, and as one of his
earliest admirers, I gladly echo those paid by Myron Tribus. My purpose
now, however, is to say something about those aspects of the concept of
information that Shannon’s theory (for good and clearly stated reasons)
chose to ignore. For it is, indeed, the case, as Shannon so often insisted, that
his theory did not define the concept of information at all. [Shannon, 1951, p.
219.]

INFORMATION AND ITS UNEXPECTEDNESS

What Shannon’s famous £p,- log (1/p,) defines is the mean statistical unex¬
pectedness of an item of information selected from a given ensemble; and
this measure is benignly agnostic as to the meaning of the items of informa¬
tion selected in a communicative process except insofar as it affects their
prior probabilities of selection.
For this reason, I am not so sanguine as Tribus in his paper that “Measur¬
ing the ability of people and animals to transmit and receive information
through various sensory channels is an obvious way of applying information
theory.” The basic problem, of course, is that the probabilities p,- in Shan¬
non’s formula presuppose a well-defined ensemble; but in the case of a
human subject, it is not at all obvious how we should define the operative
ensemble, let alone its relative frequencies. It is still less clear that what we
really want to understand about human information-processing is always
illuminated by appeal to the particular measure of unexpectedness that
Shannon developed for the assessment of communicative channel capacity.
To try to translate every reference to information (whether in biology or
elsewhere) into a statement about unexpectedness would be as inept, and as
conceptually Procrustean, as translating all references to a house into state¬
ments about its size.

485
486 DONALD M. MACKAY

THE SCOPE OF INFORMATION THEORY

Information theory, in the more general sense it has developed over the past
forty years, is concerned with all processes in which the spatio-temporal
form of one set of objects or events (at A) determines the form of another set
(at B) without explicit regard for the energetics involved. These are situa¬
tions in which we say that information flows from A to B. In this operational
context, then, we can define information as that which determines form, in
much the same way as force is defined in physics as that which produces
acceleration.
In all cases, information is operationally defined (like energy) by what it
does. [MacKay, 19546; 1969.] But whereas the work done by energy is
physical in character, the work done by information is logical work. In
talking about information, there is always a suppressed reference to a third
party, since, as in the physical theory of relativity, we have to relate our
definitions to an observer, actual or potential, before they become opera¬
tionally precise.
In communication engineering, the form of a received message can be
determined in one of two ways: (1) by a process of construction, as when the
form of a television picture is built out of light-spots; or (2) by a process of
selection from a range of preconstructed forms in obedience to a code signal,
which has no necessary isomorphism with the form selected but merely
specifies its address, as in Morse code.
Shannon’s theory was concerned with processes of type (2), where the
prime question is how many (or how few) elementary code signals can be
sufficient (on average) to identify each form required to be selected. His
answer—brilliantly conceived and elaborated—was that the average num¬
ber of elementary signals required is proportional to the weighted mean
unexpectedness log (1 //?,); but that in “noisy” situations, reliability can
be increased by using more signals per selection and designing a redundant
code to allow errors to be corrected automatically. One of Shannon’s great¬
est achievements was to show in a precise sense that a given noise level
could, in principle, be offset by introducing a corresponding degree of redun¬
dancy, so as to achieve practically error-free communication of sufficiently
long messages.
An ingenious analogue of this theorem, due to Winograd and Cowan, has
shown how structural redundancy in a computing network can be exploited
to achieve error-free computation with unreliable elements. [S. Winograd
and Cowan, 1963.] This has important implications for the theory of informa¬
tion-handling networks, including the central nervous system, where distrib¬
uted processing may be organized so as to optimize reliability.
Processes of type (1), in which forms are not selected but constructed, are
typical of scientific experimentation. For example, data are accumulated and
processed into the form of a graph, or light is transmitted through a micro¬
scope or telescope and forms an image on a photographic plate. In this
THE WIDER SCOPE OF INFORMATION THEORY 487

context, the possibility of information measurement had been recognized for

some time before the publication of Shannon’s papers. The English statis¬
tician Ronald A. Fisher described in 1935 an additive measure of the amount
of information in a statistical sample, which reduced itself in the case of a
Gaussian distribution to the invariance 1/ct2, and quantified in effect the
sharpness of estimation. [Fisher, 1935, p. 188.] In 1946, Dennis Gabor, who
later received the Nobel Prize for his work on holography, defined a quite
different quantum of information in signal transmission, which he termed a
“logon.” The number of logons in a signal represented the number of de¬
grees of freedom of its structure—the minimum number of independent
measurements mathematically necessary to define its form under given lim¬
iting conditions of frequency bandwidth and duration. [Gabor, 1946.]
About the same time, I was independently groping after some abstract
way of quantifying the information derived from a physical experiment. Two
distinct measures of the process of form-construction emerged at an early
stage. [MacKay, 1950; 1951; 1954b.] The first, analogous to Gabor’s logon-
content, has to do with the number of degrees of freedom of the constructive
process: For example, the number of practically discriminable elementary
areas within the field of view of a microscope. This is an a priori measure,
which in an optical instrument is related to the notion of resolving power and
defines the complexity or logical dimensionality of the framework within
which construction of the form is to take place.
The second, analogous to Fisher’s measure of amount of information, has
to do with the weight of evidence associated with each degree of freedom of
a constructive process. This is an a posteriori measure, closely related to the
concept of signal-to-noise (power) ratio, and illustrated in the case of a
photographic image by the density of the silver grains. I gave it the name of
metron-content.
The details of this formalism do not matter here. The point is that for
different purposes, the question “How much information?” may need to be
answered in quite different ways, according to the quantifiable aspects of the
process of form determination that are relevant. As a typical example, con¬
sider the recording of electrical signals from the scalp of a human subject—
the electroencephalogram (EEG). A band-limited amplifier gives an output
with a finite number of degrees of freedom (1 + 2rA/), where t is the
duration and A/the bandwidth of the amplifier. This is the logon-content of
the signal. If signal-to-noise ratio is poor, the operator may deliberately
narrow the bandwidth, reducing the logon-content for the sake of a greater
signal-to-noise (power) ratio per degree of freedom, or metron-content. The
latter measures the weight of physical evidence contributed as to the mag¬
nitude of the signal, and its total over a signal is invariant under such pro¬
cesses as averaging.
Although logon-content and metron-content are quantities that can be
bartered for one another in the foregoing sense, each is clearly a measure of
a different and important aspect of the ability of a signal to determine form.
488 DONALD M. MACKAY

Furthermore, in a generalized sense of the term, we can speak also of the

logon-content of the spatial distribution of EEG activity over the scalp. The
important informational question here concerns the number of substantially
independent spatial areas of the scalp (from an electrical standpoint). This
logon-content determines the maximum number of electrode-placements
necessary and sufficient to record all the structural information present in
the distribution of scalp potential.
By contrast with the foregoing measures, the familiar Shannon-Wiener
measure of amount of information (for which selective information content,
SIC, is a useful distinctive term) would fasten on the mathematical improba¬
bility of the signal and ignore its structure except insofar as it bears on its
probability.
It would clearly be absurd to regard these various measures of amount of
information as rivals. They are no more rivals than are length, area, and
volume as measures of size. By the same token, it would be manifestly inept
to take any of them as definitions of the concept of information itself.
At the time when Shannon’s paper [Shannon, 1948] and Wiener’s book
[Wiener, 1948 and 1961] appeared, then, some progress had already been
made with the quantification of processes of form-construction, to which
Shannon’s quantification of the statistics of form-selection was valuably
complementary. By 1949, a group of us in England had set about organizing
the First International Symposium on Information Theory, which was held
in the rooms of the Royal Society in London in 1950. Shannon was a leading
participant, as was Gabor; and Fisher chaired one of the main sessions. The
need for complementary measures of information-content was illustrated in
several papers, and the scope of the new concepts was explored in such
diverse fields as psychology and optics. [Proceedings, 1950.]

INFORMATION THEORY AND THERMODYNAMICS

A major topic of discussion at the 1950 meeting was the relation between
Shannon’s measure and the thermodynamic concept of entropy, first in¬
troduced into the theory of heat engines in the nineteenth century as a
measure of the unavailability of thermal energy. The expression, derived
from the statistical theory of gas molecules as a measure of their nearness to
thermal equilibrium (and, hence, of their inability to work a heat engine),
contains 1 pj log (1/p,) as a factor. What it signifies here, as Tribus has
explained, is the amount of information (in the Shannon-Wiener sense of
selective information content), concerning the detailed microstate of the
molecules, that remains undeducible from the macrodescription available (in
terms of pressure, volume, temperature, etc.)—in short, the extent to which
the microstate is underspecified.
Now, every unit of missing information (in this sense) means that a corre¬
sponding amount of energy cannot be pinned down to do work. It turns out
THE WIDER SCOPE OF INFORMATION THEORY 489

that this amount of energy is proportional to the temperature, so we can say

that each unit of missing information costs us (i.e., deprives us of the use of)
one unit of energy per unit of temperature. The physicist’s name for energy
per unit of temperature is entropy, and the unit in which it is measured
(when logs are taken to base e) is called Boltzmann’s constant and written k.
Thus, the entropy of a mass of gas molecules is written

Unfortunately, although it is not the mathematical factor X p,- log(l/p,)

but the unit k that here brings in Clausius’s notion of entropy, the habit has
grown of attaching the label entropy to the mathematical factor. It is rather
as if, having used the expression M x (XuA'/Xn,-) for the mean mass of a
population of objects, n, of which had mass a, x M, and so on, we were to
forget about the factor M and call the expression Xa/n//Xfl/ by the name
mass. If, then, we later found someone using the latter expression to calcu¬
late the mean income of a population, we can imagine the confusion that
would be caused by talking of his or her measure as mass. [MacKay, 1964b.]
But such, alas, is the state of our subject. If it is too late to clear up the
muddle by dropping the use of entropy for the information measure, the best
we can do may be to call the expression k X p, log( 1/p,) physical entropy, and
use some such term as mathematical entropy to distinguish the dimen¬
sionless expression X p, log(l/p,). To call the latter by some simple and
descriptive name such as mean unexpectedness or statistical variety would,
however, seem a better way of removing the recurrent confusion. It would
have the further merit of reminding us of the essentially relative character of
the mathematical measure, which clearly depends for its magnitude on the
particular body of evidence (or if we like, the particular observer) by which
the probabilities p, are determined. Any convention that helps avoid the
notion that information is a kind of absolutely measurable “stuff’ (like
water) might be worth a little trouble to secure.
Curiously enough, thermodynamic entropy also turns up in a key role in
calculating the metrical information content (metron-content) of a physical
measurement. The ultimate limit to the accuracy of determination of a signal
amplitude, for example, is the thermal energy per degree of freedom (i.e.,
per logon), which is classically \kT (k = Boltzmann’s constant; T = abso¬
lute temperature). This provides a natural scale-unit in terms of which to
quantify signal energy per logon. For a unit signal (1 metron) on this scale,
the physical-entropy increase is thus AX = \kT!T or \k. In general, then, it is
the limiting metron-content that is proportional to the increase in physical
entropy brought about by a received signal. If the signaling system uses all
discriminable signal levels with equal frequency, then the average selective
information content per logon [Shannon’s X p, log( 1/p,j] will be proportional
only to the logarithm of the average physical-entropy increase. There is
nothing paradoxical about this, once we remember that the relative probabil-
490 DONALD M. MACKAY

ities pi are defined by quite different considerations in the respective ensem¬

bles of the thermodynamic state-description and the communication en¬
gineer’s repertoire; but it underlines the importance of not identifying
Shannon’s measure uncritically with physical entropy.

INFORMATION THEORY IN BIOLOGY

The early 1950s saw a rush of would-be applications of Shannon’s measure

in biological contexts. A pioneering example was the work of Hick on the
influence of the selective information content of stimuli on reaction time,
where the significant feature of stimulation was not the physical specification
per se, but rather the number of alternative forms that it might have as¬
sumed but did not. [Hick, 1952.] The idea that the range of forms not taken
by the input might be relevant to its specification is a characteristic contribu¬
tion of information theory, and one that might not have been thought so
obvious without it.
Again, the concept of the redundancy of an input pattern as an experimen¬
tal parameter has suggested a class of stimuli (such as optical patterns of
near-parallel lines) that have revealed interesting new phenomena and may
throw light on the neural mechanisms that encode those features that are
redundant. [MacKay, 1957u, b; I960.]
At the other extreme, the use of random noise (acoustic or visual) to
provide a stimulus of minimal redundancy has helped to uncover another
class of phenomena rich in clues to neural processes. [MacKay, 1965.]
As Myron Tribus has mentioned, such concepts as economical encoding
seemed at first to permit straightforward transfer. With hindsight, however,
a snag becomes obvious. Concepts like economy or efficiency can be defined
only in relation to specific criteria of cost, and the problem is to determine
what counts as costly in a biological system. To an engineer, cost is normally
at least proportional to (or may even rise more rapidly than) the number of
elements in a network. In assessing a biological system that can multiply
cells at a relatively small genetic cost, however, it is obvious that our criter¬
ion of economy or efficiency may have to be drastically modified. Moreover,
since Shannon’s measure was designed for a context in which the engineer
must be able and willing to adjust his or her repertoire of code signals
according to the known statistics of the traffic, and install correspondingly
complex decoding processes ad hoc, it can here give a quite misleading idea
of efficiency unless the biological cost (and likelihood) of implementing these
processes is included in the calculation.
All this adds up to the suggestion that in the biological realm, theorems
and concepts of information engineering can seldom, if ever, be transferred
directly to make predictions. Their value, as we have seen earlier, lies in
their power to suggest new questions from a new angle and with a new
thrust. [MacKay, 1959; 1964b; 1965.] As always, it is from nature’s answer
THE WIDER SCOPE OF INFORMATION THEORY 491

to these questions, rather than from our armchair theorizing, that our knowl¬
edge grows. Equally clearly, it is the quality of our thinking in the armchair
that will determine whether the questions we ask receive a worthwhile
answer.

INFORMATION THEORY AND MEANING

As early as our 1950 Symposium on Information Theory, it was felt to be

somewhat scandalous that the theory of information seemed to have so little
working contact with such concepts as the meaning and relevance of infor¬
mation. Once the framework is widened from the simple source-channel-
sink model to include the sender and recipient of messages as goal-directed
agents, however, the links are not hard to find. [MacKay, 1954b; 1969.]
Briefly, the key is to think of the act of communication as goal-directed to
bring about certain states of affairs in the recipient. The recipient at any
given time has (1) a store of factual information, (2) a repertoire of skills, (3)
a hierarchy (or heterarchy) of criteria of evaluation and priorities. These
three interlock in a complex manner so as to set up at any given time a
certain total state of conditional readiness (SCR) for all possible action,
including planning and evaluating action. It is this SCR that can be thought
of as the operational target of a communication.
The meaning (intended, received, or conventional) of a particular com¬
munication can now be defined as its readying function (intended, actual, or
conventional) on the SCR of the recipient. Distinctions then emerge auto¬
matically between indicative, imperative, interrogative meaning, and so
forth, according to the particular aspects of the SCR that are the target in
each case. [MacKay, 1961b.] Functional criteria of meaningfulness (or
meaninglessness) follow in obvious ways, and the relative nature of all the
foregoing concepts, including that of information itself, becomes explicitly
obvious. By quantifying the degrees of freedom of the SCR in its various
aspects, it is in principle possible on these lines to attach numerical mea¬
sures to the transfer of information between cognitive agents, in essentially
the same way as in an inanimate context. [Ibid.]

CONCLUSION

It is undeniable that much of the popularity of the information-theoretical

approach, especially in the 1950s, can be attributed to the seminal work of
Claude Shannon and Norbert Wiener on the quantification of uncertainty. It
may also be true that some of the later disillusionment with that approach,
especially among psychologists, stems from an initial lack of public clarity
about the scope and limits of Shannon’s theory (despite his best efforts) and
especially from a widespread failure to distinguish the concept of informa-
492 DONALD M. MACKAY

tion per se from various measures of its amount. It should be added that
some of the most generally useful qualitative concepts of information en¬
gineering cited by Tribus, such as feedback and feedforward, have quite
independent origins and owe nothing to the mathematical theory of informa¬
tion-measurement. [MacKay, 1954 b.]
This said, however, I agree with Tribus in expecting information-
theoretical concepts, both qualitative and quantitative, to play an increas¬
ingly fruitful role in the construction, testing, and (doubtless) demolition of
theories of brain function and kindred processes in a multitude of other
contexts.
 INFORMATION THEORY
IN PSYCHOLOGY

George A. Miller

Some enterprises are defeated by success. I like to think that the application
of information theory to experimental psychology illustrates this oxymoron.
It is true that Shannon’s measure of information—the mean value of the
logarithmic probability [see Tribus’s Equation (1)]—is seldom seen these
days in psychological publications. It enjoyed a vogue in the 1950s, became
widely known among experimental psychologists in the early 1960s, and
disappeared by the 1970s. Students, noting this history, usually assume that
it disappeared either because it was proven false or, worse, because the
intellectual position it represents became uninhabited. Neither assumption is
correct.
What students cannot appreciate is what life was like before information
theory. Perhaps an example will help. I recall a series of experiments in¬
tended to determine the absolute threshold for the perception of visual form.
Alternative shapes were displayed in near darkness, and illumination was
gradually increased until viewers were able to identify the shape correctly.
Different experimenters obtained wildly different results. Today, it is almost
inconceivable that such experiments could have been run without control¬
ling the viewer’s expectations of what shapes might occur and without in¬
structing the viewer on the acceptable set of alternative responses. No won¬
der the results were all over the lot!
The point is that before anyone tried to apply information theory to such
experiments, there had been no reason to think of the viewer as a channel or
of stimuli as inputs and responses as outputs. If you wanted to estimate the
amount of information transmitted by such a channel, however, it im¬
mediately became apparent that you had to form a matrix with the stimuli as
rows, the responses as columns, and the cells as the frequencies of occur¬
rence of particular stimulus/response combinations. How many rows? How
many columns? In order to estimate the probabilities that were required in
order to compute the transmitted information, answers to such questions
had to be fixed and known. One consequence of trying to apply information
theory to psychology, therefore, was a principled increase in the rigor of
psychological experimentation.
493
494 GEORGE A. MILLER

No one today wants to give up these conceptual and methodological

advances. The notion that people are channels through which information
flows into storage or behavior is today a familiar part of most introductory
courses in psychology.
So, why has information measurement disappeared? For various reasons.
For one thing, the basic ideas of information measurement are now part of
the general culture. Everyone learns about bits and bytes as part of learning
about computers, which are everywhere. Explanations in psychological
journals or texts are no longer needed. For another, it has turned out that the
probabilities themselves are psychologically more interesting than their
logarithms. Mathematical psychology has increased greatly in sophistication
since 1950, and interest now is less in measuring channel capacities than in
characterizing the processes that limit them. Generally speaking, ideas first
introduced to experimental psychology through information theory have
now become foundational assumptions that everyone takes for granted, their
historical origins now irrelevant to their present role.
Psychological interest was not confined to transmitted information. Re¬
dundancy, the extent to which the future of a message can be predicted from
its past, was also important. Formally, redundancy is equivalent to transmit¬
ted information—both are best viewed as nonparametric measures of corre¬
lation; in one case, a correlation between the input and the output of a
channel; in the other case, a correlation between the history of a message
and its next symbol. Shannon illustrated his theory of the message source by
computing upper and lower bounds for the amount of information per letter
that is encoded in printed English messages, and the techniques he used
introduced psychologists to some new ways of thinking about the sequential
organization of behavior.
Shannon’s assumption that the sequential statistics of printed messages
can be regarded as the product of a stationary stochastic process—
specifically, a higher order Markov process—was peripheral to the more
fundamental ideas of information measurement, but it served to alert psy¬
chologists to theoretical possibilities they had not previously discovered.
Language was not the only place where psychologists were concerned with
the sequential aspects of behavior. Here, too, information theory fed the
growth of mathematical psychology.
This peripheral assumption soon came under attack, however, and the
impression some students have that information theory was somehow dis¬
proved may result from misinterpretations of that controversy. Chomsky
argued that any device that generates sequences of message elements must
be a grammar; he proved that Markov processes are not adequate grammars
for natural languages. Shannon, of course, had not proposed that his sto¬
chastic generator should replace grammar—it was (and still is) a useful
engineering approximation—and so he felt ill treated by implications that he
had somehow failed in doing something that he had never tried to do. In any
case, by 1960, the basic statistical parameters needed for that kind of ap-
INFORMATION THEORY IN PSYCHOLOGY 495

proximation of English were known, and further research into that question
was unnecessary. By 1960, therefore, psycholinguists had moved on to new
research on the psychological reality of various types of generative gram¬
mars.
* * *

According to Shannon’s measure of selective information, the less prob¬

able a message is, the more information it contains. This definition some¬
times conflicts with one’s notion of semantic information. Of the two sen¬
tences, Rex is a dog and Rex is a mammal, for example, the first is surely
more probable in normal, conversational English. So, according to Shannon,
Rex is a mammal must communicate more information than does Rex is a
dog, which is more probable and therefore less informative.
Stated so baldly, it is apparent that something about Shannon’s measure
is odd. To say Rex is a dog implies that Rex is a mammal, but to say Rex is a
mammal leaves it open whether Rex is a dog, cat, horse, elephant, or some
other mammal. Semantically, therefore, Rex is a dog must communicate
more information than does Rex is a mammal, which reverses the conclu¬
sion based on Shannon’s measure.
At the very least, therefore, it must be admitted that Shannon’s is not a
measure of semantic information. It is sometimes referred to as a measure of
selective information, since it has to do with selecting a transmitted message
out of the set of all other messages that might have been transmitted instead.
Could a measure of semantic information be developed? In 1952, Rudolf
Carnap and Yehoshua Bar-Hillel published a technical report from the Mas¬
sachusetts Institute of Technology Research Laboratory of Electronics, en¬
titled An Outline of a Theory of Semantic Information. In that report, they
dealt with semantic information only insofar as it applies to declarative
sentences, or statements. The general idea was that the content of a state¬
ment is given by the class of possible states of the universe that it excludes.
Rex is a dog excludes more possible states of the universe than does Rex is a
mammal and so has more content.
Detailed development of this idea is beyond the scope of this comment,
but some of the more obvious consequences can be stated. If statement 5,
logically implies statement S2, then the semantic information carried by
Si includes the semantic information carried by S2 and possibly more. If S|
logically implies S2 and S2 logically implies Si, then Si and S2 contain equal
amounts of semantic information. A tautology is logically implied by every
statement and so contains no semantic information. A contradiction logically
implies all other statements and so contains the maximum amount of seman¬
tic information. The amount of semantic information in the conjunction of
two statements is equal to or less than the amount in the first plus the amount
in the second, and so on.
An interesting question for a theory of semantic information is whether
there is any equivalent for the engineer’s concept of noise. For example, if a
496 GEORGE A. MILLER

statement can have more than one interpretation and if one meaning is
understood by the hearer and another is intended by the speaker, then there
is a kind of semantic noise in the communication even though the physical
signals might have been transmitted perfectly.
Semantic information never enjoyed the vogue among psychologists that
selective information once did. It is worth reviving here only for the perspec¬
tive it offers on Shannon’s theory.
 ENTROPY AND THE
MEASURE OF INFORMATION

Peter Elias

Information theory in Shannon’s original technical sense as a mathematical

theory of communication was derived in part from Shannon’s work on cryp¬
tography during World War II and in part from earlier work on communica¬
tions. The technical theory has had a considerable evolution since Shan¬
non’s early publications. [Shannon, 1948; 1949.] Key technical steps in that
evolution appear in the volume of collected papers edited by Slepian.
[Slepian, 1974.] The texts of Fano [1961], Gallager [1968], and McEliece
[1977] give successive views of the field. Current work is published in many
journals, but the greatest focus is in two, IEEE Transactions [1955] and
Problemy [1965],
Tribus cites Shannon’s feeling expressed in a 1961 conversation, that
applications of information theory outside of the theory of communications
are suspect and of no fundamental significance. I agree in part and will
explain more precisely just why many such applications seem suspect to me.
I feel that it is particularly important to get this point straight in the context
of the present study. It is necessary to go into a bit of technical detail in
order to make the more precise statement.
Entropy, the information measure that Shannon defined for communica¬
tions purposes and Tribus discusses in his paper, is one of many possible
measures of the width of a probability distribution. Any number of measures
of this sort can be constructed. By the excercise of sufficient ingenuity, one
can construct (as Shannon and others have) a set of axioms that any rea¬
sonable information measure must satisfy and then prove that entropy is the
measure that satisfies them all and is uniquely determined by the axioms up
to a multiplicative scale factor, the constant K in Tribus’s Equation (1),
which can be absorbed in the choice of a logarithmic base in that expression.
(It is usual to choose the logarithmic base 2, set K = 1 in Equation (1), and
call the unit of information in the resulting measure the bit [binary digit], the
information required to select one of two equiprobable messages.) However,
R. A. Fisher [1934] had come up earlier with a quite different measure of
information, which uniquely satisfied some requirements that seemed to
497
498 PETER ELIAS

Fisher to be natural to impose on an information measure to be used for

some statistical purposes. The real basis on which Shannon—and later
everyone else—concluded that entropy is the uniquely correct measure for
the purposes of communications is not the fact that it satisfies some axioms
but, rather, the fact that by adopting a new, intrinsically probabilistic model
of communication and making use of his measure, he could prove two re¬
sults that do not hold for other measures.
Shannon’s model of an information source is a random process that gener¬
ates an infinite sequence of messages. In the simplest case, which I assume
for the purpose of explanation, that process just chooses the next message
from a fixed set of messages, using a fixed probability distribution, indepen¬
dent of prior choices. His first result says, roughly, that if a source selects
messages from a probability distribution that has an entropy of H bits then,

1. There is a uniquely decodeable encoding of sequences of N source

symbols into sequences of binary digits that, for any e greater than 0
and all sufficiently large N, takes on the average no more than N(l +
e) binary symbols to represent the N source symbols.
2. There is no uniquely decodeable encoding of sequences of N source
symbols into sequences of binary digits that for any N takes on the
average less than NH binary symbols to represent N source symbols.

Shannon’s second result requires a channel model and some additional

definitions. Shannon’s model of a channel has an input alphabet and an
output alphabet and a conditional probability distribution that gives the
probability of receipt of each of the possible output symbols conditional on
the transmission of each possible input symbol. If the transmitter uses differ¬
ent input symbols with a probability distribution known to the receiver, then
the natural measure of how much information is transmitted per use of the
channel, on the average, is the average difference between the entropies of
the receiver’s estimate of the probability distributions on the transmitted
symbols before and after receipt of each symbol. That is, the average
amount of information the channel should be credited with sending from
transmitter to receiver is the average difference between two quantities: (1)
the total amount of information the receiver initially needs in order to find
the value of the next unknown transmitted symbol and (2) the amount of
information the receiver still needs after a noisy version of that symbol has
been received, in order to remove the uncertainty introduced by the noise in
the channel and inform the receiver exactly what symbol was in fact trans¬
mitted. If the channel is noiseless, then the second entropy—the sub¬
trahend—vanishes: On receipt of the output symbol, the receiver knows
exactly what was sent. But when the channel is noisy, receipt of a symbol
reduces the receiver’s uncertainty about what was sent on the average, but
not to zero.
This difference-of-entropies measure is called the mutual information be-
ENTROPY AND THE MEASURE OF INFORMATION 499

tween the transmitted and received symbols. Its value depends not only on
the channel but on the probabilities with which different input symbols to the
channel are used. If the transmitter chooses to send only one of the possible
input symbols all the time, then the mutual information will, in fact, be zero,
since the a priori and a posteriori distributions and entropies will be the
same. If, instead, the transmitter chooses input symbols from the probability
distribution that maximizes the mutual information, then that maximum
value is called the capacity of the channel and is usually denoted by C and
measured in bits per channel use.
Shannon’s second result can now be stated. It says, roughly, that if a
communications channel has capacity C bits per channel use (in the sense
just defined), then for any positive e,

1. It is possible to encode sequences of N binary digits into sequences of

at most N/(C - e) channel input symbols in such a fashion that for all
sufficiently large N, the receiver, knowing the encoding codebook,
can decode the noisy sequence of received output symbols into a
sequence of N binary digits that matches the original binary sequence
perfectly with probability greater than 1 — e.
2. If sequences of N binary digits are encoded into sequences of less
than NIC channel input symbols, then any decoding procedure used
by the receiver gives an output binary sequence that has probability
at most e of matching the original binary sequence perfectly.

The first of these results was revolutionary when it was discovered. Its
import is that lack of reliability in a channel does not limit the reliability of
the information that can be received over the channel: It limits only the rate
at which information of arbitrarily great reliability can be received. And
unlike the result of repeating each message an arbitrary number of times to
obtain arbitrarily reliable reception, which gives arbitrary reliability only at
zero rate, Shannon’s second result says that the rate at which information of
arbitrarily great reliability can be received is, in fact, as large as the max¬
imum rate at which unreliable information can be received.
Together, Shannon’s two results allow the solution, in principle, of any
problem of matching a message source to a communications channel, noisy
or noiseless, in two stages. First, take the message source with entropy H
and encode sequences of its messages into sequences of binary digits. Then,
encode each sequence of binary digits into a sequence of channel-input
symbols, and send it over the channel at the allowed rate, decoding the
received noisy sequence into the input binary-digit sequence with probabil¬
ity near 1 and then decoding the input binary-digit sequence correctly into
the original message sequence.
The net effect of these two results is to justify the use of a probabilistic
model of communications for systems in which coding and decoding of
messages is allowed and to demonstrate that in such a model, the entropy H
500 PETER ELIAS

has a unique role as the measure of information. It is the common de¬

nominator between different message sources and between sources and
channels. The entropy of a source determines how much in the way of
channel resources must be devoted to its transmission, and no other infor¬
mation measure can make that claim.
The reason for looking with caution at the use of entropy and the rest of
the formal apparatus of information theory in fields other than communica¬
tions should now be obvious. Anyone who wants to use the measure is
certainly free to do so, and some of those users—for example, Boltzmann—
antedate Shannon. However, the fact that entropy has been proved in a
meaningful sense to be the unique correct information measure for the pur¬
poses of communications does not prove that it is either the unique or a
correct measure to use in some other field in which no issue of encoding or
other changes in representation arises. The measure may, indeed, be useful
or even unique for another application, but that fact must be demonstrated in
the target field. The statement or implication that information theory has
proved that entropy is the proper measure to use is not legitimate.
Psychology provides two interesting examples. Psychologists have used
information-theoretic concepts in designing and analyzing experiments in
human communication that clearly fit within Shannon’s framework.
[Attneave, 1959.] They (and Tribus) have also made use of the mutual infor¬
mation measure previously mentioned (and other related information-
theoretic measures) to measure the degree of independence of two or more
variables—especially when the variables are not number-valued but
categorical, so that correlation coefficients cannot be computed. [McGill,
1954.] There is no case for uniqueness in that application. It is true that
mutual information vanishes when (and only when) there is independence
between the input and output but so do many other measures. Kendall gives
an example of the kind of comparative discussion of a variety of information
measures that is relevant for statistical purposes. [Kendall, 1973.]
The use of entropy in thermodynamics and statistical mechanics, as pre¬
sented for example by Tribus [1961], has its own justifications, which ante¬
date Shannon’s work. In the related mathematical field of ergodic theory,
entropy has been a subject of great recent interest, in part, because it is an
invariant under a very broad class of transformations—that is, under any
reversible encoding of one set of sequences into another. A recent book by
Nathaniel Martin and James England presents developments in ergodic
theory and other branches of mathematics that have evolved from Shan¬
non’s work and its extensions by Kolmogorov and Sinai. [Martin and En¬
gland, 1981.] The fact that it is so easy to find the distribution that maximizes
entropy subject to a set of constraints, as illustrated in the Tribus paper in
Equations (8)—(11), is a great mathematical convenience and in the absence
of other compelling reasons may itself be adequate justification for using this
solution to the philosophically difficult problem of choosing prior distribu-
ENTROPY AND THE MEASURE OF INFORMATION 501

tions in fields other than thermodynamics. In thermodynamics, there is (at

least in principle) an available independent experimental check by making
measurements of temperature and heat flow. It is necessary to be careful in
applying the technique of entropy maximization subject to constraints to the
choice of probability densities in the case of continuous random variables,
since the integral analog to the entropy expression takes a value that is not
invariant under scale or other variable changes and is not itself an informa¬
tion measure. The difference of entropies that is the mutual information
between two random variables is invariant under scale changes in the con¬
tinuous case, however, and careful treatments of the continuous case are
available; see Gallager [1968] and Pinsker [1964] for early careful versions.
In biology, Tribus lists some fairly recent papers; for historical interest, I
will add references to two early symposia. [Quastler, 1955; and Yockey,
1958.] Some of the biological work deals with questions of the genetic stor¬
age of information—its transcription and translation into protein structure—
and bit-counting seems highly relevant. Most of it, however, is of a statisti¬
cal character coupled much more loosely, if at all, to Shannon’s
communications model.
In computer science, there have been various applications of information-
theoretic ideas and answers to specific technical problems. In particular,
there has been work on the economical storage and retrieval of information
in computer memory by Peter Elias and Richard Flower, where bit-counting
arguments are very relevant. [Elias and Flower, 1975.] Information theory
has been used to prove the optimality of algorithms for one purpose or
another, in particular, for demonstrating that sorting algorithms use a
minimal number of comparisons. [Knuth, 1973, Section 5.3.] There has been
work of commercial importance in the use of error-correcting codes (which
arose in a communications context, although their later development has
been more algebraic than information-theoretic) in providing reliable
memories, both semiconductor main memories and magnetic tape and disk
backup memories. Elwyn Berlekamp edited a collection of leading papers in
this field and has recently summarized its current state. [Berlekamp, 1974;
1980.] There has also been some work of a different character of consider¬
able philosophical interest, which makes use of a quite different information
measure although it still gives bit-counting results. In this work (e.g., Sol-
omonoff [1964], Kolmogorov [1965], and Chaitin [1977]), unlike Shannon’s,
an information value (possibly infinite) is defined for each particular se¬
quence of symbols rather than being defined only on a probability distribu¬
tion. The information in such a sequence is defined, roughly, to be the length
(in bits) of the shortest Turing-machine program that can be used to generate
the sequence.
Neither Shannon’s measure nor the Solomonoff-Kolmogorov-Chaitin
measure (sometimes called algorithmic information) seems to be of direct
relevance for use in treating information as a commodity in economics. In
502 PETER ELIAS

both cases, for example, the amount of information contained in an edition

of a million copies is the same as the information contained in one copy, and
a bit of information that says that the price of a stock is going up rather than
down has the same measure—one bit—independent of the size of the price
change. It is possible, however, to consider channels where different input
symbols not only define different conditional probability distributions on the
received symbols but have different costs. [McEliece, 1977, chap. 2.]
 THE ENTROPY FUNCTION
IN COMPLEX SYSTEMS

Elliott W. Montroli

Myron Tribus has clearly surveyed the evolution and numerous applications
of information theory and maximum-entropy formalism. I elaborate on the
theme by first making a few historical remarks on the early appearance of the
expression

(1)
j

commonly called the entropy function. Then I display its appearance in

certain observations on several sociotechnical systems.

HISTORICAL REMARKS

The combination p log p was apparently first published in de Moivre’s The

Doctrine of Chances, where we can learn how to calculate the odds of
situations encountered in the popular eighteenth-century games of Bassette,
Pharaon, Quadrille, Raffles, Hazard, and Piquet, [de Moivre, 1756.] The
combinatorics relevant to those games were often expressed in terms of
binomial coefficients, thus motivating de Moivre to estimate n\ for large n.
On November 17, 1733, he distributed privately a few pages of his findings,
containing what is now called Stirling’s approximation. [Stirling, 1730.] Few
copies of the small pamphlet have survived, but its contents are reprinted in
the third edition of his book. James Stirling’s minor role in the enterprise is
mentioned, [de Moivre, 1756 and 1967, p. 244.]
De Moivre estimates log n\, the binomial coefficients, and the ratio of a
given term in the expansion of (1 + 1)" to the middle term in that expansion:

I have also found that the logarithm of the ratio, which the middle term of a
high power (n) has to any distance from it by an interval /, would be denoted by
a very near approximation, supposing m = \n, by the quantities (de Moivre,
1756, p. 244)
(m + / — z) log (m + / — 1) +
(m — / + |) log (m - / + 1) - 2m log m + log (m -f 1 )/m. (2a)
503
504 ELLIOTT W. MONTROLL

For large (m + /) and (m - l), this combination is approximately

(m + /) log im + I) + (m - l) log (m - /) — 2m log m

fm + l , (m + l\ m — l
- log - + log (2b)
m \ m j m

= m ipi log pi + p2 log p2), (2c)

with p\ = (m + [)/m and p2 — (m - l)/m.
Although the expression for log n\ became widely applied in the next
century, there seems to have been little interest in the combination in these
equations until the publication of Boltzmann’s famous paper “On the Rela¬
tion between the Second Law of Thermodynamics and Probability Theory
and the Laws of Thermal Equilibrium.” [Boltzmann, 1877.] There, the ex¬
pression corresponding to —S of the entropy function appears in
Boltzmann’s derivation of the canonical energy-distribution function of sta¬
tistical mechanics from an artificial model of kinetic-energy transfer by
molecular collision.
In Boltzmann’s first model, he postulated molecules to limit their kinetic
energy to the set 0, e, 2e, . . . , pe. On collision, energy might be redistrib¬
uted subject to energy conservation. Let vv,- be the number of molecules of
energy je. Then the probability of a given state distribution is
n!

p
fW
o

with Wy = n and total energy = L = Wj. (3)

o
Boltzman sought the distribution of the total energy into its various states
that maximized the state probability [minimized the denominator of the
product in Equation (3) subject to the other restrictions in Equation (3)]. On
taking logarithms of the product and applying Stirling’s approximation to the
logarithm of the factorials, the quantity he minimized was

M -1 Wj log Wj — n. (4)
Today, he would maximize — M, which is in the spirit of the entropy func¬
tion [Equation (1)]. The combination wj log Wj represents the analog of the
terms in de Moivre’s expression [Equation (2)] except that Boltzmann’s
terms followed from the multinomial coefficients in Equation (3), while de
Moivre’s came from binomial coefficients.
Boltzmann then let vv, = /(ye), chose e to be small, proceeded to the
continuum (e = 0) limit, let p —» °°, and minimized

M' - fix) log/(x) dx (or maximize -M') consistent with (5a)

Jo
>
n fix) dx and L = x fix) dx. (5b)
o Jo
THE ENTROPY FUNCTION IN COMPLEX SYSTEMS 505

By exploiting the method of Lagrange multipliers, he found f(x) = h exp -

hx. This result is equivalent to the statement that the function /(x) that
maximizes an entropy integral —M' under the restriction that the first mo¬
ment of/(x) is specified is a decaying exponential when 0 < x <
Then Boltzmann considered real molecular systems and observed that if/
were the Maxwell velocity distribution of perfect gas molecules that exist
with equal probability anywhere in a container of volume V, the quantity

n - f\ogfd\ d3x (6)

became proportional to the entropy of a perfect gas known from thermody¬
namics.
Notice that, if the variable x in Equation (5a) has the range (- °°,o°) and ct2
= f x2 f(x) dx, the second moment being specified, the various limits of
integration go from to °°, and we find that the maximization of — M' by
Boltzmann’s method yields the Gauss distribution

/(x) = c exp ( — x2)/2cr2. (7)

As quantum theory became a starting point for statistical mechanics (fol¬

lowing Planck, Einstein, Sommerfeld, etc.), one specified the energy states
of a system by Ej. In an assembly of systems in contact with each other, the
fraction of systems fj in state Ej became determined by maximizing the
entropy

S = -k'Yjfj log ft

with constraints ^ = 1 and E = Ejfj. (8)

The/) are then found to have the canonical distribution

fi=c exp ( ^)■

As a student in the 1930s, I was introduced to this simple attractive style by

John Slater’s text. [Slater, 1938.] Since Slater was a popular lecturer at the
Massachusetts Institute of Technology at that time, it must have become
widely known to MIT students.
It was Claude Shannon’s genius in elaborating the pioneering but more
primitive ideas of Ralph Hartley and Harry Nyquist that identified the en¬
tropy function as an ideal measure of information transferred in communica¬
tions systems. [Shannon, 1948.] Thus, through the strange manner in which
scientific thought evolves, we find that a simple mathematical construct that
appeared without special comment in de Moivre’s gambling manual became
a highly visible quantity in Boltzmann’s kinetic theory of gases and even¬
tually basic in the statistical-mechanical interpretation of the properties of
gaseous and condensed matter. Surprisingly, it also emerged as the domi¬
nant mathematical structure for the optimization of human communication
systems as they developed through the application of a century of engineer-
506 ELLIOTT W. MONTROLL

ing ingenuity. E. T. Jaynes was one of the first to attempt to unify informa¬
tion theory and statistical mechanics. [Jaynes, 1957a, b.]

THE ENTROPY FUNCTION IN SEVERAL SOCIOTECHNICAL SYSTEMS

Communication systems considered by Shannon were composed of mes¬

sage-input elements, transmission channels, and message-output elements.
Since the entropy function appeared in a natural way for the information-
transfer rate in such systems, we might seek the importance of the function
in other sociotechnical systems composed of three analogous components.

Entropy Function in a Traffic Stream

A natural example is a highway-transportation system with an input provi¬

sion for travelers and a road providing a channel for travel to exit points.
About twenty-five years ago, Robert Herman, Elliott Montroll, and col¬
leagues at the General Motors Research Center performed car-following
experiments and made numerous observations on flow on single-lane roads
(and multilane highways under high-density conditions that prevented weav¬
ing from one lane to another). [Chandler et al., 1958; Gazis et ah, 1959;
Herman et ah, 1959.] I now show that an entropy function evolves naturally
from the observed stimulus/response equation that describes car-following
in a platoon.
Let us consider a platoon of N cars identified as 1, 2, . . . , N flowing along
a long single-lane highway void of traffic signals. An equation found to
describe with remarkable accuracy the response of a follower (identified by
n + 1) to the behavior of a leader (identified by n) is

dvn+x{t + A) = fv„(Q - v„ + i(QJ

dt ° Un(0 - *n+l(0_'
Here, vn(t) is the speed of car n at time t, xn{t) is the location of the front end
of that car at time t, and A is the time lag between the stimulus provided by
the lead car and the response by the follower. The time lag A, which varies
from person to person, is about 1.5 sec. Equation (9) is a quantitative expres¬
sion of the fact that the driver of the follower car (n + 1) accelerates when
his or her relative speed is too slow and decelerates when it is too fast. When
the driver being followed is far ahead, the response is not so sensitive as
when close.
Integration of equation (9) is the first step in the derivation of an equation
of state for traffic, a relation between vehicular flow-rate and density in
single-lane traffic:

vn+l(t + A) - X0 log dn + i(t) = constant GO)

THE ENTROPY FUNCTION IN COMPLEX SYSTEMS 507

with

dn + l(t) = xn(t) - xn+l(t) (11)

being the space available per car at location of «th car at time t. The traffic
density at the location of car n, pn(t), is the reciprocal of the space per car: p„
= 1 ld„ = number of cars per unit-length. In a freely moving stable stream of
traffic vn{t + A) with A — 1.5 sec is approximately v„(t), and if pc. is bumper-
to-bumper close-packing density at which vn = 0, Equation (10) becomes

Vn(t) = -\0lOg-^-. (12)

Pc
The local-traffic-flow rate (dropping explicit dependence on time) is

Qn = P„vn = -Xopc l°g (-jp-)- (13)

Notice that 0 < p„/pc s 1 and that the dimensions of our variables might be
cars per hour for q, cars per mile for p, and miles per hour for v. By
averaging over N cars in a line of traffic, the mean flow rate is proportional to
an entropy function in the variables (p„/pc). [Montroll, 1981.] Thus,
N
-A
Pn
Q — X0pc (14)
=1 o0 ' '°8 o )/
Pc
N
1
= \0pc. log —-1— log X P» '°g P")' (15)
P N ~ n= 1 >
where p is the average traffic density and pn — p„/pN. While the variables p J
pc are not normalized, the set pn is, so that Xpn = 1. The flow rate is a
maximum at a given traffic density if all p„ are identical, so that p„ = p and pn
= VN. Then the term in the bracket in Equation (15) vanishes. If drivers
behave differently from each other (as of course they do), so that some p„
deviate from p, the entropy term in Equation (15) does not achieve its max¬
imum value and the term in the parenthesis is negative, yielding a reduction
in flow rate q.

Entropy in the Catalogs of Sears, Roebuck, and Company

Another important sociotechnical system is the merchandising system.

Goods flow into a warehouse or distribution center of a retailing firm, remain
temporarily as inventory, and finally flow to the consumer. A company’s
profit depends on the flow-through rate of goods and the pricing of the
goods. The analogy between merchandising flows and communication and
transportation flows suggests that the entropy function may appear in the
process.
A further consideration of this idea requires merchandising data. By good
508 ELLIOTT W. MONTROLL

fortune, the firm of Sears, Roebuck, and Company has provided a rich
legacy of information on this subject in its annual catalogs, which form a
magnificent database of Americana of the past 85 years.
The preparation of the catalogs was a major concern of Sears Roebuck.
Basically, each page was audited to produce its share of the profit. [Emmet
and Jeuck, 1950.] For example, in 1930 the goals set ranged from $5,000 to
$20,000 per page, depending on the responsible merchandising department.
Since the profit that year was $14,300,000 and the catalog ran 1000-1500
pages, the profit per page averaged about $10,000. Many pages reserved a
small space for the tentative introduction of new products. Favorable public
response motivated increased allocation the next year. With declining sales
of an item, its space allocation decreased; sometimes it even disappeared
from the catalog. Various department heads, anxious for raises and promo¬
tions, were very competitive in preparing pages that listed items that they
hoped would outsell those of their colleagues. Robert Herman and I found
the distribution function of prices by year listed in many of the catalogs.
[Herman and Montroll, 1977.] Since prices ranged from a few cents to hun¬
dreds of dollars, we “expanded” the scale of low-cost items and “con¬
tracted” that of higher priced ones by recording the data as the logarithm of
the price (to the base 2) log2f\ Of course, we were aware (as many before us
dating back to Daniel Bernoulli) that log P is psychologically a more impor¬
tant variable than the price itself because we are especially sensitive to
relative price changes (AP)/P — A log P.
Examining the price distribution from many catalogs indicates that, in a
given catalog, the distribution of log2E, (P, being the price of the /th item) is
very close to the normal distribution [Equation (7)]. Important distribution
moments are
N

log P = (log2P) = -L X logzT5,, (16)

•/v ; = i

and
N

Q°&pi ~ lo§2T*)2. (17)

•/v /= i

The findings for (log2P) and cr for eighteen years appear in Table 1. The
quantity a is the dispersion in the logarithm of prices relative to their mean
value. The larger the a, the broader the range of prices in the catalog.
The variation in (log2P) over the years reflects changes in cost of living
through the twentieth century. Catalog prices changed in two manners: (1)
by the change in price of an invariant item such as a clothespin, a 1910
specimen being indistinguishable from one in 1940; and (2) by the change in
the nature or quality of the item listed to reflect evolving technology and
varying public taste. Many interesting deductions follow from changes in
(log2E), but it is to the third column a of Table 1 that I wish to direct
THE ENTROPY FUNCTION IN COMPLEX SYSTEMS 509

Table 1. Standard Deviation of log2Pfrom Mean (log2P) for Various Years in

the Period 1900-1976

Year" (log2 P) a Year (log2 P) cr

1900 0.150 2.43 1939-40 0.627 2.62

1902 0.212 2.34 1946-47 0.532 2.15
1908 -0.228 2.29 1948-49 1.336 2.37
1916 -0.068 2.38 1951-52 1.785 2.34
1924-25 0.422 2.32 1962 2.403 2.24
1929-30 0.998 2.26 1972-73 3.030 2.27
1932-33 0.691 1.91 1973-74 3.322 2.05
1934-35 0.673 2.22 1974-75 3.870 2.12
1935-36 0.537 2.39 1975-76 4.060 2.03

“An entry identified by a single year corresponds to a spring-summer catalog; an entry

identified by a number such as 1924-25 corresponds to a winter catalog.

attention. We have been as much impressed by the existence of an almost

invariant statistical quantity—an “economic constant of the motion’’—for
the marketing operation. It is remarkable that for more than seventy-five
years, the dispersion a [defined by Equation (17)] has hardly changed. The
average value of cr is 2.26 with ((cr — cr)2)* = 0.17.
Having observed the constancy of a, let us construct a simple mathemat¬
ical model to explain it. Let us suppose that in a given year the zth item in the
catalog experiences an inflation factor a,, so that its price over the year
changes from Pt to a,P,. Also, suppose that a is the mean inflation rate, so
that a, = a + A a,-, where presumably A a, (with 2 A a, = 0) may be treated as
a small number. Then, to first order in (Aa,/a), log Pt for one year is trans¬
formed in the next to

log ajPi — log Pi + log a + (18)

Hence, to second order, from Equation (17), cr2 of one year transforms to
N

O’2 + Jj X (lo§2 Pi - l°g2 P)

i=l
Act,-
a
+ if
N !
1 = 1
Act,-
a
(19)

The simplest assumption that might be made is that all items inflate in the
same manner, so that Act, = 0 for all i. Then, cr2 would remain completely
constant over the years. Table 1 indicates that this is almost but not pre¬
cisely the case. If we were to assume that A a,- were independent of the price
P,-, then the middle term in Equation (19) would vanish and a2 would increase
every year, which was not observed. If, however, we assume that low-
priced items were inflated more rapidly than high-priced ones, Aa, would
be positive for small P, and negative for large P,, so that the middle term
510 ELLIOTT W. MONTROLL

in Equation (19) would be negative in order to cancel the contribution of

the third term (a positive one) in Equation (19). This type of inflation is
commonly experienced. Clearly, if a2 remains constant, so does its square
root, a.
It was observed in the text between Equations (6) and (7) that if a random
variable has a range (-00,00) and if its dispersion is given, then the entropy
function is maximized by the Gauss or normal distribution. We have also
observed that the dispersion a2 of the log2E; distribution in Sears Roebuck
catalogs has been essentially constant for seventy-five years. Since the dis¬
tribution function of log P, has been Gaussian, the entropy function associ¬
ated with that variable has been maximized. Hence, in their marketing wis¬
dom, Sears, Rosenwald, their staff, and their successors created a business
based on catalogs with goods priced so that year after year the price distribu¬
tion maximized the entropy function associated with log P,.
A broad log-normal price distribution represents a balance between highl¬
and low-priced items, so that there is something of interest to all potential
customers. Originally, Sears Roebuck attempted to make merchandise avail¬
able in a medium-sized town to the farmer. They so well succeeded that they
soon not only successfully competed with stores in such towns but also
those in large cities. Circa 1910, several of the great department stores,
including Macy’s, Wanamaker’s, and Filene’s, unsuccessfully entered the
catalog marketing field in competition with Sears. It has been observed that
for a given type of merchandise, Sears stocked models that were cheaper
and also more expensive (a broader distribution) than those listed in catalogs
of the more traditional department stores. [Emmet and Jeuck, 1950.]

CONCLUDING REMARKS

The magnificent mathematical construct, the entropy function defined by

Equation (1), made its first inconspicuous appearance in de Moivre’s sys¬
tematic investigation of developing a strategy for winning gambling games
with the aid of probabilistic analysis. It was explicitly introduced by
Boltzmann to furnish a probabilistic interpretation of the second law of
thermodynamics, a law based on the thermal behavior of materials. After
communications engineers devoted a century to inventing and improving
communication systems, Claude Shannon showed that their drive toward
optimization could be measured mathematically in terms of an entropy func¬
tion with probabilities measured appropriately in terms of frequencies of
coding symbols or in terms of Fourier components obtained by the Fourier
decomposition of continuous signals. I have shown here that motor-car
operators have developed a style of car-following in traffic platoons that
yields a vehicular throughput that is measured in terms of an entropy func¬
tion obtained from ratios of local-traffic density to close-packing traffic den¬
sities. Finally, one of the most carefully planned and financially rewarding
THE ENTROPY FUNCTION IN COMPLEX SYSTEMS 511

retailing operations in the United States in the first half of this century seems
to have evolved in a manner (unknown to the planners) that maximized an
entropy function in distributing the logarithm of the item prices in its catalog,
subject to dispersion in the square of the logarithm of the prices being
specified and kept constant.
Notice that the last two examples have nothing to do with information
theory or thermodynamic entropy. They seem to reflect the manner in which
a group of people, trying to optimize their productivity, have unwittingly
fallen into a behavior pattern that may be described by an appropriate en¬
tropy function. The diversity of examples of sociotechnical systems previ¬
ously listed suggests that an astute observer should be able to find further
examples of the entropy function in the complex systems generated by peo¬
ple who wish to optimize their productivity.
■
 ENTROPY, PROBABILITY,
AND COMMUNICATION

Myron Tribus

The main idea of my paper was that the entropy concept had been used not
only as Shannon originally intended but in novel ways unforseen by him.
Four distinct uses were detected in the literature:

1. To examine human communication processes within the nervous sys¬

tem as well as input/output processes of humans.
2. To consider how entropy assignments could be used to determine
values and meanings for probabilities. Shannon defined entropy in
terms of probabilities. Jaynes defined probabilities in terms of en¬
tropies, and this had led to new lines of inquiry in several fields not
associated with information theory.
3. Defining entropy in a way devoid of concern with meaning has made
it possible for problems of communication and information to be sep¬
arated from problems of meaning, making it possible to discuss them
either together or separately, with less confusion of concepts.
4. To define conditions for reproducibility in statistical physics.

The reviewers have added specific examples of some of these ideas and
because of their specialized expertise and experience, have given us special
insights. In each case, what makes Shannon’s contributions so significant is
the casting of old questions in a new and more interesting light.
I especially appreciate Donald MacKay’s remark, “every unit of missing
information . . . means that a corresponding amount of energy cannot be
pinned down to do work.” The essergy concept, derived in the paper, dem¬
onstrates just that. MacKay also brings out the importance of the constant
K, which in the case of essergy is set equal to A.T0, just as he seems to
suggest.
Although the editors have given me more space for a rejoinder, I do
believe that in view of the reviewers’ remarks, it is better to stop here and
not increase the ratio of words to ideas.

513
SECTION 8
 ON SYSTEM THEORY AND
ITS RELEVANCE TO PROBLEMS
IN INFORMATION SCIENCE

Hassan Mortazavian

My task is to offer explanatory remarks and present some questions, distinc¬

tions, and ideas that may be useful for a better understanding of system-
theoretic concepts, in particular as they relate to problems in information
science. My approach is entirely informal, although occasionally I shall have
to provide formal definitions and technical details, partly in footnotes. The
text should be intelligible to those without prior background in either system
theory or information science. Writing a paper of this type is not an easy
task. Perhaps one should not even attempt it, for there is no general agree¬
ment on the scope of either system theory or information science; and some
issues on which consensus seems to have been approached cannot be accu¬
rately presented without using mathematical language.

SETS AND RELATIONS

A set is a collection of objects—sense-objects or thought-objects. If a set

contains no objects, it is called empty; if it contains all objects of particular
reference, it is called the universal set. When the number of objects con¬
tained in a set is finite, the set is called a finite set. If we consider the set of all
human beings as the universal set of our particular concern, then the set of
all male humans is a subset of the original set, and obviously it is finite. It is
clear that a subset of a set is itself a set.
A relation defined on a set is nothing but a way of relating the elements or
objects of the set to one another. For example, considering the set of all
men, one may define a brotherhood relation and obtain a set of pairs, where
the elements of every pair are brothers. Note that the same man can be a

Research for this paper was carried out while the author was on the staff of the Institut National
de Recherche en Informatique et en Automatique, Paris, France, and he wishes to acknowledge
the many useful comments received from his colleagues at INRIA.

517
518 HASSAN MORTAZAVIAN

member of more than one pair simply because he may have more than one
brother. Imagine now that we relate men in terms of brotherhood, but we
also want to distinguish between older brothers and younger brothers, ex¬
cluding those who are twins. This amounts to first partitioning the elements
of the set into two subsets, one consisting of all those men who are another
man’s older brother and the other consisting of all those who are another
man’s younger brother. Note that the same man then can be a member of
both subsets, in that he may have both an older brother and a younger
brother, hence, being also both an older and a younger brother himself.
Therefore, membership in the two subsets need not be exclusive. After this
partitioning is done, we relate every element of one subset to every element
of the other subset with respect to the relation of brotherhood if such a
relation exists between the two elements of concern. If the first subset
consists of older brothers and the second subset of younger brothers and we
choose the first element of every pair from the first subset and the second
element from the second subset, what we get is a larger set consisting of
what is referred to as a set of ordered pairs.
We need to define one more operation called the Cartesian product of two
sets and denoted by x —the customary sign for multiplication. By the Carte¬
sian product of two sets we mean a set consisting of all ordered pairs whose
elements are chosen from the first and the second set, respectively. To be
more precise, consider the two sets U = {father, mother} and Y = {son,
daughter}. Then the Cartesian product of U and Y is constructed as follows:

U x Y = {(father, son), (father, daughter), (mother, son),

(mother, daughter)}.

Note that the first element of every ordered pair of the product set belongs to
the first set and the second element of every ordered pair of the product set
belongs to the second; thus an ordering between elements is established.
A set S is said to be contained in a set T if every element belonging to S
also belongs to T, and then one writes S C T. Conversely, we may say that T
contains S.
These primitive concepts are sufficient for an elementary definition of the
concept of system, meaning a relation defined on a family of sets.
Consider a simple example adopted from Casti with some modifications.
[Casti, 1979.] Let U be a set of consumer items and Y a set of service
facilities. For example,

{bread, milk, stamps, shoes}

*1 *2 *3 *4
department
{market, store, bank, post office}
y\ y2 J3 >>4
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 519

Let us define a relation R on the Cartesian product U x Y by the rule:

x, is related to y\ in the sense of relation R if and only if good Xj may be

obtained at facility y;.

Thus,

R = {(*1, yi), (*2. y 1). Us, yfi, U4, Tb)}-

A convenient way to represent the relation R is by an incidence matrix:

R y1 Y2 Y3 Y4

X\ 1 0 0 0
*2 1 0 0 0
*3 0 0 0 1
x4 0 1 0 0

where if the relation R exists between xt and yh there is a 1 at the intersection

of the ith row and the ith column; otherwise, there is a 0.
Now, considering that R as a relation is contained in the Cartesian prod¬
uct of the two sets U and Y, that is, R C U x Y, the relation R may be
considered to define a system in the most abstract sense of the term. (This is
obviously not sufficient for characterizing most real-world systems, and a
more mathematical structure is required, but we need not go into any detail
concerning such technicalities.)
Now if we consider a decisionmaker who, after choosing the items he or
she would like to obtain, determines the facilities that should be used to
obtain every specific item chosen, then what we have previously described
may as well be considered this person's decision-making system. (This ele¬
mentary characterization of a system should not be generalized to more
complex situations, although the basic idea remains the same.)

THE CONCEPT OF SYSTEM

There is no general agreement among system theorists about how to define

this concept. The situation is not unlike the familiar one in mathematics
where there is no consensus among mathematicians about the definition of
such terms as number, set, point, or even infinity. We may argue, however,
that the absence of a generally accepted definition of number in mathematics
has not caused any significant delays in the progress of number theory or
mathematics in general. Although we may not all agree on how to define
number, for all practical purposes we all have a clear idea of what numbers
are.
520 HASSAN MORTAZAVIAN

As long as we have no generally accepted definition of system, it suffices

for all practical purposes to characterize and specify various instances of
systems constructively. Progress in system theory depends for the most part
on such constructive characterization. Before we say more about definition
and constructive characterization, we need to clarify a few methodological
issues.
It is clear that there is a difference between what is considered to be
known and what is considered to be defined. Not all known objects are
defined, and not all defined objects are necessarily known. By known ob¬
jects I mean objects of which we have at least one example at hand. Thus,
the number it as the ratio of the circumference to the diameter of a circle is a
defined mathematical object. Its numerical value, however, is not com¬
pletely known although it is known to certain degrees of approximation. On
the other hand, a point in Euclidean geometry is not defined, but it is known.
The definitions of line, plane, and volume depend on the primitive concept of
point. Systems are known, and particular instances of systems may indeed
be defined, but the very concept of system itself has not been defined to the
satisfaction of most, let alone all, theorists.
This is what Bertrand Russell had to say about attempting to define num¬
ber:

Many philosophers, when attempting to define number, are really setting to

work to define plurality, which is quite a different thing. Number is what is
characteristic of numbers, as man is what is characteristic of men. A plurality
is not an instance of number, but of some particular number. A trio of men, for
example, is an instance of the number 3, and the number 3 is an instance of
number; but the trio is not an instance of number. This point may seem elemen¬
tary and scarcely worth mentioning; yet it has proved too subtle for philoso¬
phers, with few exceptions. (Russell, 1919, p. 11.)

Russell’s remark is also relevant to our discussion of the definition of sys¬

tem. To see what system is, we have to study the characteristics of various
instances of systems.
According to Russell’s methodological position, a serious inquiry may
proceed in two opposite directions. One may be called the constructive
direction, toward gradually increasing complexity from deterministic sys¬
tems to stochastic systems; from time-invariant systems to time-varying
systems; from linear systems to nonlinear systems; and so on. The so-called
analytical direction, on the other hand, proceeds to greater and greater
abstraction and logical simplicity: For example, from concrete signal-flow
representations of electrical circuits to graph-theoretical and topological rep¬
resentations of electrical circuits; from matrix representations of determin¬
istic systems to algebraic representations of deterministic systems; and then
to categorical representation of such systems. (By categorical representation
of system, I mean the approach that uses category theory, a highly abstract
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 521

branch of mathematics.) The choice between the two directions is not dic¬
tated by subject matter, but by the inclination of the investigator, the pur¬
poses of the research, and the state of development of a particular type of
inquiry.
A definition of system is, as a rule, needed only when systems are pursued
in the analytical direction. For merely constructive characterization, specific
definitions of particular types of systems can be provided that suffice for all
practical purposes. We should distinguish among:

1. A concrete system (i.e., an existing system in the real world or one to

be constructed).
2. Various abstract (mathematical) systems or system-types.
3. The abstract (mathematical) concept of system.

By a concrete system, I mean any set of objects—sense-objects or thought-

objects—that interact among themselves or are interrelated. In this sense,
almost everything is a system of one kind or another. This broad notion of a
system, however, is not the one with which system theory is concerned.
Lack of clarity on this issue in the writings of some system theorists has
caused misunderstandings and confusion among outsiders. System theory is
not a theory of concrete systems in the sense just described. System theory
is the theory of (mathematical) models. Rudolf Kalman made this issue clear
when he wrote:

“Does system theory deal with the real world?’’ To this obvious question the
answer is no—that’s quite simple. System theory studies models and does not
accept responsibility for the accuracy or relevance of these models. That head¬
ache is reserved for the practitioners, and it should not be exaggerated. (Kal¬
man, 1980, p. 7.)

This statement means that system theory deals with certain universal
mathematical questions about (abstract or mathematical) models and as such
is independent of particular fields of application. It contributes to other
sciences by providing techniques for generating a class of models for any
well-defined system at hand, based on data resulting from experiment, ob¬
servation, and so forth. But it does not specify which model out of a class of
models is the model for a particular concrete system.
In this sense, system theory is a theory that develops mathematical mod¬
els for various abstract (mathematical) systems. Examples of such systems
or system-types are: linear versus nonlinear systems, continuous versus
discrete systems, deterministic versus stochastic systems, and so torth.
These abstract systems can properly be defined only by appropriate mathe¬
matical tools. Nevertheless, I shall try to offer nonmathematical illustra¬
tions—at the risk of some inaccuracy.
To illustrate what a linear system is, consider the trajectory of the flight of
522 HASSAN MORTAZAVIAN

an airplane. The plane is our system, and we assume that at every point of
time there are only two controls or influences exerted on it: the wind and the
control exerted by the pilot. Each one of these two inputs causes a particular
change in the trajectory of the flight—a particular output. Now, this system
is said to be linear if and only if the output corresponding to any linear
combination of these two inputs is a linear combination of the outputs corre¬
sponding to those inputs. Otherwise, the system is said to be nonlinear.
Roughly speaking, by a linear combination of two variables, we mean a
combination performed only by addition (subtraction) and multiplication by
a Scalar. Note that according to this definition, it is not true that a system is
nonlinear if it contains a nonlinear element. A linear system may indeed
contain nonlinear elements.
A system is said to be continuous if its variables change continuously over
time. A system is discrete if its variables change only at certain specific
points in time. A car traveling on a road is an example of a continuous
system; the number of people entering a shopping center is an example of a
discrete system.
In computing control schemes for real-world continuous-time systems, in
practice, we are forced to treat them, in some sense, as discrete-time sys¬
tems. There are essentially two reasons for this: (1) We can make only a
finite number of measurements of the output values in any time interval. (2)
In computing an appropriate sequence of inputs (controls or decisions) to be
applied to a system to control its behavior in some desired way, we generally
use a digital computer, which is a discrete-time system, and can generate
only distinct control signals at discrete times. The practical necessity of
dealing with continuous-time systems as if they were discrete-time systems
raises important questions about the treatment of sampled-data systems,
that is, systems that should be considered as continuous-time but with which
we can interact only at discrete times when samples of outputs are gathered
and a sequence of discrete controls based on the sampled outputs is applied.
Before providing definitions for deterministic and stochastic systems, I
ought to say a few words about determinism and causality. The word deter¬
mined has often been used in an ambiguous way. In one sense, a quantity—
say the output of a system—is determined when it is measured. In another
sense, an event—say the occurrence of a specific sequence of outputs—is
determined when it is caused. Although there are many reasons to believe in
some principle of causality, it is not altogether easy to define the notions of
cause and causal laws. It must be made clear, however, that what distin¬
guishes deterministic systems from stochastic systems is not the assumption
of causal relation between inputs and outputs or past and future behavior in
one system to the exclusion of the other. Both deterministic and stochastic
systems may or may not be considered as causal (although disbelief in cau¬
sality causes many conceptual difficulties). Deterministic systems are deter¬
mined, and stochastic systems are not determined, in the first sense, which
is independent of the notion of causality.
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 523

These considerations suggest the following definitions:

Deterministic systems are those in which exact knowledge of outputs is

ascertainable from exact knowledge of inputs. In other words, in deter¬
ministic systems, future behavior can be predicted with accuracy. Sto¬
chastic systems are those whose future behavior is not predictable.

Thus, roughly speaking, a system is stochastic if it exhibits no regularities in

its behavior. These irregularities may be inherent in the internal mechanisms
of the system; they may be caused by disturbances from the environment or
simply a result of inaccuracies in our measurements, and so forth; or they
may be a combination of these. In system theory, all such irregularities are
referred to as noise. The word noise in this sense covers a much wider range
of elements than in its electrical-engineering sense. In the present usage,

noise is equivalent to what cannot be accounted for or explained in the

behavior of a concrete system by the deterministic part of a model
adopted for that system.

Thus, for example, possible nonlinearities in a concrete system for which a

linear model has been constructed are simply considered as noise. This
point, though trivial, has proved too subtle for many practitioners.
At last we come to the concept of an abstract (mathematical) system.
Various concrete systems, such as cars, cells, or trees, are instances of
particular types of abstract (mathematical) systems, and these systems
themselves are instances of system as a general concept. System theory is
properly a theory of systems in this sense.
Mesarovic defines a system as an abstract relation defined on a set of
objects. [Mesarovic, 1964a; 1970; 1975.] The objects can be anything. A
system in this sense is a general notion, defined independently of any partic¬
ular set of objects. The justification for the adoption of such a definition lies
in the universal existence of objects and relations, and thus the definition is
general enough to represent all sorts of systems. Starting from such a general
notion of system, we proceed by assuming more structure for the objects
and investigate the properties induced by various types of relations.
To clarify this definition, let us recall the example of obtaining certain
consumer goods from different service facilities, given in the first part of this
paper. In terms of this example, decisions to use the four different service
facilities are the inputs to our shopping system, and the foui consumer items
obtained are the outputs (making the oversimplifying assumption that at
every service facility, one and only one consumer item can be obtained).
This definition of system is the starting point of Mesarovic’s approach to
system theory and what he calls general system theory. In his view, the
latter is concerned with the most fundamental, abstract, and general aspects
of various systems, whereas system theory, at least as reflected in the work
524 HASSAN MORTAZAVIAN

of some authors, tends to deal with more specific questions about more
specific types of systems (e.g., dynamical systems, automata, control sys¬
tems, game-theoretic systems, etc.). The main reason for adopting the name
general system theory has been to emphasize the point that system theory
should be considered as a general theory, not as a theory of specific types of
systems, for instance, control systems. In my view, the term system theory
is comprehensive enough to capture all aspects of the field.
David Berlinski has criticized Mesarovic’s definition of a system as a
relation defined on a set of objects on the grounds that this definition col¬
lapses the notion of a system into the notions of a sequence and a set.
[Berlinski, 1976, pp. 28-29.] My response to Berlinski’s criticism is that
almost every abstract mathematical concept collapses into a few elementary
notions such as sets, relations, points, and so on. The notion of a graph, for
example, simply collapses to the notion of points and edges (or elements and
relations between them), and yet graph theory is a powerful mathematical
theory with a wide range of applications. An important advantage of this
definition of system is that it stresses the structural characteristics of phe¬
nomena.

WHAT IS SYSTEM THEORY?

In the second part of this paper, I discussed the concept of system; now, I
shall examine the essential questions of system theory, in particular: (1)
What is system theory? (2) How and why did it emerge as a new discipline?
(3) Why do we need to learn about its findings?
The terms system, systems approach, system analysis, system science,
and so on, are so broadly used, and occasionally misused, that outsiders
sometimes wonder what we really mean by system theory. Moreover, we
often see systems engineering, general system theory, mathematical system
theory, and so on. These terms all refer to various aspects of the same
discipline, to wit, system theory.
System theory does not aim at unifying all or special types of scientific
disciplines. It does, however, provide a methodological base for solving a
number of problems common to various disciplines. Roughly speaking, sys¬
tem theory is the theory of phenomena that are “system-bound,” “system-
dependent,” or “system-related.” By this, I mean those phenomena that
depend on the conditions of the environment in which they exist, interact
with this environment, and thus cannot properly be studied in isolation.
Let us consider some examples of such system-dependent phenomena.
The rate of growth of population in a country, for example, generally de¬
pends on such factors as average income of individuals, income-distribution
patterns, age distribution of population, distribution of male and female
populations, level of education, and so on. Therefore, a model of population
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 525

growth for a country cannot be developed unless we also consider the rele¬
vant variables of socioeconomic environment in that country. Or, consider
the problem of choice of an optimal energy policy for a nation. Such an
optimal policy, if it exists, obviously depends on a number of national as
well as global economic, political, environmental, natural, and other factors
that should be specified for every particular nation.
All such phenomena are system-dependent, and the aim of system theory
is to develop methods that can be used in analyzing, understanding, ex¬
plaining, modeling, controlling, and predicting such system-dependent phe¬
nomena. But techniques of system theory are general in nature, not related
by any means to the subject matter of specific disciplines. System theory
aims at providing a common complement of methods that deal with such
concepts as dynamics, stability, complexity, catastrophe, hierarchy, struc¬
ture, and so on.
Dynamical systems are those that change their state over time. If the state
of a dynamical system changes in the same way over time, it is called time-
invariant, and if the way in which the state changes varies over time, it is
called time-varying. Consider some specific examples: a traveling space
vehicle, a home-heating system, an operating computer, a growing popula¬
tion, the flow of traffic on a road, genetic transference, ecological decay, and
so forth. While each one of these examples refers to a specific dynamic
behavior, the general concept of dynamics transcends all such particular
situations. Countless examples of such dynamic situations in various areas
can be supplied, but they can all be represented by a small number of
mathematical models. System theory is concerned (among other things) with
specifying all such general mathematical models that can represent dynam¬
ics in various situations. This is done by exploring that part of mathematics
that is suitable for modeling dynamic behavior. It turns out that at least the
elementary aspects of smooth dynamical systems can be almost completely
studied by differential and difference equations—the first used when time is
considered to be continuous, the second when time is considered to be
discrete. The velocity of a car, or the flow of water into a tank, changes
continuously; therefore, velocity and flow are modeled in continuous time,
hence, by differential equations that relate the rate of change of a variable to
its level at every point in time. The number of people entering a shopping
center, on the other hand, changes discretely; therefore, it is modeled in
discrete time, hence, by difference equations that relate the value of a vari¬
able at one time to values at adjacent times. That part of system theory that
provides models for such dynamical phenomena, using mostly differential
and difference equations, is called the theory of dynamical systems.
[Kalman, Falb, and Arbib, 1969; Brockett, 1971; Mesarovic and Takahara,
1975; and Luenberger, 1979.]

‘Of course, at a higher level of sophistication, the theory of dynamical systems requires a
number of algebraic, geometrical, topological, and catastrophe-theoretic concepts as well.
526 HASSAN MORTAZAVIAN

In analyzing complex phenomena, we have to consider more than one

variable at a time. Indeed, complex systems are characterized by interac¬
tions both among their variables, elements, or subsystems (all referring to
the same thing) and with their environment. The ability to deal effectively
with large numbers of interrelated variables is one of the most important
characteristics of system theory. System theory provides mathematical tools
for manipulating large numbers of simultaneous relations. Part of the defini¬
tion of a complex system is that it is multistate. On the other hand, multi¬
input/multi-output systems are called multivariable, though multivariable
systems are not necessarily complex.
Consider, for example, the problem of controlling air traffic at a modern
airport. There are aircraft arriving, say, every five minutes. They have to be
assigned a location in one of the available queues before they can land. Some
of these flights obviously have priority over others; for example, interna¬
tional flights may have priority over domestic flights in landing. At the same
time, a number of aircraft are queued up on the strips on the ground, waiting
for their turn to take off. Again, some of these may have priority over others.
Now, if the same strip is used for both landing and take-off, we have to
coordinate two groups of queues, since no aircraft can land unless the take¬
off path is already cleared. The fact that there may be more than one queue
for landing and take-off and that there are internal priority policies to be
considered already makes the problem of coordinating and controlling land¬
ing and take-off schedules quite complicated. To be sure, there are at least a
dozen more factors to be considered in a real situation. (For example, emer¬
gency situations may arise; too many aircraft may arrive at the same time;
weather conditions may change unexpectedly; and so on.) The problem of
control at an airport thus presents a number of theoretical problems to be
solved by control theory, which is part of the discipline of system theory.
System-theoretic techniques thus have a wide range of applications,
though system theory itself is essentially a theoretical discipline. Whenever
we deal with a situation where there are too many variables, patterns of
interrelations change over time, the value of a variable has to be predicted, a
model has to be developed to represent a system, or a system is to be
controlled in a specific manner to achieve a particular goal, system theory
surely has something to contribute.
The following is a random list of fields that are or will have to be included
in system theory: game theory, computer science, programming theory,
cybernetics, operations research, computation theory, modeling, control
theory, econometrics, optimization, artificial intelligence, pattern recogni¬
tion, algebraic coding theory, and circuit theory.
How did system theory emerge? I submit that the following factors were
involved:

1. There was a general awareness among scientists in a variety of disci¬

plines, mainly engineering, that the same types of problems are dealt with in
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 527

apparently unrelated areas. For example, goal-seeking behavior, communi¬

cation, and control patterns were found to be common among machines,
biological systems, and even organizational and social systems. This aware¬
ness led Norbert Wiener to the idea of cybernetics as a general theory of
control and communication in machines and animals. [Wiener, 1948; 1950.]
2. At the same time, Bertalanffy [1950] and others proposed the idea of a
general system theory as a philosophy of science that transcends all aca¬
demic disciplines, even mathematics. Even though Bertalanffy’s idea had an
influence on the later development of system theory, it was not successful,
mainly because it rejected mathematics as the language of science. Conse¬
quently, what is presently referred to as system theory does not have much
in common with Bertalanffy’s general system theory.
3. In the early 1960s, a new trend appeared, aimed at developing a
mathematical theory of complex interactive phenomena, which led to the
development of mathematical system theory. (I hasten to add that system
theory and mathematical system theory refer to one and the same thing.
There is no such thing as a nonmathematical system theory.) The major
efforts in this direction are reflected in the works of a long list of research¬
ers.2 Although these authors have approached system-theoretic problems
from different perspectives, using different analytical tools, they have all
contributed to the development of system theory. As a result of these vari¬
ous contributions, and several other significant developments by others,
system theory at the present time has found a more or less definitive form.

Although I cannot in the allotted space give more than only the sketchiest
outline, I wish to emphasize that whenever we deal with complex interactive
phenomena, we cannot avoid using the results of system theory, in the same
way that we cannot avoid using the results of physics or chemistry when we
want to study the nature of physical objects.

A system is a precise mathematical object and system theory a mathemat¬

ical theory, though not entirely a branch of mathematics. Any attempt to
apply system theory outside its mathematical context is then not proper.
This remark makes it clear that much of what is considered as applications of
system theory to various other disciplines, mainly social sciences, is not
system-theoretic. (Not that system theory could not ideally be applied to

2 Among the most important of the contributions are: Kalman, Falb, and Arbib [1969], Bensous-
san [1971]; Bensoussan and Lions [1978; 1981]; Bensoussan, Delfour, and Mitter [1982]; Bell¬
man [1971]; Brockett [1971; 1978]; Brunovsky [1970]; Byrnes and Falb [1979]; Fuhrman [1979];
Hazewinkel and Kalman [1976]; Hermann and Martin [1977]; Martin and Hermann [1978];
Kalman [1960a; 19606; 1962; 1965; 1971; 1972; 1976; 1979; 1980; 1981; 1982]; Klir [1969];
Luenberger [1964; 1966; 1971; 1979]; Mesarovic [1964]; Mesarovic, Macko, and Takahara
[1970]; Mesarovic and Takahara [1975]; Pontryagin et al. [1962]; Popov [1972; 1973]; Rosen-
brock [1970]; Rouchaleau and Sontag [1979]; Tannenbaum [1981]; Thom [1972; 1974; 1977;
1980]; Willems [1979]; Wonham [1967; 1974]; Zadeh and Desoer [1963]; Zadeh and Polak [1969].
528 HASSAN MORTAZAVIAN

social sciences; I am simply emphasizing the need for, but want of, appropri¬
ate applications.)
Before proceeding further, I would like to suggest the following definition
for the discipline of system theory:

System theory (or system science) is the theory (science) of mathemat¬

ical models of complex interactive systems.

The true meaning of this definition, and the justification for its adoption, will
become clear in the following pages.
In order to present some of the specific results of system theory, I need to
return occasionally to the discussion of the specification of systems though
this will involve some repetition. I will first take up a particular type of
specification, called input/output description because the system is de¬
scribed in terms of inputs and outputs. If inputs and outputs of a concrete
system are distinguishable, then it can be modeled or represented by an
abstract (mathematical) input/output system.
Following the definition of a system as a relation (or relations) defined on
a set, we may now partition the elements of the set into two object sets, one
consisting of input elements, the other of output elements. In this way, a
constructive definition of a system can be provided. A concrete input/output
physical system is a part of the world isolated naturally or artificially from
the rest of the world and considered as a box in which something (matter,
energy, or information) enters at certain times and that itself puts out some¬
thing at certain times. At each moment of time, the system, be it continuous
or discrete, receives some input and produces some output. When the sys¬
tem produces output discretely, we assume that the output between two
points of time is simply zero. We assume that values of the input are taken
from a specific set, which for every system will be defined depending on the
circumstances. For example, a country’s imports are often restricted by
both governmental regulatory policies and the conditions of the interna¬
tional-trade system. Or, another example, the amount of information a par¬
ticular computer can process in every second is limited to the capacity of its
central-processing unit (CPU). Such restrictions on inputs are quite natural
and related either to bounds on inputs or the nature of the inputs. Thus, we
define a set of permissible or allowable inputs to the system. In the same
way, we define a set of permissible outputs from a system, though with much
milder restrictions.
A philosophical justification for such an input/output description of a
system helps make it clear that the notion of an input/output system is not
simply a practical device. If we consider the system S contained in a specific
domain D of the world with walls around it, then we need somehow to get
inside this domain in order to be able either to control it or observe it. This
remark paraphrases one made by Rene Thom:
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 529

... a system is never completely independent of us, human observers (if it was,
we had better forget about it . . . ); hence the walls of D have to be provided
with windows, through which we may act on the system or, conversely, re¬
ceive information about its inner state. This justifies introducing the notion of
Input and Output of the system, or—in slightly different terms—of control
parameters and of observables. (Thom, 1980, p. 3.)

In the input/output description of a system, the assumption is that we do not

know anything about the internal structure of the system. All we observe (or
know for sure) are the inputs and outputs. This type of description is also
referred to as external description.
Now, an interesting and natural question arises. Can we say what the
output of a system would be by just observing its input at the present time?
The answer is generally no. In other words, we may also need to know the
past history of the inputs of the system. Depending on the structure of the
system, we may need to go back one time-point, two time-points, or more, in
the past. Thus, the output of a system generally depends on both its present
input and its past history, meaning some or all of its input/output pairs in the
past. It is, however, best not to differentiate between present input and past
input. The reason for this is that we would like to introduce a notion of state
fundamental to our understanding of systems and containing those parts of
the past and the present history of the system that are relevant to determin¬
ing present and future outputs. We say, therefore, that the present output
depends on the state of the system, and the (present) state of the system is
informally defined as that part of the present and the past history of the
system that is relevant to determining present and future outputs. The pres¬
ent state of a system, then, must be thought of as some (internal) attribute of
the system at the present moment that determines the present output com¬
pletely and affects future outputs. This idea is expressed in the following
statement:

Intuitively, the state may be regarded as a kind of information storage or

memory or an accumulation of past causes. We must, of course, demand that
the set of internal states of the system be sufficiently rich to carry all informa¬
tion about the history of the system needed to predict the effect of the past
upon the future. We do not insist, however, that the state be the least such
information, although this is often a convenient simplifying assumption. (Kal¬
man, Falb, and Arbib, 1969, pp. 4-5.)

A system is finite-dimensional if its so-called state space is finite¬

dimensional. By this, we simply mean that the number of variables used to
describe the state of the system at every point of time is a finite number.
Most real-life systems are finite-dimensional. In large-scale systems, the
number of such variables is relatively larger, but it is still finite. For ex¬
ample, the state of an economy can (supposedly) be described by a finite
530 HASSAN MORTAZAVIAN

number of such variables as output rates of production sectors, consumption

rates of consumption sectors, prices of various commodities, level of sav¬
ings, and so on.
By a finite-state system, on the other hand, we mean a system that
evolves in a finite number of states or takes only a finite number of states. An
economy cannot be modeled as a finite-state system (i.e., as an automaton),
whereas a sequential digital computer is a finite-state system.
Several more concepts of importance in system theory remain to be
discussed: reachability, controllability, observability, realization, para¬
meterization, identification, structure, and model. Since system-theoretic
concepts are purely mathematical concepts, my attempts to make them
comprehensible without using mathematical notations may prove unsuc¬
cessful. I should first recall the need to make three distinctions: (1) a “con¬
crete” system, sometimes called a process or a plant; (2) an “abstract” or
mathematical system, which is usually meant if system is used without a
modifying adjective; and (3) an (abstract mathematical) “model” or “class
of models” for a concrete system.
Suppose a concrete system is given or to be constructed. We describe this
system by a set of mathematical equations. This set of equations defines a
(mathematical) system and may be taken as a model for that particular
concrete system. Therefore, our basic assumption here is that a mathemat¬
ical model for a particular concrete system—existing or to be constructed—
is already at hand. (I shall later return to the discussion of model building.)
We can think of all information-processing taking place inside a concrete
system (or, equivalently, inside an abstract system that represents it) as
expressed by transformations of states. Then, we may ask the following
basic questions: (1) What effect does a specific input have on the state of the
system? (2) What effect does the state have on output, or what do the
outputs tell about the system? The answer to these two questions lies in
the explication of two basic system-theoretic concepts: reachability and
observability.

A system is (completely) reachable if every part of it is accessible to

inputs. In other words, if any state can be produced by applying a
suitable input, then the system is reachable.

For example, if we can change every significant situation in an economy by

applying a suitable control policy (input or set of inputs), then the mathemat¬
ical system-model representing the economy is said to be (completely)
reachable.
On the other hand, as a dual of the concept of reachability, and with
strictly dual reasoning, we can show that

if we can always determine the internal conditions of the system—that

is, the instantaneous values of all its state variables—from the infor-
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 531

mation contained in the output, then the system is (completely) observ¬

able.

Observability of a system is an abstract mathematical concept and should

not be mistaken for the ordinary process of observation. In the language of
mathematical system theory, a system is said to be completely observable if
its states can be inferred from the output structure. To determine the com¬
plete mathematical conditions of reachability and observability of all classes
of systems is a fundamental question of system theory. The answer to this
question has been found for several classes of systems.
In order to make these two abstract notions clear, consider a simple
example (adapted from Rugh [1975]). A system of two water buckets, one
above the other, consists of two states, one control, and one output. The
two states are the depth of water in each bucket; the control is the water flow
into the second bucket; and the output is the water flow out of the second
bucket. This concrete system can be represented in an input/output form,
depicting a single-input/single-output system. We need not assume knowl¬
edge of the (internal) states of the system.
If no water flows from the second bucket into the first, the depth of water
in the first, that is, the first state-variable, cannot be influenced by inflow into
the second bucket. Thus, we do not expect the mathematical system repre¬
senting this system to be completely reachable. (Remember that we have
reserved the concepts of reachability, observability, and so on, for mathe¬
matical systems representing concrete systems and shall not apply them to
concrete systems themselves.) Thus, if the mathematical system represent¬
ing a concrete system is not reachable, we conclude that the states of the
concrete system should not be reachable either—assuming, of course, that
the mathematical system is a correct representation of the concrete system.
To repeat, system-theoretic concepts apply only to mathematical systems
and have implications for concrete systems only in an indirect way.
In a different system, where water from the outside flows into the upper
bucket and from there into the second bucket, and the output is the outflow
from the upper bucket, we find a different situation. Since the output from
the first bucket is not influenced by the water-level of the second—which
implies that not every state of the system influences an output—the mathe¬
matical system representing this (concrete) system is not completely observ¬
able.
Reachability is a necessary condition for controllability of a system and
thus it is the fundamental system-theoretic concept underlying all questions
of control. Observability, on the other hand, is the fundamental concept
underlying all questions of estimation of practically measurable variables.
We sometimes undertake to estimate unmeasurable variables from their
effects on measurable variables. Therefore, if the system is observable,
unmeasurable variables can be deduced by computation from the measured
ones.
532 HASSAN MORTAZAVIAN

Complete controllability (observability) means that all states are control¬

lable (observable). The concepts of complete controllability and complete
observability are fundamental theoretical notions, and many techniques of
modern control design rely on satisfying these conditions. From a practical
point of view, however, we do not always need to go through laborious
formal calculations to check these conditions. The reason is that in practical
applications, a good analyst can often decide from certain structural consid¬
erations whether a system is controllable (observable). I should mention that
complete observability is also important in the context of system control.
The reason is that we usually determine the controls to be applied to a
system on the basis of measurements of available outputs. Inputs affect the
states, and states affect the outputs; thus, the concepts of controllability and
observability relate the input and output structure to the internal state mech¬
anism. If the output structure is deficient in that it does not convey sufficient
information about the internal state of a system, then it may not be possible
to devise suitable control strategies to make the system behave in a desired
way. Thus, in general, good control requires both the ability to get sufficient
knowledge about the internal state of the system (observability) and the
ability to change the values of states and thereby the behavior of the system
(controllability).3
If a system is not reachable as presented, it can always be reduced to a
reachable one by throwing away all its unreachable states. Likewise, if a
system is not observable as presented, it can always be reduced to an ob¬
servable one by throwing away all its nonobservable states.
The intimate relation of the concept of observability with the question of
measurement of information in information sciences should be given proper
consideration. If we apply both types of reductions just mentioned in a
combined form, any system can be reduced to a form both reachable and
observable. Such a form, or such a reduced-form system, is called
canonical.
Neither the first nor the second bucket-system just described can have
canonical mathematical representations, since these systems contained non-
reachable or nonobservable states, unless we disregard such states. This
means that in the first bucket-system we must disregard the depth of water in
the first bucket, because it is not influenced by the control, which makes the
system nonreachable. In the second bucket-system, we must disregard the
depth of water in the second bucket, since it does not influence the only
output of the system (that is, the outflow from the first bucket), which makes
the system nonobservable. After such reduction is performed, the system is
canonical in form. Thus, canonical means that a system is both reachable
(every state can be reached from an input) and observable (every state can
be observed from an output).

’For a more detailed discussion, see Luenberger’s very instructive elementary textbook.
[Luenberger, 1979.]
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 533

A third fundamental concept in system theory is realization. The problem

of realization is the basic problem of identification, and, hence, modeling.
Modeling is one of the main aims of system theory, with many implications
for various fields of application from physics to econometrics.
Why do we need to develop models? Classical physical sciences dealt
with situations where systems under study were relatively simple. The solar
system, for example, with all its apparent complexity, is a much less com¬
plex system than a computer. The solar system can be studied using only the
laws of physics. Although various parts of a computer obey the laws of
physics, the functioning of a computer cannot be understood by these laws
alone. It depends not only on the structure of its parts but also on the way
these parts are interconnected. The same holds true for a biological system,
such as the human brain, an economic system, a political system, and so on.
Modeling is indispensable if complex systems are to be understood.
There are basically two approaches to modeling, each suitable to a partic¬
ular type of modeling problem. The first approach is based on the assump¬
tion of laws governing the functioning of systems. The second is based on
the assumption that the laws, if any, governing the functioning of a system
are not known and, therefore, that a model must be constructed of the
relevant data representing behavior of the concrete system to be modeled.
The first approach is essentially theory-dominated, and the second approach
is data-dominated. Both these approaches have advantages and disadvan¬
tages.
The first approach consists of writing down equations describing the func¬
tioning of a concrete system based on laws, such as Newton’s law, Kirch-
off s law, and so forth, and then fitting equations to data to determine values
of the so-called free parameters that are not accessible to direct measure¬
ments or observation. Kalman has made interesting remarks about this ap¬
proach (although he is essentially inclined toward the second approach), and
I paraphrase him. [Kalman, 1974, pp. 496-500.]

1. Knowledge of physical laws is not enough, we must deduce their

consequences in complicated situations. If the real system is too com¬
plex, we build a model. This allows deductions to be made more
easily or more economically.
2. If the laws on which the model is based are valid, fitting the model to
real data provides a method for determining values of parameters that
are not accessible to direct measurement or observation. By means of
the model, such parameters may be determined indirectly. If the
model has no theoretical basis, then the parameters have no concrete
meaning even if they have a well-determined, stable numerical value.
3. If model building is divorced from a priori laws, the existence of the
model, however accurately it fits the data, has no implications beyond
a kind of efficient tabulation of data.
534 HASSAN MORTAZAVIAN

4. It may be possible to contradict a law by the impossibility of fitting

models based on it to real data. The converse situation occurs much
less frequently: A well-fitting model does not establish the law on
which it is based.

Kalman describes the main features of theory-dominated modeling; again,

I paraphrase him:

1. The relevant behavior of a real system must be isolated and described

in a formal mathematical way.
2. A class of models must be given, constructed on the basis of laws
supposed to govern the relevant phenomenon and containing a suffi¬
cient number of free parameters to allow the behavior of the model to
be fitted to real behavior.
3. The class of models must include the behavior in question (now inter¬
preted as a model), so that we are assured a priori of a realization of
the behavior within the given class of models.
4. Conceptually, models and real systems must be sharply distin¬
guished. System theory is the study of the former, not the latter.

Kalman’s main criticism of this philosophy of model construction is that

the models developed are strongly dependent on laws, which has at least
three implications. First, that laws built into a model usually constitute a
more important body of quantitative data than the quantitative knowledge of
the free parameters of the model, and, therefore, nothing essentially new can
be learned from the model. The structure of the system is predetermined by
the laws and only certain parameters have been estimated. This process
should not be called system identification in the true sense of the term, since
it basically consists of parameter estimation. The second limitation is that
this type of modeling is not possible in areas of science where we do not
know of any laws governing the behavior of systems. A third limitation of
this modeling approach is that because it is theory-dominated and therefore
disciplinary, the methodological principles developed cannot be easily car¬
ried from one discipline to another.
These considerations motivate the development of the second approach
to modeling, which is basically data-dominated. Here, Kalman takes an
approach that he admits to be purely Platonic. He believes that:

The only possibility of freeing the modeling process from over-dependence on

“laws” is to convert the laws into purely mathematical axioms. Thus laws of
physics (“acceleration proportional to force,” “conservation of momentum”)
are replaced by mathematical statements (“linearity,” “convexity”). As a
consequence, the entire modeling process may be built up in a rigorous mathe¬
matical (hence “mechanizable”) way. (Kalman, 1974, p. 500.)
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 535

A potential advantage of this approach would be that models are then no

longer confined to the boundaries of rigidly separated disciplines and are
equally applicable to physics, biology, computer science, economics, sociol¬
ogy, and so forth. This approach to modeling is based on the assumption that
the fundamental concepts and structures (Kalman uses the word universal
underlying truths, which I shall avoid for the sake of clarity) underlying all
models of dynamical phenomena are purely mathematical in form, whatever
the concrete embodiment of these phenomena (physical, biological, . . .).
Kalman takes these mathematical laws to be certain algebraic rules (such as
linearity).4
The second approach to modeling then conists of finding an equivalence
class of models for a concrete system directly from input/output data (that is,
behavior), which are assumed to have been generated by that system.
Given a fixed amount of data and a specific type of model to be constructed
for the system at hand, say a linear deterministic model, two main questions
to be asked are (1) Could these data have been generated by a linear deter¬
ministic system? and (2) Does the class of all realizations constructed for the
given data contain elements that are essentially different? These two ques¬
tions are referred to as questions of the existence and uniqueness of realiza¬
tions, respectively. The answer to both depends on many formal mathe¬
matical conditions. It can be shown that for finite-dimensional linear
deterministic systems—assuming that the data are complete—a unique re¬
alization can always be constructed. When the data are not complete, the
problem is more complicated, and the answer to this question is only par¬
tially known. Also, the problem of stochastic realization has a rich litera¬
ture. For nonlinear systems, however, the problem of realization is very
complicated and generally unsolved. To explain these abstract concepts,
some degree of formalism would be indispensable; I therefore dispense with
the explanatory exercise.
The realization problem for linear deterministic systems can be formu¬
lated in this question: “Given a complete external, or input/output, descrip¬
tion of a system, can we determine a class of models for the internal struc¬
ture of the system?” The answer is yes. More precisely, given a fixed
amount of input/output data about a (concrete) dynamical system, canon¬
ical realizations always exist; they depend on the data, and only on the data,
and any two canonical realizations based on the same data are isomorphic.
(The reader will find detailed expositions of this classical result in many
places. See, for example, the original paper of Kalman.) [Kalman, 1962.]
The next step is to define model. We must admit the possibility of only
partial success in the process of identification, in that several different sys¬
tems may have to be accepted as the model for a concrete system to be
identified. The basic problem of identification is that of realizations. Identifi-

4For a dissenting view, see Thom [1972],

536 HASSAN MORTAZAVIAN

cation is limited only by the nonuniqueness of realizations. [Kalman, 1981,

p. 5.]
Let us summarize our discussion of the second approach to modeling.
The variables to be considered as the state variables of a specific concrete
system are chosen by the modeler to begin with. This set of state variables,
among other things, determines which data-set must be considered as repre¬
senting the behavior of the concrete system to be identified (i.e., the con¬
crete system for which a model is to be constructed). The realization of these
data generates an equivalence class of models for the concrete system at
hand. This stage of the modeling process is the system theorist’s job. Then,
it is the modeler’s job to choose a model out of the equivalence class of
models generated for this system.
Thus, the class of models identified for a concrete system is dependent on
the data and only on the data describing the behavior of the concrete system
at hand. Considering that the modeler chooses a specific data-set as the one
describing the behavior of a concrete system according to his or her assump¬
tions about the relevance of the data to the concrete system, and in view of
the specific form of arrangement of the data (e.g., input/output form) and
other assumptions, it is clear that the end result of the modeling process
depends on the modeler’s judgments.
The assumption of causality, for example, obviously lies behind all input/
output descriptions of dynamical systems, inputs explicated ab initio as
causes and outputs as effects. In many situations, however, and particularly
in identifying socioeconomic systems, it may be argued convincingly that
inputs and outputs are not distinguishable beforehand. Hence, “the identifi¬
cation of which variables constitute causes and which effects is itself a
modeling question that should be dealt with on a conceptual level.’’ (Wil¬
lems, 1979, p. 72.) To be sure, the existence or nonexistence of causal
relations cannot be ascertained from data. Yet, once such an assumption is
made, identifying which variables constitute causes and which effects
should be possible in principle, though only as part of the modeling process.
Several difficulties impede the second approach to modeling:

1. Complex systems, which contain the class of large-scale systems,

cause serious difficulties. For such systems, it is not at all simple to con¬
struct a reduced model. Indeed, model reduction is an open area of active
research. When the model is small (i.e., canonical), its construction does not
depend on the computational procedure used. When the model is too large
(i.e., not canonical), it will contain parts that do not depend on behavioral
data and, therefore, not on the concrete system to be modeled, but on the
method used to obtain the realization. This dependence is not trivial, nor is it
trivial to recognize the canonical part of a noncanonical model. Indeed, it is
not easy to decide whether a model is canonical or not. On the other hand, it
is mathematically dangerous to try to describe the behavior of a system by a
noncanonical model.
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 537

2. When sufficient behavioral data are not available (the word sufficient
in this context is a well-defined term in realization theory), the class of
constructed realizations among which a model is to be chosen becomes
unreasonably large. Research in this area is in progress.
3. Data-dominated modeling is mathematically precise and safe, but to
some extent, it lacks the transparency of theory-dominated modeling in the
sense that it may not be easy to give concrete interpretation to the relations
contained in the model. Whether or not this is to be considered a serious
difficulty depends partly on the type of problem at hand and partly on the
purpose of the modeling.

A final remark is pertinent at this point. It is extremely important to

distinguish between a concrete system and a model representing it. Many
practitioners, especially social scientists, including economists and econo¬
metricians, have not clearly distinguished between the two. This lack of
clarity as to what is the system and what constitutes a model for it has led to
much confusion. For example, we read in the text of Jan Tinbergen’s Nobel
lecture: “A first subject to be dealt with refers to the necessity to introduce
the element of space into socioeconomic models.” (Tinbergen, 1981, p. 19.)
This use of the word model is not entirely consistent with our definition of
model. All relevant variables of a concrete system, say, a specific socioeco¬
nomic system, must already be included in our description of that system,
and, indeed, this is the only way we can talk about a system as a well-defined
object. Once a system is well-defined, with all its state variables listed and
sufficient information about them provided, a model, that is, a mathematical
representation of that system, can be constructed. Thus, it is not correct to
talk about introducing an element or a variable into a model. In our setting,
the whole problem is reduced to providing the right description of a concrete
system in terms of its state variables, which is the job of the specialist (in this
case the social scientist), then constructing an equivalence class of models
realizing that system, which is the job of the system theorist, and finally
choosing a model from the equivalence class of models constructed and
interpreting and using it, which is again the job of the specialist.
It is important to have the right description of a system before the model¬
ing starts. Without this right description, modeling is bound to be a useless
mathematical exercise. In our setting, then, providing the right description
of a specific economic system is the job of the economist, and constructing
the right mathematical model for that system is basically the job of the
system theorist; thus, a basic part of econometrics becomes essentially a
part of system theory. To many econometricians, this last remark may
sound provocative if not absurd. Yet I shall argue, in line with earlier re¬
marks in this paper, that the main objectives of econometric research, that
is, identifying and modeling systems, belong to the domain of system theory.
For, to repeat once more, system theory is the theory of mathematical
models of (complex) systems.
538 HASSAN MORTAZAVIAN

Returning to the problem of realization, I wish to emphasize that for most

deterministic systems, the realization problem has a unique solution,
whereas for most stochastic systems, the solution of the realization problem
is nonunique.5 A deeper understanding of the problem of realization requires
studying the problem of parameterization of data and model, but I can dis¬
cuss it only briefly, with a reference to a more detailed discussion. [Kalman,
1981.]
A sharp distinction should be made between two types of parameters,
descriptive and intrinsic ones. Descriptive parameters are only subject to
implicit constraints, whereas the constraints on intrinsic parameters are
explicit. Descriptive parameters specify the parameters of a particular sys¬
tem, whereas intrinsic parameters are concerned with only the equivalence
class of realizations. Therefore, a basic problem of identification is specify¬
ing the intrinsic parameters of various classes of systems (e.g., linear deter¬
ministic systems that are finite-dimensional).

The intrinsic parameters of a class of systems should be characterized by two

properties:

(i) There is one and only one parameter set corresponding to any system in the
class.
(ii) The constraints on the parameters are explicit.

... so, in an oversimplified way, we have arrived at the conclusion that

Identification = Realization + Parameterization

(Kalman, 1981, pp. 13 and 19.)

To illustrate the difference between the descriptive and intrinsic parame¬

ters of a model, we consider the example of rotation on a plane (two-space).
[Kalman, 1981.] Roughly speaking, the descriptive parameters of the model
of a system involving rotation in space are functions of an angle. But the
only intrinsic parameter of this model is the angle itself. This parameter is
not free, but subject to an equivalence relation. “What has been generally
referred to as ‘structural parameters,’ particularly in econometric literature,
are the ‘descriptive parameters,’ whereas the truly structural parameters of
a system are its unique intrinsic parameters.” [Kalman, 1981.]
From our discussion of modeling based on input/output data, it should be
clear that such modeling activities are intimately related to the study of
properties of time-series data. However, the intrinsic properties of data—
say, the patterns of jumps in time-series data—have not been discussed in
time-series literature until recently. [Kalman, 1979.] A large number of open
problems are to be investigated in the future. “At present even the prelimi¬
nary question of parameterization of the data raises deep-seated theoretical

5Stochastic realization theory has a rich literature: Kalman [1965]; Rissanen and Kailath [1972];
Faurre [1973]; Akaike [1974]; Picci [1976; 1977]; Faurre et al. [1979]; Van Putten and Van
Schuppen [1979].
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 539

questions. The entire literature of time-series analysis will have to undergo a

certain revision . . . (Kalman, 1981, p. 38.) Intrinsic parameterization is
a rather complicated topic, meriting detailed and technical discussions.
[Kalman, 1982; Tannenbaum, 1981.]

THE CONCEPT OF INFORMATION

A proper discussion of the notion of information should begin with a distinc¬

tion between the concepts of information and amount of information. The
first concept generally refers to content, thus indicating what information a
message or an experiment has supplied, whereas the second answers the
question of how much information has been supplied. A second and most
crucial distinction is that between the concepts of semantic and nonsemantic
information; the latter has been called physical information.
A comprehensive discussion of the meaning of the concept of information
lies outside the scope of this paper. One thing is certain from the literature:
There is no consensus, no clarity, but plenty of confusion. Yet, it is perhaps
possible to draw some distinctions among various denotations and interpre¬
tations of the term.
The central notion in the set of alternative concepts referred to as infor¬
mation is that of semantic information. According to Carnap [1950] and Bar-
Hillel [1964], semantic information conveyed by a statement, which is re¬
ferred to as the content of that statement, is the “class of those possible
states of the universe which are excluded by this statement, that is, the class
of those states whose being the case is incompatible with the truth of the
statement.” (Bar-Hillel, 1964, p. 299.) If we replaced the word states by
state-description, which is technically preferable, the definition for the con¬
tent of a statement becomes the class of all state-descriptions excluded by
this statement.
One basic advantage of this shift to state-description is the necessity of
realizing the treatment with respect to given languages. (The word language
is used in its logical sense, and in this sense, it contains both ordinary
[natural] languages, such as English and French, and specialized [artificial]
languages, such as computer languages, mathematical systems, and so on.)
A basic disadvantage of the work of Carnap and Bar-Hillel in defining
semantic information on languages is that their approach is based on subjec¬
tive probability versus empirical probability and apparently has not led to
significant progress. More recently, it has been suggested that the concept ot
information is actually more fundamental than the concept of pi obability.
[Ingarden and Urbanik, 1962; Forte and Kampe de Feriet, 1967, Cerny and
Brunovsky, 1974.] Indeed, it has been suggested that probabilities exist only
for special cases of information measures. [Cyranski, 1979.]
Based on such an assumption, a new approach to the idea of semantic
information has been suggested, starting with a concept of system that corre-
540 HASSAN MORTAZAVIAN

sponds to the definition of the abstract notion of system previously dis¬

cussed, meaning a relation defined on a set. [Ibid.] Thus, the basic assump¬
tion in this approach is that a system of objects is defined (i.e., completely
characterized) by some class of empirically definable relations on the family
of objects.6 I cannot in this paper undertake to explain the concepts of
information and measurement and the attempts to extend and generalize
these classical notions.
I should not, however, ignore the question of the relation between the
theory of semantic information and the theory of communication or trans¬
mission of information or, even more appropriately called, the theory of
signal transmission, as originally developed by Shannon and Weaver. There
has been a deplorable confusion here due mainly to the misuse of terminol¬
ogy. The two theories have entirely distinct subject-matters and aims. The
theory of semantic information deals with the question of what and how
much content and meaning a proposition, sentence, statement, message,
signal, or observation has. In short, we may ask what and how much infor¬
mation is conveyed. The theory of communication or signal transmission, on
the other hand, is concerned with the question of what happens to a mes¬
sage, signal, or sentence when it is transferred through a channel from a
source to a receiver and what type of message can carry more content or
information. This latter theory is not concerned with the information or the
content of a message or sentence in itself and, thus, should not have been
called information theory.
Although the original writers on the mathematical theory of communica¬
tion, or theory of (transmission of) information, have made it explicit that
they are not concerned with the semantics of communication, others have
failed to issue such a necessary warning. [Shannon and Weaver, 1949.]
Claude Shannon states, “These semantic aspects of communication are ir¬
relevant to the engineering problem.” (Ibid., p. 3.) And Colin Cherry said,
“It is important to emphasize, at the start, that we are not concerned with
the meaning or truth of messages; semantics lies outside the scope of mathe¬
matical information theory.” (Cherry, 1951, p. 383.) Still, Cherry could not
help using this word information in a misleading way.
Although the mathematical theory of communication deliberately ne¬
glects semantic aspects of the process of communication, that is, the ques¬
tion of the meaning of messages, the underlying theoretical distinction
creates problems in application. The reason is that in most real-world situa¬
tions, we are often concerned with almost all of such aspects simulta¬
neously, and, therefore, a comprehensive theory of information dealing with
all these different aspects in a coherent manner is badly needed. Such a
theory does not exist at present. Bar-Hillel tried to develop a calculus of
information, with the aim of showing that it could be regarded as a common

hBy “empirically definable relations,” these authors mean those relations that can be estab¬
lished among objects as a result of measurements performed on them.
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 541

formal system containing both a theory of semantic information and the

theory of signal transmission. [Bar-Hillel, 1964.] He was not entirely suc¬
cessful, mainly because his approach was based on subjective versus empir¬
ical probability. More recent efforts in this direction show that the problem
can be approached without any probabilistic assumptions. This approach
requires the use of a new type of mathematics that has not ordinarily been
used in classical information theory; it may be considered as essentially
system-theoretic though the authors have not made any explicit remark to
this effect. [Cerny and Brunovsky, 1974; Cyranski, 1979.]
In the theory of communication (or signal transmission), we measure the
relative rarity of each signal in a given set and in this way, find a measure of
distortions in the content of what is carried by the signal; we sometimes call
this the amount of information a signal can carry. This approach will be
explicated by a simple example.
Imagine that a man who has two sons A and B finds a message on his desk
indicating that his son had ’phoned him while he was away. If there is no
other information, he would wonder which one of his sons had phoned. This
message can indeed be considered as two potential messages, each referring
to a distinct event: (1) A ’phoned; (2) B ’phoned. Since the event-set does not
have a unique element, the information we can derive from it is obviously
not maximum. In this case, the probability of the occurrence of any one of
the two events is Vi. If the man had three sons, the probability of the occur¬
rence of any one of the events would have been V), and the information
conveyed by the message in this case would be less than in the previous
case. Thus, we can say that the more improbable an event or signal, the
more information it can carry. This is the sense in which the communication
engineer uses the word information and relates it to the concept of probabil¬
ity in order to be able to measure the amount of information. The job of the
communication engineer is, then, basically to design codes that can carry the
maximum possible information.
But this is totally distinct from the problem of analyzing and measuring
the meaning or the semantic information conveyed by messages or con¬
tained in them. It is, of course, also meaningful to compare the semantic
information content of two messages. For example, we may assert that a
statement such as John is a translator contains less semantic information
than a statement such as John translates from several languages; similarly, I
will see you sometime next week contains less semantic information than /
will see you next Monday. But the communication engineer is only inter¬
ested in the comparative rarity of statements of this kind expected to come
from a certain source and would like to have a quantitative measure of these
rarities. It must be perfectly clear, however, as Bar-Hillel has put it, that:

There is no logical connection whatsoever between these two measures, i.e.,

the amount of (semantic) information conveyed by a statement and the mea¬
sure of rarity of kinds of symbol sequences ....
542 HASSAN MORTAZAVIAN

Notice that the concept of semantic information has intrinsically nothing to do

with communication. If an explication for this concept can be found, then the
proposition that all apples are red will carry a certain amount of information
entirely independently of whether a statement to this effect is ever transmitted
. . . (Bar-Hillel, 1964, pp. 286-287.)

It is high time that we ask ourselves what we, and others, too, mean by
the term information science. By information science, we mean the assem¬
blage of systematic studies aimed at understanding, interpreting, analyzing,
and measuring information; and modeling, organizing, and utilizing the pro¬
cess of transferring information, or more generally, knowledge—be it among
humans, humans and machines, or only among machines. From this per¬
spective, communication theory and the theory of semantic information are
considered only as parts, though significant parts, of the broader discipline
of information science.

SYSTEM THEORY AND INFORMATION SCIENCE

Throughout this paper, I have tried to point out the relations between system
theory and information science. It may be helpful if I attempt a synthesis of
my scattered remarks on this question.
An information system in the nonsemantic sense of the term is a special
type of relation or transformation defined on two reference sets: inputs and
outputs. In this case, information generated by the information source may
be considered the input to the system and information derived from the
system the output of the system. The information system can be defined in
this manner if it is both continuous (e.g., nondigital measuring-devices such
as a voltmeter; or devices for measuring pitch and amplitude of a sound
wave, color and brightness of a graphic pattern, field strength of a radio
emission, etc.) or discrete (e.g., a written language, a digital computer, etc.).
Also, as Thom remarked, there is a conceptual relation between an input/
output representation of a system and the requirement to interact with the
outside world (observe the outside world) in an information-theoretic sense.
Indeed, Thom goes as far as relating the apparent time reversibility of physi¬
cal laws with the concept of semantic information. Thom states:

I would claim—as a principle—that any phenomenon is associated with some

kind of irreversibility: for a phenomenon has to appear, hence it has to emit
something which can be seen (or detected through some apparatus amplifying
human vision). Then the apparent time reversibility of physical laws only
shows that these laws do not describe phenomena by themselves, but more
accurately change of frames between observers. They describe, so to speak, how
the same local irreversible phenomenon may be perceived by different ob¬
servers ....
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 543

Putting things more abruptly, I would dare to say that the time reversibility of
physical laws is probably no more than the expression of a sociological con¬
straint, namely communication between several observers. For this constraint
is nothing more than the linguistic constraint between members of the same
linguistic community: when people speak the same language, they share the
same semantic universe: because, to the same sentence, they have to put the
same meaning (or at least, approximately the same). In fact, any observer has
to communicate with himself—with his own past. Flence he needs to have the
possibility of comparing his way of looking at the universe at time t\ with the
look he had at time t0< tp, this requires a common standard of description, a
permanent way of parameterizing the states of the world. Hence reversibility
of the dynamics. (Thom, 1977, pp. 193-194.)

I cannot discuss here Thom’s remarks and their far-reaching philosoph¬

ical implications. I would like to emphasize that the questions of the rela¬
tions between information science and system-theoretic concepts cannot be
resolved without deep study of a number of extremely complex problems.
For example, a theory of semantic information has to be developed in full
consideration of automata theory, artificial intelligence, mathematical
linguistics, hence, system theory as such.
The problem of modeling in system theory has turned out to be related to
the question of deriving information about the state of a system from the
given set of data. More specifically, two basic problems of information sci¬
ences, namely, the problem of analysis and measurement of the information
content of a set of data derived from an experiment, and the problem of
modeling information systems, have become intimately related to realization
theory. I have discussed the significance of the results of realization theory
in analyzing time-series data and concluded that the classical literature of
time-series analysis needs an overall revision to incorporate the recent sys¬
tem-theoretic results. The problem of modeling should be considered as a
very general problem. Models are used as tools for policy analysis and
decision-making. They are also used to design information systems, and,
therefore, they play a crucial role in providing us with proper information
about a situation.
In the third part of this paper, I discussed the basic distinction between
descriptive and intrinsic parameters of both models and data. This distinc¬
tion implies another, namely, between two types of information that can be
derived from the mathematical model developed for a concrete system,
depending on whether this information is based on values of the descriptive
or the intrinsic parameters. The former is called descriptive information, the
latter structural information. Descriptive information derived from a system
is relevant to only that particular system; structural information derived
from a system is relevant to all members of the class of systems to which that
particular system belongs. To illustrate, consider the class of all circles.
Those properties of a particular member of this class that make it distinct
544 HASSAN MORTAZAVIAN

from other circles (i.e., its radius) are descriptive and, hence, can provide us
with only descriptive information about the circle. Those properties of a
circle, on the other hand, that are common to all circles and make it distinct
from anything that is not a circle are intrinsic and, hence, can provide us
with structural information about the circle. The set of properties that distin¬
guish circles from noncircles is invariant under changes in descriptive
parameters. A similar argument applies to various classes of systems.

INFORMATION, SYSTEM STRUCTURE, AND MODELING

As society becomes complex, it requires more structure, organization, regu¬

lation, and information. Thus, the question of how information should be
processed is becoming increasingly important. However, in order to deal
with complex systems, we need to know how to capture complexity. Al¬
though it is perhaps impossible to provide a formal definition of complex
systems, most of us have an intuitive idea of what a complex system is.
One fundamental requirement for identifying the structure of a complex
system is the existence of a minimal information set concerning the struc¬
tural aspects of that particular system. Without adequate information about
inputs, outputs, processing, and location of books in a library, for example,
we cannot develop a model of the operation of a library as a system. The
question of what is the minimal information set required to identify the
structure and develop a (structural) model of a system is, then, an extremely
important but as yet open question. There is, at present, no general theory
that determines such a (unique or nonunique) minimal information set. This
is a fundamental problem in developing general-purpose problem-solving
systems and beyond that, in developing a general theory of modeling. I
believe that the development of such a theory of modeling must be pursued
in the context, and as a natural continuation of (mathematical) system
theory. System theory does have the capacity to represent systems of vari¬
ous forms regardless of their nature. The next step, then, would be to de¬
velop software that transforms mathematical representations into computer
programs.
Although there is no causal link between a model and the reality that is
supposed to be modeled, the selection of a model for a particular subject is
not entirely arbitrary either. Therefore, a process of selecting relevant infor¬
mation is a necessary part of the process of modeling. The specific informa¬
tion set required to model a system depends entirely on the purpose of the
modeling. There is no single model of any system in the world, be it man¬
kind, history, or nature. Thus, there is no unique information set that can
provide a comprehensive model of a system in its entirety. Moreover, there
are no absolute criteria for preferring one model to another.
SYSTEM THEORY AND ITS RELEVANCE TO INFORMATION SCIENCE PROBLEMS 545

CONCLUSIONS

System theory applies to any branch of science in which we deal with com¬
plex interactive phenomena irrespective of the context in which the problem
may arise. System theory is a mathematical discipline though not entirely a
branch of mathematics. It is the theory of mathematical models and applies
to various concrete situations by providing practitioners with tools and tech¬
niques to develop models for concrete systems. More specifically, if a con¬
crete system is described in terms of a sequence of input/output data, the
task of the system theorist is to provide answers to the following questions:

1. How can we construct a mathematical model, describing the internal

structure of the system, for a given input/output sequence (i.e., data
describing the behavior of a system over time), and where several
models can be constructed, how can we obtain the simplest model?
(This simplest model is not necessarily the true model, and it is the
practitioner who has to choose a model out of the class of models to
be constructed.)
2. Given a mathematical model that describes a concrete system and a
set of admissible inputs, how can we determine the set of possible
outputs of the system? In other words, how can we determine the set
of possible behavioral modes of the system?
3. With a prescribed mode of performing measurements on the outputs
of a system, is it possible to determine uniquely the (internal) state of
the system at any time?
4. Given a criterion of performance for the behavior of a concrete sys¬
tem described by a mathematical model that uses a set of admissible
inputs (i.e., controls), what is the best value that this criterion can be
made to assume?

Answers to these questions are known for certain classes of systems. How¬
ever, we are still far from being able to provide complete answers to all these
and several other questions for all classes of systems.
Let us emphasize, however, that whenever we have to deal with complex
interactive phenomena we cannot avoid using the results of system theory, in
the same way that we cannot avoid using the results of physics or chemistry
when we want to study the nature of physical objects.
 SYSTEM THEORY, MATHEMATICS,
AND QUANTIFICATION

Kenneth E. Boulding

Within its rather limited framework, Mortazavian’s paper is very instruc¬

tive. The argument is easy to follow, and I learned a good deal from it. It is,
however, very narrow in scope, defining systems only in terms of mathemat¬
ical models, and, as he says, general system theory goes far beyond this.
The distinction between semantic and nonsemantic information (what I
myself call Bell Telephone or B.T. information of the Shannon and Weaver
type) is obvious, but also very important, though I do not think the mathe¬
matics of semantics goes very far. The real trouble is that as we move into
more complex systems, existing mathematics becomes more and more in¬
adequate. It is remarkably deficient, for instance, in verbs. I confess I regard
mathematics as one of the least developed of the sciences in terms of its
potential. It is very inadequate at the present time for dealing with systems
beyond at least the moderately simple. The human brain, for instance, is a
system of complexity far beyond the scope of any known mathematics.
Classical mathematics, especially algebra and calculus, is fundamentally a
generalization from the properties of numbers. Unfortunately, the real world
does not consist of numbers or even very much of quantities. It consists of
shapes, sizes, patterns, and “fittings”—that is, structures that fit into each
other. I have sometimes said, I confess a little in jest, that there are not much
more than a dozen numbers in the real world—e, pi, the velocity of light,
Planck’s constant, Eddington’s mysterious 137, and a few things like that
plus zero, 1, 2, 3, 4, 5, 6, 7 plus or minus 2, according to the psychologists.
Beyond that, numbers are a figment of the human imagination or at least a
crutch—though a useful and necessary crutch—to the human mind. The
importance of numbers and numerical mathematics is that they can be
mapped into topological structures, so that by their means we can analyze
topological structures beyond the complexity with which our minds can
deal. It is still topological structures, however, that are the real world, and
the topology of n dimensions is still a fairly primitive science. Even Thom’s
catastrophe theory is still bound to continuous functions, whereas in the real
world, we have system breaks, discontinuities, constant change of parame-
547
548 KENNETH E. BOULDING

ters, and innumerable relations for which there are no mathematical equiva¬
lents. The idea, therefore, that mathematization and quantification are the
only hallmarks of science seems to be quite fallacious. It has done substan¬
tial damage even in the biological sciences, but especially in the social sci¬
ences. I was glad to have my attention called to a remarkable article by B. C.
Brookes entitled “A New Paradigm for Information Science?” that dis¬
cusses these considerations in more detail. [Brookes, 1976.]
In the early days of what is now called the Society for General Systems
Research (of which I was a founding father and first president), I recall
defining a system as anything that was not chaos, and I could hardly blame
Mortazavian if he felt that such a definition was too broad and philosophical
to be very useful. We did, I also recall, define a general system as any
theoretical structure that was of interest to more than one academic disci¬
pline. This, perhaps, is a concept too sociological, especially in light of the
rather accidental nature of the academic disciplines themselves, to have
much prestige. Nevertheless, we cannot avoid the epistemological over¬
tones of the systems concept; if system is equivalent to order (that is, non¬
chaos) or even some useful subset of things, structures, or relations that
exhibit order, the critical question is how we come to perceive, recognize, or
know about the order that presumably exists in the real world. This is nearly
the same question as how do we get order—patterns and structures—inside
our skins, and especially inside our skulls, that represent or map onto corre¬
sponding structures in the real world. The system of human learning is
immensely complex and still very imperfectly understood, but we must
know something about it or we would not be able to transmit the knowledge
structures from one generation to the next, even though we do not know
much about how these knowledge structures are coded inside the body.
In the formation of our knowledge structures, perhaps three processes
can be identified. One is the building of knowledge structure in the nervous
system by the genes, which used to be called instinct. The second is thought,
that is, the internal building up of new knowledge structures by processes
that involve consciousness. A very important element in this is the percep¬
tion of identities—that is, relations that cannot be other than what they are.
Mathematics has a peculiarly important role to play in perceiving identities;
it might almost be defined, indeed, as the sophisticated pursuit of the obvi¬
ous. Mathematical proof is simply the process of finding out what is indeed
obvious, which is not always obvious at first. There is, indeed, a slight rattle
from a skeleton in the closet here, partly from Russell and partly from
Godel, even perhaps partly from some Godot for whom we are still waiting
and who might point out that what is obvious is not always true. It seems
obvious to me that through a point we can draw only one line parallel to
another line not going through the point, but it was not obvious to Riemann
or Einstein. Similarly, it seems obvious that minus minus is plus, but I worry
about the fact that not doing harm is a very different affair from doing good.
It seems obvious, too, that infinity times zero is any finite number we like to
SYSTEM THEORY, MATHEMATICS, AND QUANTIFICATION 549

name though this may involve definitions of both infinity and zero as things
toward which we might travel hopefully but never arrive at, especially as
infinity, and I presume also zero, has more addresses than we used to think.
1 worry though about calculating price indices when the price of a color
television set in 1910 was infinite and the quantity was zero!
The third process in the formation of knowledge structures is experience,
which represents a very complex structure of inputs into and outputs from
the body, certainly involving B.T. information, but also involving much
more than that in terms of structure, for instance, in language or even sense
perception. We suspect that if information input is to change knowledge
structure within the body, it must itself be structured in such a way that it
“fits” into some internal pattern. (I believe strongly in the survival of the
“fitting” rather than of the fittest!) If someone addresses a remark to me in
Chinese, a language with which, I regret, I am unacquainted, it will have a
very different impact from the same remark translated into English, even
though both remarks may contain the same number of bits of B.T. informa¬
tion. Even the expression B.T. information will have a different impact on
my reader in the last sentence than it would have had before reading the
paper.
In the learning process, the impact of current inputs, whatever they are,
depends on the existing structure of the mind, and this depends in no small
measure on the existence of records of the past. We go to a familiar place
and are able to find our way around it, because in our minds there is a
structure (memory) that has recorded not only past inputs but past struc¬
tures of knowledge. The success of science as a knowledge-expander de¬
pends in no small measure on that aspect of the scientific subculture that
consists of keeping careful records of the past, whether this results in maps,
the paths of the planets, national income statistics, or experimental results.
Both observation and experience, the two great sources of the inputs of
science, involve records of the past. Our perception of patterns in the
records of the past is the foundation of both prediction and experiment,
which is usually prediction of some future pattern of small contrived sys¬
tems. Our ability to interpret such records and patterns, however, depends
largely on whether they fit into previously discovered obvious relations,
which are usually, though not always, mathematical in type. I have argued
that the great laws of science are either truisms (mathematical identities) or
near truisms, relations that we are almost sure have to be this way, which
might rank as axioms. Much empirical work is devoted not to finding empir¬
ical laws, which are always unreliable, but in spotting where, in the empir¬
ical fields, the basic truisms apply. A good example is the principle of con¬
servation, which simply says that if there is a fixed quantity of anything, any
increase in one place or form must be offset by decreases in other places or
forms. The empirical question is “What is there a fixed quantity of? En¬
ergy? Matter? Money? And so on. A related identity is what I have called the
“bathtub theorem”—that the increase in anything in a given period is equal
550 KENNETH E. BOULDINC

to the additions minus the subtractions. Demography is based on this; so is

Keynesian economics, which starts with the truism that everything that has
been produced in a given period has either been consumed or is still around.
I know of no mathematical model that does not start off with truisms. To
have properties, however, a model has to have behavioral relations, usually
in the form of equations, though they can be more complex (inequalities,
limits, set functions, etc.), that have constant parameters. Indeed, a parame¬
ter is defined as a constant of the system. One of the great problems of the
real world, however, is that very few systems have constant parameters;
indeed, about the only good example is the solar system and its parallels,
which is why celestial mechanics has been so successful. Evolutionary and
information systems, however, have constantly changing parameters. The
ardent determinist chases rates of change and rates of change of rates of
change in the usually vain quest for stable parameters, but the awful truth is
that the real world has strong elements of inherent unpredictability. Sto¬
chastic models do not solve the problem, because probabilities also are un¬
predictable and variable. If the real world, as Ackoff has suggested, is a
mess, it is a great mistake to be clear about it. Mathematicians, however,
have a certain passion for clarity, being devotees of the obvious, which is
why they should probably be kept on a suitable leash. Even the theory of
fuzzy sets is about rather nice fuzzy sets.
Perhaps much of this discussion revolves around what is noise and what is
not, as Mortazavian has hinted. Just as there are concepts beyond B.T.
information in know-how, know-what, and even know-whether, so there
must be concepts beyond B.T. noise, though I find it hard to identify them
exactly. I am convinced, however, that the orderly loss of information, in
some sense or another, is crucial to the human learning process just as
selection—the moderately orderly loss of species—is crucial to the evolu¬
tionary process. The parallels between learning and evolution have, indeed,
been pointed out many times. This is why we have to be extremely careful
about the incautious application of the more traditional mathematical mod¬
els. Decision processes are a good case in point. Mathematical models usu¬
ally involve some kind of maximization of an objective variable or measure
of value or utility. They not only easily lead to suboptimization (finding the
best way to do something that should not be done at all), but they lead to a
neglect of the agenda problem—that is, among what alternatives are we
choosing? The elimination of agenda items, often by quite nonrational pro¬
cesses like boredom or anger—may be a much more important decision¬
making process than any kind of optimization with given agenda. Informa¬
tion overload and noise in the form of the seven deadly sins (pride, greed,
sloth, etc.) may be much more important in actual decisions than rational
optimization. Yet, these are systems, too, mixtures of order and chaos, even
though they are intractable to mathematical treatment, at least with current
methods. This, however, may be an argument for more and better mathe¬
matics rather than less!
 CAN MATHEMATICAL SYSTEMS
BE CONCRETE?

Richard N. Langlois

In a certain sense, Hassan Mortazavian’s paper is as instructive for what it is

not as for what it is. It is not aljout the relevance of system theory to
problems in information science. Apart from a few stray tidbits about infor¬
mation (mostly repeated stern warnings not to confuse semantic and non-
semantic information), the paper has virtually nothing whatever to say about
the relevance of the former discipline to the latter. We might even be led to
think that there is no connection between them. And that would be too bad.
For it seems to me that the constructs of system theory actually provide us
with some potentially useful insights into concepts like knowledge and infor¬
mation, meaning and information content.
What Mortazavian’s paper is about is something called mathematical sys¬
tem theory. This is a well-defined (albeit tacitly defined) intellectual spe¬
cialty. It has its own style, techniques, and attitudes, which Mortazavian’s
piece reflects quite accurately. Despite Mortazavian’s desire to transform
the phrase mathematical system theory into a pleonasm,1 however, this
specialty by no means includes all that system theory is conventionally held
to contain. To many—particularly the so-called general systems theorists—
system theory is a philosophical enterprise, an all-encompassing Weltan¬
schauung and philosophy-of-the-Whole.2 Nonetheless, there is some truth—
descriptively if not normatively—to the contention that mathematical
modeling is the sine qua non of system theory.
In fact, system theory operates by what I tend to think of as a kind of
Platonism. We begin by assuming that there exist various concrete or real-
world entities called systems. The system theorist's problem, like Plato s, is
that we do not have direct knowledge of the behavior of these sensible
things. But we can, instead, study various pure “forms”—in this case,

1 “I hasten to add that system theory and mathematical system theory refer to one and the same
thing. There is no such thing as a nonmathematical system theory.’- (Mortazavian in his paper
in this volume.)
2I touch briefly on these issues in my own paper in this volume.

551
552 RICHARD N. LANGLOIS

mathematical representations called models—in order to learn about con¬

crete systems.
It is characteristic of system theorists to assume tacitly that the problems
of connecting the model with the concrete are difficult in practice but not in
principle. While shuttling in and out of the mathematician’s cave success¬
fully may require hard work and years of training, it is not a procedure that
poses any inherent epistemological difficulties. To the system theorist, all
the important questions are to be answered by interrogating the pure forms
themselves, which, conveniently, can normally be done with mathematics.
In engineering (where simple realist ontologies always work the best), this
approach has proven spectacularly successful. And, as a historical matter,
most of the mathematics of system theory grew out of electrical engineering,
principally circuit theory.
While it may be that there are as many conceptions of system theory as
there are system theorists, it nonetheless seems to me that practitioners fall
into two identifiable if overlapping groups: those who wish to leave the cave
and those who do not.
One well-known manifestation of system theory goes by names like sys¬
tems analysis, the systems view, or the systems approach. Broadly speak¬
ing, this version of system theory subsumes the discipline of operations
research and is concerned with applying mathematical analysis to areas of
military, business, and (yes) social management. It is in this guise that sys¬
tem theory has intruded most forcefully into modern thought; and I will
resist the temptation to offer my own views on the subject.3
The second broad group of practitioners comprises those whose business
it is to scrutinize the pure forms. Mortazavian is firmly in this camp. System
theory, he says, is the theory of (mathematical) models. And he is quick to
quote a warning that this theory “does not accept responsibility for the
accuracy or relevance of these models.” (Kalman, 1980, p. 7.) Yet, while
insisting on the absolute mathematical purity of his discipline, Mortazavian
wishes simultaneously to claim for it all-encompassing relevance. In con¬
secutive paragraphs, we read both that “Any attempt to apply system theory
outside its mathematical context is . . . not proper” and yet that “whenever
we deal with complex interactive phenomena, we cannot avoid using the
results of system theory, in the same way that we cannot avoid using the
results of physics or chemistry when we want to study the nature of physical
objects.” The system theorist, it seems, is that most fortunate of people who
can have his cake and eat it too; like Aphrodite ever-virginal, he can descend
to embrace the worldly without endangering his mathematical purity.

3The literature here is voluminous and wide-ranging. For an intelligent defense of the systems
view, see C. West Churchman [1968b] and for a hysterical one, see Simon Ramo [1969]. For
representative critiques, see Laurence H. Tribe [ 1972] from a philosophical perspective and Ida
Hoos [1972] from an empirical point of view.
CAN MATHEMATICAL SYSTEMS BE CONCRETE? 553

This miraculous circumstance warrants a closer look, for it is the key to

understanding much of Mortazavian’s discussion. The solution to the mys¬
tery, I would argue, lies in recognizing how seriously mathematical system
theorists (of the school Mortazavian represents) take their Platonism.
Apparently following Rudolf Kalman, Mortazavian discusses two ap¬
proaches to modeling concrete systems. The first approach, he writes, "is
based on the assumption of laws governing the functioning of systems. The
second is based on the assumption that the laws, if any, governing the
functioning of a system are not known, and, therefore, that a model must be
constructed on the relevant data representing behavior of the concrete sys¬
tem to be modeled.” This first approach appears to be nothing other than the
hypothetico-deductive method, the approach that lies at the heart of the
modern philosophy of science. This method views models or theories or,
indeed, paradigms or "research programmes”—as logical systems imposed
by the mind to explain natural or social phenomena to that mind. Observed
data are but one element—and not always the most important element that
goes into evaluating and modifying hypothetico-deductive models.4 5 By con¬
trast, the second approach described by Mortazavian implies a naive faith in
data long ago renounced—with good reason—by the philosophy of science.
The important point to notice about this data-dominated approach to
system theory is not its somewhat atavistic empiricism but the Platonism
that is connected to it. The system theorist who adheres to this second
approach sees the mathematical model not as a hazy adumbration of the
concrete system, but as a more-or-less exact replica of it: The model is a sort
of homunculus, identical in all essential ways to the concrete system. To put
it another way, the Platonic system theorist believes that all concrete sys¬
tems actually possess true, unique, knowable mathematical structures. The
job of system identification—note the choice of word—is thus to interrogate
the data and determine directly from these data which pre-formed homun¬
culus (or at least which equivalence-class of acceptable homunculi) accu¬
rately captures the concrete system s true mathematical form.
It should be clear why system theorists should prefer this latter view of
the world: It serves in the quest to unify all sciences under the banner of
system theory, and—not incidentally—it maximizes the role and impor¬
tance of the system theorist in the scientific process. As Mortazavian ex¬
plains, the role of the disciplinary scientist ("the specialist”) in the Platonic

4On this growth-of-knowledge approach—which dominates the modern philosophy of sci¬

ence—See especially Imre Lakatos and Alan Musgrave [1970].
5It is in this context that much of Mortazavian’s discussion and terminology must be under¬
stood. The notion of observability, for example, which sounds as if it has heavy epistemological
meaning, actually has to do solely with the logic of mathematical models and not at all with the
phenomenology of observing a system. The problem of observability is whether, once I already
know all about a system and how it operates, I can always tell what the (known) state variables
are doing just by watching the (known) output variables.
554 RICHARD N. LANGLOIS

method is simply to describe the system with which he or she deals. The
scientist is thus a handmaiden to the great generalist, a kind of graduate
research assistant to the system theorist, whose task it is merely to provide
the right description of the concrete system. It is the system theorist who
performs the real scientific function, spinning the mathematical wheels to
identify the correct equivalence-class of models, which is then given back
into the grateful hands of the scientist.
Fortunately or unfortunately, however, we do not live in this simple
Platonic world. Complex phenomena of nature, mind, and society (concrete
systems) do not themselves possess variables; only mathematical models
have variables. The job of the scientist is not to describe such phenomena in
some nonproblematical fashion; rather, the scientist’s task is one of active
interpretation, mental construction. And mathematical models are often—
but by no means always—useful tools in this creative process. Thus, the
mathematical system theorist, who studies the logic of operation of such
models, retains an important role in the scientific endeavor—even if it is the
same role mathematics has always played in science rather than the sup¬
posedly new and distinctly baronial role the mathematical system theorist
would wish to be assigned.
 SYSTEM THEORY AND
INFORMATION SCIENCE
Further Considerations

Richard Mattessich

The claim that system theory and mathematical system theory refer to one
and the same thing, and the attempt to identify the broad discipline of system
theory with one and only one of its many subareas, is bound to lead to
confusion. Thus, it is important to realize that whenever Mortazavian, in his
paper in this volume, speaks of system theory, he limits himself and his
audience to the relatively narrow area of mathematical system theory. Al¬
though authors should be given freedom to choose their own terminology (or
that of their circle), that freedom must be curtailed whenever convention is
deeply enough rooted in a different meaning—otherwise misunderstanding
is inevitable. But there might be more behind this narrow definition. Is it,
perhaps, an attempt to declare everything beyond the mathematical part in
system theory as trumpery or unscholarly endeavor?

SYSTEM THEORY AS AN EMPIRICAL AND

METHODOLOGICAL CONCERN

Even if we were to accept Mortazavian’s statement that Bertalanffy’s

idea of general system theory “was not successful,” it can hardly be main¬
tained that the system-theoretical works by Ackoff and Emery [1972],
Boulding [1956a], Churchman [1961; 1968a; 19686; 1971; 1979], Simon [1969
and 1981], J. G. Miller [1978], Bunge [1979a], and many others do not belong
to system theory or are less successful than contributions by leading mathe¬
matical system theorists. And yet, Mortazavian, in his paper, asserts that
“A system is a precise mathematical object and system theory a mathemat¬
ical theory, though not entirely a branch of mathematics,” thereby invoking

Additional support from the Social Sciences and Humanities Research Council of Canada is
gratefully acknowledged.

555
556 RICHARD MATTESSICH

Mesarovic’s definition according to which “a system is a relation S C X x Y

on abstract sets X and T.” (Mesarovic, Macko, and Takahara, 1970, p. 66.)
But Mesarovic seems to offer this definition strictly within a “framework of
mathematical system theory” without identifying the latter with system
theory in general. Mortazavian is quite correct when pointing out the impor¬
tance of distinguishing between a concrete system and a model representing
it. But instead of admitting that the study of concrete systems belongs to
system theory or systems science, he compartmentalizes the academic divi¬
sion of labor in such a way that the empirical scientist’s task is to provide the
right description of a concrete system specifically for the system scientist,
who, in turn, constructs an equivalence class of models into which this and
similar models fit. But such an unconventional division of labor, and above
all the exclusion of empirical, methodological, and philosophic aspects from
the system approach is, 1 submit, unacceptable for the following reasons:

1. System theory cannot be derived from the type of a priori assump¬

tions or intuitions that form the basis of logic and mathematics (for example,
the assumption that the negation of a negation of an assertion is identical
with the original assertion) and therefore cannot be part of pure mathemat¬
ics. Furthermore, system theory cannot be conceived without factual obser¬
vation and thus belongs to the realm of empirical science. This is not to deny
that model building is an analytical concern; but the fact that those models
might represent the phenomena of an empirical discipline in a somewhat
unrealistic fashion does not convert the propositions expressed by such
models into analytical ones. This would be too strong an attempt to make a
virtue out of the shortcoming of being unrealistic. Any proposition that lends
itself (in principle) to refutation (in Popper’s sense), whether true or false,
whether realistic or unrealistic, is an empirical proposition.
2. Just as theoretical physics, which also consists of (applied) mathe¬
matical theories (and likewise “is not entirely a branch of mathematics”), is
first of all part of the empirical science of physics, so system theory remains
meaningful only as part of one or more, or even all, of the empirical sciences
apart from also belonging to applied mathematics. Let us remember that
Albert Einstein was not a mathematician but a theoretical physicist with a
strong philosophic bent. And if Bogdanov and Bertalanffy had insufficient
appreciation for mathematics, we must not fall into the trap of reserving
system theory for mathematicians, who lack training in the methodology of
pure and applied empirical sciences. However, this is not meant as a rebuff
to mathematics, which will always be a requisite for system theory.
3. To test whether a conceptual system “fits” a specific factual phenom¬
enon is an indispensable and empirical-methodological concern of system
theory. That is to say, the choice of fitting criteria requires scientific-empir¬
ical as well as normative-methodological expertise that can hardly be rele¬
gated to any other area but system theory and its philosophy.
SYSTEM THEORY AND INFORMATION SCIENCE 557

.
4 A thorough study of the emergence and evolution of system theory
offers clear evidence that the original impetus for this novel viewpoint arose
from the inadequacy of the traditional atomistic and reductionistic approach
when dealing with such factual phenomena as emergent (holistic) properties,
environmental interactions, feedback information, and so forth. [Bogdanov,
1913 and 1922; Bertalanffy, 1928; 1950; 1968.] This inadequacy was hardly
bothersome in physics but was strongly felt in the life sciences; and it is no
coincidence that of the two pioneers of system theory, Bogdanov and Ber¬
talanffy, one was a physician-philosopher, the other a biologist. The fact that
both of them envisaged this new area as a superscience instead of mere
methodology is unfortunate but affords no justification for mathematicians
to usurp system theory and claim it entirely for themselves or deprive it of its
empirical and methodological content.
.
5 The attempt to banish the factual and normative-methodological es¬
sence of system theory stands in crass contradiction to the present state of
the sciences. System methodology has been so widely accepted that system
theories not only evolve in operations research and applied mathematics,
but emerge, in one form or another, in accounting theory, organization
theory, and economics, the management sciences, electronic data process¬
ing, and engineering, and even in epistemology and ontology, of which
Reseller’s and Bunge’s rigorous and integrative work are excellent exam¬
ples. [Rescher, 1979; Bunge, 1979a.]
6. Finally, the system approach, due to its goal orientation, seems to be
particularly suitable as a basis for formulating instrumental hypotheses
(means/end relations), which are of special importance in the applied and
social sciences. This entails the emergence of novel methodological and
other epistemological problems, which I examined in a previously men¬
tioned work. [Mattessich, 1978; 1974a; 1982.]

With regard to Mortazavian’s suggestion to dispense with model-building

when dealing with relatively simple systems, I propose he ask himself
whether it was not model-building par excellence that led Johannes Kepler
and later Isaac Newton to the discovery of fundamental laws of the solar
system and of mechanics in general. Mortazavian’s suggestion that “the
solar system can be studied using only the laws of physics’’ would have been
poor advice for Kepler and Newton, who still had to search for some of
these laws. But even today, when the laws of mechanics and relativity
theory are available, the scholars at NASA still do plenty of model-building
in dealing with the solar system and its exploration. Thus, we might wonder
about Mortazavian’s or Kalman’s distinction between the two approaches to
model-building and how basic this distinction really is. Are we not dealing in
both cases with the same process of model-building, in one case searching
for fundamental relations, in the other case searching for less basic ones? In
both cases, the models are theory-dominated, but in the first case, the
558 RICHARD MATTESSICH

theory is already well supported, while in the second case, it is either (1) a
provisional hypothesis or (2) a mere conjecture that is not expected ever to
fulfill the function of a scientific law.
Furthermore, the paragraphs devoted by Mortazavian to the genealogy of
system theory leave something to be desired. Although Bertalanffy is briefly
referred to, no mention is made of Bogdanov’s more original work [1913 and
1922] nor of the contributions to this area by Ackoff [1963; 1971], Ackoff and
Emery [1972], Bunge [1979a], Churchman [1964; 1968a; 19686; 1971; 1979],
Mattessich [1974a; 1978], J. G. Miller [1978], Simon [1969 and 1981], and
many others. Indeed, even within the subarea of mathematical system
theory, Mortazavian’s references seem to be somewhat lopsided, since his
bibliography contains only a single article from the important journal Mathe¬
matical Systems Theory.
All this is neither meant to deprecate contributions made by Mortazavian
nor to deny the usefulness of mathematical system models. On the contrary,
Bunge’s ontological investigations, for example, provide a comprehensive
axiomatic system-framework and thus acknowledge the importance of math¬
ematical formulations in system theory. But Bunge, choosing the term sys-
temics—as he holds that there is no single system theory but many of
them—emphasizes that

Systemics has two related motivations, one cognitive and one practical. The
cognitive or theoretical rationale of systemics is, of course, the wish to dis¬
cover similarities among systems of all kinds despite their specific differences.
. . . The practical motivation for systemics is the need to cope with the huge
and many-sided system characteristics of industrial societies. . . . (Bunge,
1979a, p. 1.)

Furthermore, it must be noted that among Bunge’s sixty-three system postu¬

lates, there are not only analytical assumptions, but also many empirical
and metaphysical (quasiempirical) assumptions. [Mattessich, 1982, p. 66.]
Ackoff and Emery [1972] also emphasize their interest in concrete systems
and, hence, in the empirical approach; and in a similar vein, I may quote
Herbert Simon, who pleads for an empirical behaviorally oriented system
theory:

The research that was done to design computer time-sharing systems is a good
example of the study of computer behavior as an empirical phenomenon. . . .
To understand them, the systems had to be constructed, and their behavior
observed. . . . Here again theoretical analysis must be accompanied by large
amounts of experimental work .... (Simon, 1969 and 1981, pp. 24-25.)

Thus the definition of system offered by Mortazavian would be acceptable

only (1) if the term abstract system were substituted for the term system, and
(2) if it were supplemented by Mario Bunge’s following definition of a con¬
crete system: “An object is a concrete system iff it is composed of at least
SYSTEM THEORY AND INFORMATION SCIENCE 559

two different connected things.” (Bunge, 1979a, p. 6.) Perhaps I should note
that, for Bunge, a thing is always a concrete object and a connection is more
than a mere relation, since “two things are connected (or coupled or linked
or bonded) if at least one of them acts on the other.”

INFORMATION ECONOMICS, AN INDISPENSABLE PART OF

INFORMATION SCIENCE

The somewhat narrow treatment by Mortazavian not only misses the empir¬
ical and methodological aspects of system theory, but equally neglects the
important area of information economics. Such an omission appears to be
particularly critical wherever an overall synthesis between system theory
and information science is invoked, as it is by Mortazavian. His broad
definition of information science gives no indication of excluding informa¬
tion economics and thus presumably would encompass it. But as Mortaza¬
vian neither discusses nor even mentions this important area, some pertinent
remarks would be due in this paper. I shall, however, deal with information
economics in connection with Richard Langlois s paper, System Theory,
Knowledge, and the Social Sciences,” elsewhere in this volume.
 PROBLEMS OF SYSTEMS THEORY

C. West Churchman

The lead papers in Sections 8 and 9, both of them on the topic of systems or
the systems approach, illustrate a phenomenon common to academics who
write from the point of view of different disciplines, namely, the ignoring of
large hunks of literature on the same topic. I shall add to the confusion and, I
hope, to the enlightenment by discussing still another systems approach,
which both authors have largely ignored. An approach is toward something,
I think (though in the case of the two papers, it seems as though approach is
equivalent to method). The systems approach I want to discuss is an ap¬
proach to planning.
Planning, a very old human activity, means the attempt to understand
enough about a human social system to be able to formulate policies that, if
implemented, will lead to an improvement in the system. This is not a
definition of planning in a reductionist sense, since it uses such words as
understand, policies, and system, which are at least as difficult to define as
the word planning itself. But the description does indicate that we, the
planners, are in the business of trying to help policymakers primarily, and
only secondarily do we strive to excite the intellectual curiosity of people
working in various academic disciplines.
The systems approach to planning is old. I date its first writings back to
the I Ching (say 2000 B.C.),1 which attempted to classify situations in which
humans find themselves and to suggest attitudes about each of them that
could create wisdom in action. Many other ancient and modern books may,
or should, be cited, that I believe say some important things about planning
and a systems approach to it.2 The reader may notice that these books are
written from an extraordinary variety of points of view and mathematics and
economics are no more dominating than psychology, biology, law, psycho¬
analysis, political science, ethics, and epistemology.

‘Compare Richard Langlois’s estimate: “Economists . . . were the first systems theorists.”
2The list should not omit the following: the Bhagavad-Gita, Plato, Aristotle, Lucretius, the New
Testament, St. Augustinus, Hobbes, Spinoza, Kant, Bentham, Edgar Singer [1948], von Ber-
talanffy [1968], Emery [1969], Habermas [1970], Van Gigch and Hill [1971], Beer [1972], Ackoff
and Emery [1972], Hoos [1972], Bateson [1972], Meadows et al. [1972], J. G. Miller [1978],
Jantsch [1980], Checkland [1981],

561
562 C. WEST CHURCHMAN

Since the systems approach to planning is old, it would be ridiculous to

judge it either a success or a failure; instead, it is a way of life for those of us
who have dedicated our lives to trying to improve the human condition by
means of the human intellect. For us, it is almost a platitude to say that we
need to be comprehensive because policies carried into action have many
ramifications outside the problem area first addressed, for example, all poli¬
cies—energy, transportation, communication, military defense, and food—
are tightly united to education. It is also a platitude to say that no matter how
hard we try to be comprehensive, we are in for surprises our intellects
completely missed. An example from recent years is the food system of the
world: Planners concentrated their attention on such matters as food pack¬
ages, green revolutions, and the like; we are now coming to realize that the
problem of starvation (perhaps as many as one billion people in today’s
world are starving) is not a problem of food only, but also of world and
national politics.
I should say that those of us who refrain from creating a discipline out of
planning are scavengers, seeking ideas, methods, and knowledge that will
enable us to increase our understanding of how to better the human condi¬
tion. Most of us have learned to be quite wary because the garbage heap
where we do our scavenging has many a nicely wrapped package with such
attractive labels as organizational development, zero-based budgeting, con¬
trol theory, management information system, or management science. We
have learned long since that these labels are highly deceptive: organizational
development leaves out most of the major issues in trying to redevelop an
organization; management information systems are often computer-based
databanks; zero-based budgeting does not start from scratch (it is zero dis¬
tance from last year’s budget); control theory says virtually nothing to a
manager whose personnel are running amuck; and management science
tends to be largely ignorant of what it is like to manage anything.
But often the most discouraging looking packages contain gems of under¬
standing. Who would think that in a book discouragingly called Foundations
of the Metaphysic of Morals, first published in English in 1799, Kant would
write out the first clear ethical guide to systems design: “Never treat human¬
ity, either in yourself or another, as means only but as an end withal.’’
[Kant, 1799 and 1959.] This is a prescription badly needed in an age like
ours, which shows strong political tendencies to design systems around
utilitarian ethical ideas like productivity and wealth, and use people as means
only in the pursuit of these goals.
Or, consider a more recent book by James Hillman, called The Myth of
Analysis. [Hillman, 1972.] The analysis Hillman had in mind is not the sys¬
tems analysis that Mortazavian describes in his paper, but rather psycho¬
analysis. But the myth (of real progress) applies equally well to either mean¬
ing of analysis.
Of course, being scavengers, some of us become fond of one kind of
package, for example, linear (or nonlinear) programming, game theory, cy-
PROBLEMS OF SYSTEMS THEORY 563

bernetics, queuing theory, and so forth; so there are as many styles of the
systems approach as there are practitioners. For the reader interested in
applications, there is the journal Interfaces, which publishes applications
together with an independent evaluation establishing that the application has
worked. For a compendium of theoretical styles, see Peter Checkland’s
recent book Systems Thinking, Systems Practice. [Checkland, 1981.]
In the remainder of this comment, I shall consider two quite related topics
that also appear in the papers by Richard Langlois and Flassan Mortazavian,
namely, systems information and the relation between systems thinking and
the logic of classes.
There is a fascinating aspect of the logic of planning for action, namely,
that if someone does x, it follows that he or she does not do non-x, assuming
that the descriptors of x are complete (i.e., that within a time span, all the
relevant things he or she does have been enumerated). This statement is not
really an axiom of systems planning because it is very close to being a
tautology. What makes it fascinating is an axiom of all systems planning
today:

Axiom 1. The class non-x not only is nonempty, but also contains a
nonempty subset of actions that could have been taken by the decision¬
maker in the time span.

We even have a technical phrase for this nonempty set: “lost opportuni¬
ties.” The etymology of to decide is very appropriate, since the verb comes
from the Latin decidere, which means to cut, or to cut off. Hence, if action x
is taken during a time span, then the actor (decisionmaker) cuts off a
nonempty set of lost opportunities—forever. That one or more of this set
could be chosen later on does not change the fatality of the lost opportunity,
especially in that large number of cases where timing is crucial. (In chess, for
example, we cannot say to our opponent, “I failed to take your queen with
my pawn on the last move, so now I would like to do it on this move even
though you have moved your queen away.”)
Why is the class of lost opportunities so fascinating to the systems plan¬
ner? Because the characteristics of this class contain a most vital piece of
information, namely, the cost of doing x rather than doing something in the
class of lost opportunities. Now I need another axiom, perhaps not so self-
evident as the previous one:

Axiom 2. For any action x taken over a specific time span, the class
of lost opportunities can be ordered in terms of their value to a class oj
human beings.

Here, I need to point out that this ordering is no simple matter. In texts on
decision-making, it is often assumed that an alternative action is like pulling
the lever on a slot machine, where the action runs itself and we simply await
564 C. WEST CHURCHMAN

the outcome. But if the action includes such matters as investing money,
firing or hiring personnel, or selling property, we have to know how to do
these things well or even optimally. Hence, in assessing the value of a lost
opportunity, we have to understand how the opportunity can best be
managed.
This is a crucial point in the systems approach to planning because in
trying to determine how best to manage a lost opportunity, we are taken out
of the original problem statement into other systems.

Definition. The value of the correctly managed best lost opportunity

is the “opportunity cost’’ of doing x over the time space.

Finally, if I can assume that the value scale that orders the lost opportuni¬
ties is additive, then

Axiom 3. In estimating the cost of doing x, the planner should add

the opportunity cost.

This last axiom need not be so strongly stated; instead, we can use the more
qualitative prescription: The planner must take full account of opportunity
costs.
So much for formalism. Now, I would like to ask an epistemological
question: “How do we come to know opportunity costs?” They are vital
pieces of information for all planners who use the systems approach, and,
hence (assuming my axioms hold), such planners should have an appropriate
epistemology for them.
In the first place, it is clear that determining an opportunity cost is not
empirical. The lost opportunities are all “counterfactual” (if y had been
done instead of jc, then . . . ). Besides, it would be a hopelessly impractical
task to test the consequences of each of the members of the correctly man¬
aged lost opportunities even if we knew what correctly managed meant. If
this comment is valid, then it appears that what is called information theory,
which assumes an empirical base for the information, is largely irrelevant as
far as opportunity costs are concerned.
Rationalism would appear to be a more attractive epistemology with ref¬
erence to Axiom 3. Suppose that the planner has a sufficiently strong theory
to predict the cost of every possible action the decisionmaker could take
over the time span. This, indeed, seems to be the case in linear program¬
ming. The geometry of linear programming is designed to map in multi¬
dimensional space all the possible costs for each set of possible actions.
Actually, we do not normally find any discussion of opportunity costs in the
literature of linear programming, perhaps because the authors think that
their measurement is more or less automatically carried out.
In practice, the opportunity costs are all there, hidden by the method of
estimating the parameters; for example, measures of the amount that a unit
of an activity contributes to the measure of total cost. This very often is
PROBLEMS OF SYSTEMS THEORY 565

estimated empirically but without determining whether the unit that contrib¬
utes to total cost was optimally managed or whether increasing or decreasing
any particular output might lead to considerable organizational disruption.
The epistemological problem of estimating opportunity costs is more eas¬
ily recognized if we turn to a very familiar system, inventory. Suppose my
job is that of a purchasing agent for a retail store. How many white shirts,
size 151/2, 33, of a certain make should I order at one time? If I order too few,
I run up ordering costs. If I purchase too many, I shall face obsolescence
costs, and more to the point, I shall reduce liquid capital. The shirts sit on
the shelf for a number of days and do not produce cash. What could best be
done with the cash they cost me is the opportunity cost of capital, and we do
try to use this number in estimating optimal order quantities. In other words,
the correctly managed lost opportunity is a matter of correctly using the
liquid capital.
But how do we obtain the estimate? Many textbooks on inventory control
tell the student to use the interest that the cash could earn. Aside from the
fact that there is no such thing as the interest in today’s money markets, the
suggestion is absurd at the policy level. If the best the retail store can do with
liquid capital is to invest it in interest-earning accounts, then the store should
liquify all its assets and set them to earning interest.
The question “What is the opportunity cost of holding inventory?” de¬
pends on determining the answer to another question, “What is the best use
of liquid capital?” which depends on the answer to still another question,
“What is the best use of any capital?” which asks for a total financial plan of
the company.
I hope the reader can begin to see why systems information does not
correspond to empirical information at all and why its acquisition is epis¬
temologically mysterious. Far from being able to reduce the question of the
appropriate opportunity cost to simpler questions, we find that it expands
into questions about the larger system. The following is not an axiom, but
more like a theorem:

The more we investigate how a system should work, the larger the
system becomes; or
Under investigation, the boundaries of systems keep breaking out¬
wards; or
The essence of the systems approach is the interconnection between
systems.

I do not want to appear pessimistic about the matter of systems informa¬

tion. The excitement of all inquiry is its mysteries. Of course, we practice
the systems approach even though the critical aspects remain a mystery. Foi
example, another critical aspect of all social systems is the way people relate
to each other; but despite years of studying human relations in psychology,
sociology, organization theory, behavioral science, and so forth, we still
know very little about how various kinds of relations are formed and how
566 C. WEST CHURCHMAN

they influence what happens. Like any other profession, we make the best
guesses we know how to, and we have plenty of hope.
In closing, I would like to make two comments, one more or less technical
and relevant to the two papers, the other more general and germane to world
systems. The technical question deals with the manner in which systems
should be depicted. Here, it seems common sense to say that systems are
made up of parts (components, departments, and the like) and that the
systems-planning and management problem is to determine how the parts
should work together for the good of the whole system (e.g., its goals). All
this common sense suggests that the logic of parts and wholes is the appro¬
priate beginning of systems theory: Belongs to is reflexive, nonsymmetric,
transitive, and so forth. Also, if a belongs to b and b belongs to a, then a and
b are the same.
However, if I want to justify an examination of part b when my interest is
in part a, I would like to define is a part of somewhat as follows: b is a part of
a if the way in which b operates influences strongly the performance of a.
Please understand that this definition is epistemological and might more
reasonably be posed as: Investigation of b is part of the investigation of a if
. . . . But now, the part/whole logic will appear with an overall symmetry.

As investigation proceeds: Every part of any system becomes a part of every

other part, and yet, despite this, the parts of a system do not become all the
same.

So my technical point is that the logic of classes is not the basis of systems
theory, as Mortazavian seems to suggest.
My other point is based on the idea that we need to concentrate on
systems interconnections. I think the point was made very well in the first
book on world modeling, The Limits to Growth. [Meadows et al., 1972.] My
chief interest in the past several years is in the social diseases called militar¬
ism and malnutrition. I and my colleagues at Berkeley have been trying to
see in what sense these are related to each other. Seeing the relation does
not necessarily solve every problem, but it may lead to more enlightenment.
Here is the way we have been thinking about the interconnection:

1. There exist some sociopolitical structures for a nation that would

enable the nation greatly to reduce starvation. (China seems to be an
example though communism may not have the only example. Taiwan
seems to have very little starvation.)
2. It serves the military (defense) interests of certain world powers to
change countries that have adopted such sociopolitical structures
(e.g., Chile) or stop countries from adopting such sociopolitical struc¬
tures (e.g., Vietnam).
3. I conclude that the world food system and the world military system
are interconnected.
 MATHEMATICAL SYSTEMS
THEORY AND
INFORMATION SCIENCES

Mihajlo D. Mesarovic

Much of what Hassan Mortazavian has written on systems in his paper I

agree with—actually, almost all of it. I do not feel I should repeat what he
has already said but rather will restrict myself to additions, comments, and
reservations. Much of what I have to say actually is about what Mortazavian
has chosen not to say. In other words, I am saying that the only criticism I
have of Mortazavian’s paper is that he has committed a sin of omission. This
is fully understandable, as I will explain near the end of this comment.
Mortazavian has given a very good account for a broad and sophisticated
audience of the aspect of mathematical systems theory that is a somewhat
glorified control theory (in particular, linear control theory), which is, in
turn, what used to be called modern control theory in disguise.
In my view, the territory is much larger, particularly in the conceptual
sense. As an illustration, we can simply refer to the journal Mathematical
Systems Theory (which I founded more than ten years ago) to realize that
mathematical systems theory is more than the theory of realization of linear
systems, observability, and so forth. Mathematical Systems Theory, as a
matter of fact, has published more articles on automata and models of com¬
puter systems than on anything else, and although I do not think that this is a
good balance either, the fact illustrates a broad concern of mathematical
systems theory.
I agree with Mortazavian that a system is a relation among objects. (How
could I disagree with this definition, which I introduced more than ten years
ago?) I also agree with Mortazavian that systems theory deals with models
of specific, concrete systems rather than the systems themselves. However,
I do not quite share Mortazavian’s requirement that systems theory has to
deal with complex interactive systems; in particular, my notion of complex¬
ity seems to be different from Mortazavian’s. For example, dynamic sys¬
tems are certainly complex, since they can have memory that can take them
a finite or infinite number of steps back, so that even a one-input/one-output
567
568 MIHAJLO D. MESAROVIC

system can be quite complex in the sense of being intricate. However, a

system can be complex even without being dynamic. Complexity is due just
as much to the disharmony among subsystems as to the intricacy in describ¬
ing the behavior of a system and its time response. It is indeed peculiar that
the least mathematical of all possible systems research, namely, that of Jay
Forrester, is in full agreement with what is probably the most mathemat¬
ically oriented research in systems theory, namely, that of Rudolf Kalman
and his associates. Both of these schools basically adhere to the philosophy
that systems theory is the theory of dynamics—and therefore, even though
they use different methods, Forrester and Kalman have the same view on
systems theory and what is the main problem in reality: the dynamics of
behavior.
I agree with Mortazavian that systems theory should be mathematical—
that is a position I have held for a long time and still consider valid. How¬
ever, I do not think that the key problems in systems theory are those for
which we can find relatively easy mathematical solutions (again, such as
dealing with linear realization problems). The implication of what I have just
said might become clearer at the end of this discussion.
Among the reservations I have about Mortazavian’s paper is, before all,
the issue of complexity. We cannot identify systems theory with the theory
of models of complex systems and then talk about major findings in systems
theory as they relate to the simplest systems we can imagine, namely, linear
systems described by first-order differential equations. (Even if they are
abstracted and couched in the framework of module theory, they are no
more complex than their original description, that is, linear ordinary differ¬
ential equations.) I would say that systems theory is indeed a theory of
models or, if you like, a mathematical theory of models. But some models
are simple while others are complex; and some models are simple models of
complex systems, while others are complex models of complex systems.
What is the nature of complexity? Here, of course, we would have to ex¬
press a personal view because there is no agreement whatsoever as to how
complexity ought to be defined. My approach to complexity is the following.
A system is complex if it consists of a family of interactive subsystems,
each one of which operates autonomously under normal conditions. (This
definition could be made mathematically precise—see my forthcoming
book, coauthored with Yasuhiko Takahara.) The definition implies that
there are two levels of systems behavior: the subsystem level and the overall
systems level. Thus, my definition of complexity implies a hierarchy. Going
back to Mortazavian’s definition of system as a relation, I would say that
whereas a system is a relation among sets, a complex system is a relation
among systems themselves—that is, the complex system is a relation among
the sets that in turn are systems themselves. This definition, though trivial,
has significant implications in practice.
Just to illustrate this point, let us assume that we have to model an
organization. Modeling an organization is just as legitimate a modeling task
 MATHEMATICAL SYSTEMS THEORY AND INFORMATION SCIENCES 569

as modeling a rotating plane (an example Mortazavian used in his paper). An

organization clearly has three characteristics:

1. It consists of a number of subsystems.

2. The subsystems are decision-making (or goal-seeking); that is, they
can be most appropriately and most efficiently described in terms of
their responses to any stimuli in reference to their pursuance of inter¬
nal goals.
3. They are organized in a hierarchy; that is, there are decision-making
subsystems that can modify, influence, or even control the behavior
of the subsystems’ units (on what are considered as lower levels).

Without going any further, it is quite clear that such a system as an

organization can hardly be described by a set of first-order linear differential
or difference equations or even by their module structure counterparts.
Relations among the subsystems are far too complex for that. There are no
identifiable state variables in the system as a whole, and the whole concept
of state transition is not necessarily directly relevant. Notice that I have said
not relevant (rather than impossible to introduce); concentration on time
transition rather than other aspects of the system’s functioning, such as
conflict, would obscure the basic nature of the system and be a completely
inadequate model of reality. The point 1 am making is that a system that in
practice would be accepted as complex can hardly be described by a set of
first-order (linear or nonlinear) differential equations.
Having said this, I have to admit that although I subscribe strongly to the
notion that systems theory has to be mathematical, there is no mathematical
theory as yet fully developed of the kind of complex systems described in
the preceding example. Of course, I have to refer to our book Theory of
Hierarchical Multi-Level Systems, which has not stimulated as much re¬
search over the last decade as I would have liked to see. [Mesarovic, Macko,
and Takahara, 1970.]
My second reservation regarding Mortazavian’s paper concerns his in¬
adequate emphasis on goal-seeking systems. A system is goal-seeking if its
behavior can be best described with reference to the pursuance of a given
goal. The simplest example is modeling a car moving on a highway. A
Forrester/Kalman system-dynamic approach would describe this in terms of
the car’s acceleration and speed, that is, as a physical system. A goal¬
seeking description would require that the driver inside the car be identified
and the moving point on the line representing the car on a highway be
represented in terms of the strategy the driver uses in steering the vehicle
along the road. A goal-seeking description, therefore, requires a description
of a goal, a description of a strategy (how to pursue this goal), and a descrip¬
tion of the environmental conditions in which this strategy is being pursued.
All of these concepts could be, and indeed have been, defined mathemat-
570 MIHAJLO D. MESAROVIC

ically. However, it is difficult to develop a full-scale theory starting from

such a concept of a system, and this, of course, is the reason why progress in
that direction is disappointingly slow.
My final reservation regarding Mortazavian’s paper is its relegation of
information sciences to Shannon’s and Shannon-type mathematical con¬
structs. Quite clearly, Shannon’s notion of information refers only to the
capacity for transmission. The concept of information is, of course, much
richer. Actually, in my view, information cannot be defined without refer¬
ence to the goal-seeking behavior of a system; therefore, information theory
and systems theory are one and the same.
Elaborating further on this, it is my view that systems theory is not just
mathematics. Systems theory is, indeed, a mathematical theory of models
(not of real systems), but the functioning and behavior of systems and the
importance of various issues with which the theory is concerned are not
questions of mathematics but of the real world. This is why stability might be
a more important consideration than finding a solution for equations. The
crucial aspect of systems theory, and where it differs from mathematics, is
that the importance of the problems with which a theory is designed to deal
(in a precise, and therefore mathematical, manner) is defined outside of the
theory itself. To repeat for emphasis, the importance of the problem comes
from the real world. In the present state of development of systems theory,
the underlying set of concepts for systems theory is from the decision¬
making and information-processing domains in a broad sense. Therefore, I
would actually equate systems theory with information theory in an appro¬
priately wider sense. In other words, to repeat this, too, systems theory goes
beyond glorified control theory. Systems theory is a mathematical theory of
models of real-life systems presented in terms of laws and concepts of deci¬
sion-making and information-processing.
In conclusion, I wish to mention one more thing. The restrictiveness of
Mortazavian’s view of systems theory is perhaps best illustrated in the glori¬
fication of the problem and solution of the realization of linear systems. The
theory of realization of linear systems is not necessarily mathematically very
difficult. In a sense, it is a restatement of dependence in the set of linear
equations. Even conceptually, it is not quite so important as Mortazavian
seems to imply. To come back to the earlier example, let us assume that we
are looking at an organization with many levels and numerous decisionmak¬
ers. Process identification really refers to identifying the levels and deci¬
sionmakers of which the system consists. It is not primarily a matter of
identifying the proper number of state variables, that is, the proper linear
structure in the sense of linear-systems theory. The problem of identifica¬
tion, therefore, is much more difficult than Mortazavian would make it ap¬
pear.
Mathematical systems theory, in the sense I have advocated here, is very
slow in developing. Whether this is in the nature of the problems involved,
their difficulty, or whether it is so because of the scarcity of institutional and
MATHEMATICAL SYSTEMS THEORY AND INFORMATION SCIENCES 571

financial support for the required research or because of a lack of leadership

in the scientific community, the future will show. At any rate, there is a
danger that mathematical systems theory might have a fate somewhat like
that of the attempt by Bertrand Russell to axiomatize mathematics.
Developing such a theory seems the right thing to do, but it may prove not
to be possible. If it has taken us twenty years to come to the realization
theory of linear dynamical systems, then it may take us another fifty to one
hundred years to develop the theory of organizational systems with multi¬
level decision-making. If this is so and mathematical systems theory is to
have an impact, then a somewhat different, more modest approach may be
in order.
 SYSTEM THEORY VERSUS
SYSTEM PHILOSOPHY

Hassan Mortazavian

System theory is at present in a state of rapid development. Progress in

system theory during the last two decades was more significant in the do¬
main of finite-dimensional linear systems (deterministic or stochastic) that
used, for the most part, the highly developed and already available tools of
differential and difference equations, analysis, algebra, geometry, and prob¬
ability theory. This was a great step forward. Many fundamental problems of
system theory, however, still remain unsolved. In a sense, the unsolved
problems are more challenging than the ones already solved. In my view,
there are at least three reasons for this: First, the unsolved problems are
generally more difficult. Second, the choice of the proper mathematics to be
used or developed to formulate and solve these problems is far from trivial.
Last, but not least, some of these problems are not even well understood at
the conceptual level. For example, the concepts of complexity, structure,
coordination, conflict, organization, goal-seeking behavior, information, and
knowledge are poorly understood.
Advances in system theory are likely to be observed in the future in at
least three directions: (1) Further development of the theory of nonlinear
systems, systems with nonsmooth dynamics, as well as infinite-dimensional
systems. (2) Elucidation and mathematical formulation of fundamental
structural aspects of analysis, control, design, coordination, and organiza¬
tion of complex systems. (3) A more unified theory of the somewhat frag¬
mented systemic disciplines, such as control theory, decision theory, auto¬
mata theory, and computer science, with a view to the development of a
higher level theory of analysis, control, and design, involving logic, abstract
algebraic structures, and languages. This is particularly important because
of the rapid development of cheap and flexible computer hardware. To be
honest, at present, we have no satisfactory scientific answer to many ques¬
tions involving the type of concepts just mentioned.
Mesarovic is right that I have committed “a sin of omission.” This was
perhaps inevitable, as I decided to write about those aspects of system
problems that we know something more or less definitive about, that could
573
574 HASSAN MORTAZAVIAN

be explained in an informal way, and that can be directly applied to various

disciplines where such system-theoretic problems arise.
Mesarovic criticizes me on the grounds that, having identified system
theory as the theory of models of complex systems, I then devoted much of
my paper to the simple case of linear systems. I decided to write more about
linear systems for two reasons: (1) The theory of linear systems is well
developed; (2) it has been successfully used in many situations where a
system is linear or can be linearized and thus can serve as a good example of
the type of system-theoretic knowledge that has been applied. But certainly,
system theory is not just “linear control theory.”
I agree with Mesarovic that “a system can be complex even without being
dynamic.” But I do not think that the definition of a complex system as one
that “consists of a family of interactive subsystems, each one of which
operates autonomously under normal conditions” is sufficiently inclusive. A
family of interactive subsystems may be considered as complex even if some
of the subsystems do not behave autonomously. Part of the complexity may
indeed arise from the mode of interaction of the subsystems. Moreover,
Mesarovic’s abstract definition of a complex system as a “relation among
the sets that in turn are systems themselves” is, in my view, not exclusive.
According to this definition, any relation on a family of sets that are them¬
selves relations must be considered a complex system. I think more mathe¬
matical structure must be introduced to avoid the possibility that this rela¬
tion becomes a trivial one.
Despite Mesarovic’s view that I have relegated “information sciences to
Shannon’s and Shannon-type mathematical constructs,” I should like to men¬
tion that my definition of information sciences in my paper (at the end of the
fourth section) was much broader. Shannon’s theory was a theory of “statis¬
tics of highly improbable events” and was called information theory only for
want of a better name. Information theory, in the sense of Shannon, has
nothing to do with information in the semantic sense or even in the ordinary
sense of the term. My discussion of Shannon’s theory was mainly motivated
by my attempt to point out some of these distinctions.
Boulding writes: “I confess I regard mathematics as one of the least
developed of the sciences in terms of its potential. It is very inadequate at
the present time to deal with systems beyond at least the moderately sim¬
ple.” I find myself in agreement with the first part of Boulding’s remark. The
reason is simple: Mathematics is only limited by the limitations of human
imagination. But the inadequacy of present mathematics in dealing with
complex systems is not the fault of mathematicians. Mathematicians were
never truly devoted to understanding complexity. They were devoted to the
search for truth, depth, generality, and simplicity. System theory, however,
deals with complexity, and to do so, we may, indeed, be obliged to develop
new types of mathematics. This new mathematics we may call mathematics
of systems. This does not mean that the existing mathematics cannot be used
to study systems. A fundamental application of mathematics—in particular,
SYSTEM THEORY VERSUS SYSTEM PHILOSOPHY 575

algebra, topology, and geometry—has been to invent an abstract conception

of structure. In its more developed form, mathematics has been and ought to
be used to treat more complex structures called systems.
Langlois writes that: “. . . most of the mathematics of system theory grew
out of electrical engineering, principally circuit theory.” Although many of
the early system theorists were from electrical-engineering backgrounds—a
fact that had a negative effect on the development of system theory, suggest¬
ing the idea that a system is a signal—none of the mathematics of system
theory grew out of electrical engineering. Langlois’s statement, in my view,
would be equally surprising to mathematicians and electrical engineers. I
wonder whether Langlois knows much about the nature of the mathematics
used in system theory. At the present time, apart from calculus and linear
algebra, system theorists use mathematical tools from various branches of
mathematics: algebra, analysis, topology, algebraic geometry, differential
geometry, combinatorial geometry, combinatorics, probability theory, and
statistics and many other fields, none of which grew out of electrical en¬
gineering. It is by no means an exaggeration that a system theorist’s progress
is entirely dependent on how much mathematics he or she knows and how
powerful his or her imagination is in making this mathematics directly appli¬
cable to system problems by discovering isomorphisms between mathemat¬
ical constructs and various types of systems. As Richard Bellman put it,
‘‘mathematicians, as the keepers of abstractions, . . . hold the keys to the
study of all systems.” [Bellman, 1971a.]
Again, Langlois writes: ‘‘Practitioners fall into two identifiable if overlap¬
ping groups: those who wish to leave the cave and those who do not.” And
then he adds that Mortazavian firmly belongs to the second group, while
those who leave the cave are, presumably among other things, ‘‘concerned
with applying mathematical analysis to areas of military, business, and (yes)
social management.” I have many things to say about Langlois’s statement
but confine myself to two issues. (1) The notion of a practitioner who does
not wish to leave the cave and ‘‘whose business is to scrutinize pure forms”
is so evidently self-contradictory that I cannot but call it as meaningless as a
square circle. (2) To leave or not to leave the cave is not the question. The
question is to do honest scientific work. Incidentally, I do not believe in a
rigid distinction between pure and applied mathematics, nor do I believe one
is inferior to the other.
From Langlois’s discussion of my remarks on modeling and the nature of
system theory, it is clear that he has misunderstood me. When I wrote that
‘‘any attempt to apply system theory outside its mathematical context is not
proper,” I simply meant that system theory cannot be applied unless the
general setting of a concrete problem is such that the numerical quantities
that are put into the dynamics of the model or from which a model is to be
constructed are mathematically well defined. In short, system theory must
be applied mathematically.
Commenting on the second approach to modeling discussed in my paper,
576 HASSAN MORTAZAVIAN

that is, modeling based on data not laws, Langlois finds that it “implied a
naive faith in data long ago renounced—with good reason—by the philoso¬
phy of science.” To be sure, observation alone, and thus modeling based on
data alone, can never discover new laws or new theories. New laws can only
be discovered by new theoretical analysis and new experimentation. What I
said about the data-dominated approach to modeling does not amount to a
“naive faith in data.”
Langlois writes that Kalman’s view of the world and to some extent my
own view “serves in the quest to unify all sciences under the banner of
system theory . ...” I have said exactly the opposite in my paper when I
wrote in the beginning of the third section, “system theory does not aim at
unifying all or special types of scientific disciplines.”
Contrary to Langlois’s interpretation of my views, I do not regard the
scientist as a handmaiden or “a kind of graduate research assistant to the
system theorist.” System theory is a science like any other science, and in
science, there is no question of inferiority or superiority.
In a footnote, Langlois tried to clarify the special meaning of the term
observability adopted by mathematical-system theorists. Langlois is right
when he says that the notion of observability has nothing to do “with the
phenomenology of observing a system.” I said this myself in my paper:
“Observability of a system . . . should not be mistaken for the ordinary
process of observation.” However, the rest of Langlois’s footnote is partly
misleading, partly wrong. The point is not that we “know all about a sys¬
tem,” or find out “what the (known) state variables are doing.” The ques¬
tion is whether we can measure the values of the state variables of a given
system at every instant of time within a time interval from the measurement
of its outputs.
I now turn to comments by Richard Mattessich. He warns the reader that
whenever I speak of system theory I limit myself to “the relatively narrow
area of mathematical system theory.” I have argued in my paper that non-
mathematical system theory does not and cannot exist and, therefore, that I
take system theory and mathematical system theory to be identical. The
term mathematical system theory was introduced, I think, for two reasons:
One, to distinguish true system theory, which must be mathematical, from
works that were not mathematical and yet were called system theory by
their producers. Second, to distinguish the part of system-theoretic research
that used more sophisticated mathematical tools from parts that used less
sophisticated and sometimes poor mathematics. I cannot think of any sys¬
tem problem that can be solved without mathematics. If such a problem
exists, it must be a trivial “nonproblem.” I hasten to add two remarks: (1)
System-theoretic ideas did not arise from mathematics, but rather from the
practical necessity of dealing with complex concrete phenomena; (2) it is
possible, and indeed quite probable, in any science to miss the real problem
and fall into the trap of too much mathematical sophistication.
Mattessich criticized me for not having mentioned the works of Ackoff
and Emery, Boulding, Bunge, Churchman, James G. Miller, and Herbert
SYSTEM THEORY VERSUS SYSTEM PHILOSOPHY 577

Simon in my paper. These authors, in my view, have all contributed to what

we may call philosophy of systems—a branch of philosophy dealing with
problems centered around the notion of system.1 Philosophy of systems,
however, must be distinguished from system theory, which is the science of
systems.
Now I shall consider Mattessich’s six specific arguments against my view
of system theory, one by one.

1. Contrary to Mattessich’s view, system theory is not an empirical

science. It is a mathematical theory. System-theoretic propositions
can only be proved or disproved mathematically, but they cannot be
tested empirically.
2. For these reasons, I entirely disagree with Mattessich’s view that
“system theory remains meaningful only as part of one or more or
even all the empirical sciences.” System theory is not part of any
empirical science.
3. The choice of “fitting criteria” is partly a system-theoretic problem
and has both mathematical and empirical aspects. The mathematical
part is a basic concern of system theory, while the empirical part
depends on the specialist’s view of the constraints, specific purposes
of modeling, degree of accuracy required, and so forth. It also raises
certain philosophical problems, which must be dealt with in what I
called philosophy of systems.
4. I entirely agree with Mattessich that the problems dealt with in sys¬
tem theory were hardly bothersome in physics. I do not believe,
however, that mathematicians have ever claimed system theory for
themselves. The fact that system theory is a mathematical theory has
attracted many mathematicians to this field. I do not want to raise the
controversial question whether by virtue of the fact that a system is a
mathematical object and system theory a mathematical theory, sys¬
tem theorists may be considered as mathematicians in some rather
broad sense.
5. Contrary to Mattessich’s claim, I have made no attempt in my paper
“to banish the factual and normative-methodological essence of sys¬
tem theory.”
6. I do not find Mattessich’s remark about the goal orientation of the
systems approach relevant to the content of my paper.

I agree with Mattessich when he writes that “Mortazavian’s suggestion

that ‘the solar system can be studied using only the laws of physics’ would
have been poor advice for Kepler and Newton, who still had to search for
some of those laws.” He has a point. I should not have said so without
qualification.

'The scientific contribution of these authors, however, is by no means limited to this domain.
SECTION 9
 SYSTEMS THEORY,
KNOWLEDGE,
AND THE SOCIAL SCIENCES

Richard N. Langlois

It is not too much of an overstatement, I think, to suggest that systems

theory has not lived up to the great promise it has long been supposed to
hold for the social sciences. Considering the magnitude of the claims, of
course, disappointment was probably inevitable. Systems theory has been
hailed by some as a way of unifying the methodology of all sciences. Others
have intimated that this approach would finally endow the social sciences
with that natural-science-like rigor and precision those disciplines allegedly
lack. And others—more modestly—have looked to systems theory as a way
of combatting the fragmentation and specialization of the sciences.
But systems theory has failed—at least so far—in its various attempts to
bring all the disparate parts of inquiry under the sway of its organizing force.
In fact, it is fairer to say that systems theory has itself succumbed to the
diversity and complexity of modern scientific inquiry. Defining what we
mean by systems theory or the systems approach is virtually impossible
outside the context of a particular discipline, and we might almost say that
there are as many versions of systems theory as there are would-be systems
theorists.
In trying to be all-encompassing, systems theory has become unsys¬
tematic; and in trying to become systematic, it has become narrowly special¬
ized. On the one hand, the attempt to describe at a broad level the system-
theoretic approach to this or that inevitably ends up sounding like a Sears
Roebuck catalog of vaguely connected concepts, models, and definitions.
On the other hand, mathematical system theory1—the most rigorous and

This research was supported in part by the Division of Information Science and Technology of
the National Science Foundation under grant IST-8110537. Note: Portions of this paper ap¬
peared under the title “Systems Theory and the Meaning of Information” in “Perspectives on
Systems Methodology and Information Research,” a supplement to the Journal of the American
Society for Information Science, vol. 33 (November 1982), pp. 395-399.
'See Hassan Mortazavian’s contribution to this volume.

581
582 RICHARD N. LANCLOIS

tightly knit version of the systems approach—has coagulated out of the

more general body of systems ideas to become a well-circumscribed mathe¬
matical specialty, even if one with certain imperialist pretensions.
I say this largely by way of apology—for 1 do intend in this paper to
provide a perspective on systems theory. But because systems theory is, in
fact, so multifarious, my portrayal of it will necessarily be selective, picking
out those pieces that I find useful and letting the rest alone. This seems to me
the only sensible compromise between the encyclopedic and the axiomatic.
My general concern will be with systems theory’s relation to the social
sciences, especially economics. More particularly, though, I will be con¬
cerned with the notions of knowledge and information and the ways in which
systems theory can help illuminate those concepts. Indeed, I am hopeful that
this essay might prove a useful preface to some research I am now beginning
on the connection between the information sciences, broadly construed, and
economic theory.

WHOLES AND PARTS

Perhaps the principal reason systems theory has failed to revolutionize

scientific methodology is that at the broadest and most general level, it has
nothing particularly revolutionary to offer.
At its most philosophical, systems theory is a confrontation with the age-
old problem of the whole and the parts. Systems theorists discovered—or,
rather, rediscovered—complexity. They saw the approach of the physical
sciences, which is supposed to be analytic or reductionistic, as applicable
only to phenomena of “organized simplicity.” In dealing with the increas¬
ingly prevalent problems of “organized complexity,” they argued, we can¬
not understand the parts in isolation from the whole. [Weaver, 1948.] Thus,
we need to study the total system, to see the big picture. These are not new
ideas, of course. In fact, it is striking the extent to which modern systems
theorists—notably proponents of so-called general systems theory—have
tended to retrace a lot of old footsteps in the long-standing debate over the
doctrine of methodological holism. [D. C. Phillips, 1976, esp. pp. 46-67.]
Notwithstanding the holist rhetoric of most systems theorists, however,
systems theory is not at all methodological holism in any strong sense. And
at its best, systems theory is, in fact, a form of intelligent methodological
individualism.
Since this is a somewhat unconventional thing to say, and since the indi¬
vidualism/holism debate is so frequently misunderstood, perhaps a brief
digression is in order.
The conventional wisdom runs something like this. There is a doctrine in
the social sciences called methodological individualism. It is a form of the
analytic or reductionist method, and it therefore holds that knowing the
properties of the parts—the individuals in society—is fully sufficient for
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 583

grasping all there is to know about the whole—society. In other words, the
properties of the whole can be deduced from the properties of the parts; or to
put it in more familiar (and more naive) terms, the whole is just the aggregate
or sum of the parts. This view is to be contrasted with methodological holism
(or, sometimes, collectivism), which insists that wholes possess “emergent”
properties that cannot be derived from the properties of constituent parts.
To the holist, the whole is greater than the sum of its parts.
This is a familiar story. But, as is often the case with familiar stories, it is
almost entirely wrong. There may well be some writers who espouse in
principle this sort of naive methodological individualism. But no one can
actually put such a view into practice. Consider the case of a neoclassical
economist analyzing the effect of a tax suddenly placed on a certain com¬
modity. In proper individualist fashion, he or she will instantly begin to
model the choice problem faced by a representative economic agent, discov¬
ering, as always, that the individual will most likely consume less of the
commodity than before. But this finding about individual demand tells the
economist nothing about the whole—total demand—until he or she adds a
global fact: that total demand is the sum of individual demand. In this (triv¬
ial) case, the whole is just the sum of the parts; but even here, the whole
could not be deduced from the parts, since the relation among them—
addition—is not logically contained in the individual-choice model itself.
Moreover, there is nothing sacred about addition, and methodological indi¬
vidualists are quite willing to specify very different sorts of relations among
the parts. In economics, for example, the aggregate result is very often
exactly the opposite of what we would have expected from considering
individual behavior alone.2
Far from denying the importance of emergent phenomena, the economist
and philosopher Friedrich A. Hayek, widely (and correctly) cited as an
archproponent of methodological individualism, reminds us that the entire
objective of the social sciences is to explain how the behavior of individuals
leads to orderly patterns and institutions that none had consciously
planned—to explain, in other words, the emergent results of individual ac¬
tion. [Hayek, 1979, pp. 146-147.]
In sophisticated discourse, the question of emergent properties is in no
way an issue between individualist and holist. Both agree that social phe¬
nomena must often be considered emergent wholes whose behavior cannot

2In his best-selling introductory textbook, Paul Samuelson (a methodological individualist in

microeconomics, at least) finds it necessary to warn the student against fallacy-of-composition
errors as early as page 11 of the text, where he presents a long list of such fallacies exploded by
economic analysis. [Samuelson, 1980.] Indeed, Kenneth Arrow, Samuelson s fellow Nobel
Laureate, has written that “the notion that through the workings of an entire system effects may
be very different from, and even opposed to, intentions is surely the most important intellectual
contribution that economic thought has made to the general understanding of social processes.”
(Arrow, 1968, p. 376.)
584 RICHARD N. LANCLOIS

be entirely reduced to the behavior of individuals.3 The issue is not whether

we can deduce the nature of the whole from the properties of the parts; the
issue is whether or not we should consult the parts at all. Holism is not the
doctrine that we should study emergent phenomena; holism is the doctrine
that we should somehow study wholes directly without considering the
workings of the parts in a meaningful way.
It should be immediately clear that this position does not follow at all
from a recognition of systemic interactions or emergent phenomena. The
methodological individualist cannot deduce such phenomena from his or her
knowledge of the parts; but—what is often overlooked—the holist is in no
better position to understand such phenomena than is the individualist. The
problem is simply a lack of knowledge about emergent properties, and call¬
ing oneself a holist does not instantly convey that knowledge. [D. C. Phil¬
lips, 1976, pp. 14-15.]
More to the point, adopting the holist stance arguably puts us in an epis¬
temological position inferior to that of the individualist. The holist differs
from the (sophisticated) individualist only in that the former insists on throw¬
ing away useful information. Here lies the real disagreement. The methodo¬
logical individualist holds that the social scientist should always keep in the
closest contact with the level of the parts (the individuals) and utilize fully
whatever knowledge of the parts—however incomplete—he or she can
bring to bear. [Machlup, 1969 and 1979/?; Hayek, 1979, esp. chaps. 4, 6.]
The best way to appreciate this may be by analogy with literary criticism.
No one would deny that a work of literature is more than the sum of the
words and sentences it comprises. Yet the modern critic insists on studying
these words and sentences carefully. Similarly, methodological individ¬
ualism in social science is nothing more than an insistence on “sticking to
the text,” whereas holism is a license to engage in that most heinous of “lit-
crit” solecisms, “reading in.”
Although naive methodological individualism is probably impossible in
practice, holism in this sense is not. Social science is rife with holistic
formulations in which hypostatized concepts like society, the capitalist
class, or the public interest take on operational significance in and of them¬
selves—formulations in which, as Jacques Barzun said of Marx, the terms of
reference become “entities stuffed with people but not composed of them.”
(Barzun, 1958, p. 182.) One result, ironically enough, is that holists are much
more prone to fallacy-of-composition errors than are methodological indi¬
vidualists. For if we throw away information about the parts, we are inclined

3 “The overall order of actions in a group is in two respects more than the totality of regularities
observable in the actions of the individuals and cannot be wholly reduced to them. It is not so
only in the trivial sense in which a whole is more than the mere sum of its parts but presupposes
also that these elements are related to each other in a particular manner. It is more also because
the existence of those relations which are essential for the existence of the whole cannot be
accounted for wholly by the interaction of the parts but only in their interaction with an outside
world both of the individual parts and the whole.” (Hayek, 1967, pp. 70-71.)
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 585

directly to read in a logic of operation for the whole from some other source;
and the logic of operation nearest at hand is that of the individual human.
[Hayek, 1979, p. 101; Langlois, 1981, chap. 3.] The notion that, for example,
a society permitting economic self-interest is therefore a greedy society is a
bit of naive individualism characteristic of holists far more than of meth¬
odological individualists.
In order to make a plausible case for holism, we have to argue that less
information is somehow better than more. This is no easy task. We might for
instance invoke the Gestalt and assert that attention to the parts destroys
our understanding of the whole. But this is to confuse perception and under¬
standing. And, as my analogy with literary criticism was meant to suggest,
the sophisticated methodological individualist has no compunction against
stepping back to survey the Gestalt—so long as the (epistemologically more
accessible) parts are also carefully analyzed. Historically, holists have taken
a rather different (if not ultimately unrelated) line of attack. Hegel and his
followers argued that the parts could not be studied in isolation because the
parts acquire their very nature from their relation to the whole, which nature
is necessarily altered if the parts are considered apart from that whole. But if
taken at all seriously, this argument leads to logical absurdity, and it falls
quickly apart when translated from the quasi-mysticism of essentialist
rhetoric into a modern nominalist vocabulary. [D. C. Phillips, 1976, pp. 5-
20.]
Proponents of general systems theory have unknowingly reinvented and
invoked this Hegelian formulation in discussing the holism of systems
theory. [Ibid., esp. pp. 45-50.] But most systems theorists—both mathema¬
ticians and practitioners alike—conceive of systems theory in a way antago¬
nistic to this Hegelian view. They are certainly inclined to recite holist cant
about “phenomena that depend on the conditions of the environment in
which they exist, interact with this environment, and thus cannot be prop¬
erly studied in isolation.” (Mortazavian in this volume.) But all they mean
by this is that the behavior of the whole cannot be understood without
knowledge of the relations among the parts. The parts are conceived of as
logically distinct elements of a mathematical set; those elements exist and
are fully defined independently of any relations that might be specified. A
system is just the set of parts plus a set of relations among the parts.4 This is
a formulation that would trouble a serious holist far more than it would a
methodological individualist.
The philosopher Mario Bunge has recently transformed this set-theoretic
definition of a system into a methodological position—systemism—that
“combines the desirable features of individualism and holism.” (Bunge,
1979a, p. 13.) To Bunge, a society (which he finds it necessary to call a) may
be represented as the ordered pair (5, R), where S is the set of individuals in

4Cf. Mihajlo Mesarovid’s “implicit (syntactical) definition” of a system in Mesarovic [1964a,

p. 7],
586 RICHARD N. LANGLOIS

the society, and R is the set of relations among them. That systemism does in
many ways combine the best of both worlds is, I believe, an entirely unob¬
jectionable assertion. What is not true is that systemism somehow repre¬
sents a new methodological alternative: The basic ideas of what Bunge calls
systemism are essentially identical to what sophisticated methodological
individualists have believed all along.5
It is important to notice that while Bunge’s definition of society qua
system is consistent with the general mathematical definition of a system
[Mesarovic, 1964a, p. 7], not all applications of systems theory are in accor¬
dance with the tenets of systemism understood as intelligent methodological
individualism. The “parts” in Bunge’s formulation are individuals in soci¬
ety, but this is by no means always the case in social science. The elements
in a typical systems model are aggregate variables of one sort or another—
dollars, commodity levels, energy use—that are represented as impinging
directly on one another with no reference whatever to the existence of
human beings. A good many systems models in the social sciences must,
therefore, be classified as instances of naive holism; and while such models
may often prove interesting and illuminating, they cannot—as is often
claimed—serve as a general foundation for economics and other social sci¬
ences. This is a point to which I will return.
Perhaps I should apologize for the length of this digression. But, in a
sense, perhaps this has not been a digression at all. After all, philosophical
issues lie nearer to the surface in systems theory than they do in most
intellectual endeavors, even if they are not for that reason more often per¬
ceived. More importantly, though, there is a sense in which it is methodolog¬
ical individualism that lies behind the systems-referential view of knowledge
and information that I now wish to present.

SYSTEMS AND THE MEANING OF INFORMATION

There is a strain of thought on matters informational that I find somewhat

disturbing. It is what I think of as the “oil-flow” model of information, with
its attendant “oil-tank” model of knowledge. According to this (implicit)
view, information is some sort of undifferentiated fluid that will course
through the computers and telecommunications devices of the coming age
much as oil now flows through a network of pipes; and the measure of our

5 Bunge does admit that Hayek and other methodological individualists recognize “the reality of
social relations” (p. 17), but he continues to paint them as naive individualists. “The individ¬
ualist might not wish to dispute the systemist’s thesis,” says Bunge, “but, if he is consistent, he
must insist that the structure of R is somehow ‘contained’ in, or deducible from, the properties
of the individual members of society” (p. 19). But, as we saw (cf. again footnote 3), Hayek for
one does not so insist, and I cannot see why he is therefore inconsistent in any way. Bunge is
simply mistaken in his characterization of the individualist position. [Bunge, 1979a.]
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 587

knowledge in this world will be the amount of “info-fluid” we have managed

to store up.6
This model of knowledge and information has some mutually reinforcing
affinities with information theory in the well-known Shannon-Weaver sense,
which developed for purposes of communications engineering a quantitative
measure of something called information. [Shannon and Weaver, 1949.]
Communications theorists are themselves quick to deny the connection be¬
tween their concept of information and the term’s everyday meaning, distin¬
guishing not only between semantic and nonsemantic information but also
between the concept of information itself and the notion of amount of infor¬
mation. The logic of systems can help illuminate these distinctions, I be¬
lieve, and may even prove able to offer a conceptual model for information
and knowledge alternative to the pipe and tank.
It is tempting to think that the distinction between the communications
theorist’s information and the more general sense of the term is at base a
matter of mechanism versus antimechanism (the quantification of informa¬
tion is something appropriate to machines—to computers and switchboards,
wires and transmitters), and that the very different character of information
and knowledge in everyday experience arises from the distinctly non-
mechanistic nature of the human system. There is more than a grain of truth
in this, and I will later suggest that such a dichotomy between mechanistic
and nonmechanistic systems is indeed in order. But, in the end, this view
gives us only part of the picture. To a large extent, I believe, the real issues
of information and knowledge actually cut across the mechanism/
antimechanism dichotomy.7
To see what this means, let us look at how even the most mechanistic
sorts of systems models use a concept of information.
Systems theorists often distinguish between terminal or causal systems
and goal-seeking systems. [Mesarovic, 1964a, p. 21; 1962, p. 13.] The former
react to their environment according, as it were, to the logic of proximate
cause: The environment affects the system’s inputs, which, in turn, affect
the behavior of the system in a strictly preprogrammed fashion. By contrast,
goal-seeking systems operate according to something nearer the final cause.
In this case, there are “certain invariable aspects of the system which reflect
its goal” (Mesarovic, 1962, p. 13); a teleological subsystem receives the
system’s inputs and guides its behavior in light of the goal. The distinction

6This oil-flow model is not without implications. More than one writer has suggested that
information policy be predicated on the inevitable dependency of future society on this "info-
substance” just as energy policy is supposed to deal with our dependence on oil. As a conse¬
quence, we should worry about the availability of “info-fluid" to disadvantaged groups like
Chicanos much as we recently used to fret about the availability of heating oil to poor New
Englanders.
7This is precisely the theme pursued by the British physicist Donald MacKay. [MacKay, 1969.]
The next section of this paper will draw heavily on his ideas.
588 RICHARD N. LANCLOIS

can perhaps best be illustrated using the familiar stimulus/response (S/R)

model from behaviorist psychology. In the causal approach, we posit some
direct mapping of the stimulus to the response; each stimulus directly causes
a particular response. In the goal-seeking approach, we assume a slightly
more sophisticated version of S/R in which an intermediate processing stage
comes between stimulus and response.8 The stimulus, of course, is the infor¬
mation with which we are concerned.
As a kind of objective correlative, we could think of a stimulus as involv¬
ing a (weak) form of energy—a small electric current, a pattern of light, a
sound wave—which elicits as response the release of another form of energy
(often stronger or different in character).
In both causal and goal-seeking models, we can talk about the meaning of
a piece of information. In the former case, the meaning of a signal is the
response it elicits. In the latter, response is also the ultimate criterion of
meaning, even if we cannot necessarily understand the meaning of a signal
without first knowing the goal that the system is pursuing.
Now, it is important to notice that not all signals will be equally meaning¬
ful. A signal of the wrong form—the wrong voltage, frequency, or code, for
example—cannot be understood by the system. At the same time, not all
meaningful signals will have the same meaning. A small voltage may elicit a
smaller response than—and thus have a meaning different from—that of a
larger voltage. A message coded ABCD may have implications for achieving
a system’s goal that are very different from those of a message coded DCBA.
And this provides a clue to the distinction between semantic information and
the nonsemantic information of communications theory. The latter is con¬
cerned only with the extent to which a message is within the set of meaning¬
ful messages; it is not at all concerned with the message’s implications for
the system. For example, the messages ABCD and DCBA—which contain
the same characters—would typically have identical information content
in the communications-theory sense, even if they had very different mean¬
ings from the system’s point of view.
To illustrate this, let me draw on an example from the (normative) theory
of economic decision-making. Here we are dealing with a goal-directed sys¬
tem, one in which the “teleological subsystem’’ takes the form of a decision¬
maker who maximizes profit (or, as we shall wish to introduce stochastic
elements, who maximizes so-called expected profit, which is, in effect, profit
weighted by probability).
Let us suppose that the decisionmaker is a farmer and that he has certain
“decision variables’’ under his control, for example, the number of acres of
wheat he can choose to plant. There are also variables—say, the weather—
that are not within his grasp. The profit experienced by the decisionmaker
depends on both the weather and the number of acres planted. If the de-

8Mesarovic holds—correctly, 1 believe—that any mathematical system can be represented

either way. A causal system can be modeled as if it were pursuing a goal; and a goal-directed
system can be reduced to a direct mapping of inputs to outputs. [Mesarovic, 1964a, p. 22.]
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 589

cisionmaker knows the weather—if he knows how much rainfall there will
be—he can easily determine the acreage appropriate for maximum profit. If
he is uncertain about the level of rainfall, he can articulate a probability
distribution for that variable, which allows him to select the acreage that
maximizes expected profit. But the decisionmaker who can anticipate the
level of rainfall perfectly will very likely elect to plant an amount of acreage
different from that selected by the farmer who is uncertain about the
weather; more to the point, the better informed decisionmaker will realize a
higher profit than his ill-informed counterpart.
If the decisionmaker who is uncertain can obtain some kind of informa¬
tion about the weather—divine revelation, perhaps, or the weather bureau’s
probability distribution for rainfall—then he can improve his acreage deci¬
sion. But if this information is costly, the decision whether to acquire it can
be represented as yet another decision-system of the same sort as the origi¬
nal acreage decision. In general, our profit-maximizing farmer should wish
to acquire the information if its expected value exceeds its cost. [See, for
example, Howard, 1966.]
This expected value of information is the interesting quantity. In order to
examine it more closely, let us simplify our example even further. Suppose
there are only two possibilities, heavy rain and light rain. The farmer must
optimize his acreage planted in light of whatever information he might have
about which of these two possibilities will occur; for example, he might
consider the amount of rain experienced in past years or various freely
available predictions by the Weather Bureau or the Department of Agricul¬
ture. On the basis of this imperfect information, the farmer estimates a
probability of, say, 0.7 for heavy rain and 0.3 for light rain. Now, suppose
that a fully reliable clairvoyant stands willing (for a stiff fee) to disclose
which of the two alternatives nature actually has in store. How much is this
revelation worth to our profit-maximizing decisionmaker? We have to con¬
sider how the new information would affect the farmer’s acreage choice and,
thus, his profits. If the clairvoyant says “heavy rain,” he can optimize the
acreage in a way that increases profits over the expected level. If the clair¬
voyant sends the message “light rain,” he can also optimize acreage—in a
different direction—to increase profit over the expected profit. In deciding
whether to buy the clairvoyant’s information, the farmer must decide if the
increase in profit will be enough to justify the clairvoyant’s fee—and he must
decide before he knows which message the clairvoyant will send. So, the
decision whether or not to buy the information must be based on the farm¬
er’s prior probability assessment of what the clairvoyant will say. Since the
clairvoyant is merely revealing what nature will do, the farmer’s weather
forecast or other prior information is exactly as relevant to predicting the
clairvoyant’s message as to predicting nature’s response—since the former
is nothing but the latter moved forward in time.
Thus, the farmer must use his original probability assessment (0.7 chance
of heavy rain; 0.3 chance light rain) in deciding whether to buy the informa¬
tion. And the expected value of perfect information is thus the sum of two
590 RICHARD N. LANGLOIS

magnitudes: the first is the increase in profits from adjusting to heavy rain if
there will, in fact, be heavy rain, multiplied by 0.7, the probability that there
will be (and, therefore, that the clairvoyant will say) heavy rain; the second
is the increase in profits from optimizing for light rain given that there will, in
fact, be light rain, times the probability that there will be (and, therefore,
that the clairvoyant will say) light rain.9 In other words, the expected value
of perfect information (EVPI) is the sum of the value to the system of a set of
possible messages weighted by the probability of occurrence of each mes¬
sage.
By contrast, the information measure of communications theory is oblivi¬
ous to the value a message holds for the system that receives it: In analogy
with the entropy measure of thermodynamics, the information content of the
message heavy rain is, to the communications theorist, proportional only to
the logarithm of the probability of that message being received.10 Although
the value measure and the entropy measure can move in the same direction,
there is no general reason why this should be so.11

9The length of this sentence suggests that there are times when mathematics has its expository
(or at least space-saving) advantages. Let El(a|w) be the (realized) profit from planting a acres
under weather conditions w. The farmer’s original problem is

max lpU(a\w = H) + (1 — p)It(a|M' = L)],

where H indicates heavy rain, L light rain, and p is the farmer’s prior probability assessment on
heavy rain. (The probability of light rain has to be (1 - p) since there are only two possibilities.)
Suppose that a* is the value of a—the number of acres—that maximizes the quantity in
brackets. Then,

EVPI = p[n(a%\H) - ri(a*|//)] + (1 - p)[U(a*L\L) - Il(a*|Z.)],

where p is the farmer’s assessed probability that the clairvoyant will call for heavy rain (which is
necessarily identical to his original assessment of the probability of heavy rain), a*H is the
optimal choice of acreage under conditions of heavy rain, and a*L is the optimal acreage under
light rain conditions.
10In the notation of footnote 9, the (nonsemantic) information content of the message heavy
rain is proportional to -log2p. The information content or entropy of the light-rain/heavy-rain
information system is the information content of each possible message weighted by its proba¬
bility,

H = ~[plog2p + (1 - p)log2(l - p)].

“In our now familiar example, it happens that the two measures are closely related: Informa¬
tion content is high when value is high and vice versa. (The reason is that a high-entropy [low
“info-content”] message is one with a high probability. If a farmer anticipated heavy rain with a
high probability, then a* is likely to be already near a*H, and [II(a^|//) - II(a*|//)] is low.
Similarly, the more unexpected message would have both a higher information content and a
higher value, since a* would be less close to a*. I believe this monotonicity can be shown
rigorously to hold if the profit function is concave.) But such a connection is merely fortuitous.
Consider an industrial research laboratory, for example, in which one experiment has a very
high “info-content” (e.g., the experiment has a 50-50 chance of resulting in the message suc¬
cess) but low profit implications for the company, while a second experiment has a low “info-
content” (e.g., a 90 per cent chance of success) but big profit implications. The research
manager who used entropy as a decision-making criterion would be sorely misguided.
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 591

MEANING AND STRUCTURE

Stimulus/response systems—what we could call cybernetic systems in a

broad sense12—are mechanistic systems in that action alone matters. It is
not so much that action—response—has taken the place of meaning; rather,
meaning itself is defined solely in terms of the action released or controlled
by the action in question.13 In the case of our agricultural decision-system,
for example, the meaning of a message about the weather is the acreage the
decisionmaker chooses to plant. Now, because this is a goal-directed sys¬
tem, we can interpret the relation between message and action in light of the
goal, and since the goal is a simple quantitative one, we can even measure
the value of the message for the goal’s achievement.14 Nonetheless, all
meaning is ultimately reflected in the system’s behavior, in its output.
Yet, if we look at the question of meaning in a slightly different way, we
may be able to generalize beyond simple stimulus/response systems. An¬
other way to say that the meaning of a message within a cybernetic system
arises from the action that message brings about is to say that the meaning is
defined by the system itself. What makes the message rain will be heavy
comprehensible to our decision-system—and the message tea tonight at
eight meaningless to it—is that, by virtue in this case of its very structure,
the system is ready to respond to the one and not the other. Furthermore,
notions of information and meaning seem as applicable to complex and
arguably nonmechanistic systems (such as brains) as to simple cybernetic
systems.
In general, then, we might follow Donald MacKay in speaking of a sys¬
tem’s structure as defining “conditional states of readiness.” It is the overall
configuration—not any particular response in isolation—that determines the
meaning of a message.

It isn’t until we consider the range of other states of readiness, that might have
been considered but weren’t, that the notion of meaning comes into its own. A

12By a cybernetic system in a broad sense, I mean any system that is concerned with informa¬
tion and control. In a narrower sense, cybernetics is concerned specifically with feedback
systems, where a monitoring signal is sent from output back to input in order to control that
output and bring the system into an equilibrium condition called homeostasis. The locus
classicus here is Norbert Wiener’s Cybernetics. [Wiener, 1948 and 1961.]
13 “Depuis longtemps, le pragmatisme et le behaviourisme ont appris aux psychologues a mettre
Taccent sur Taction plutot que sur la conscience. La cybemetique adopte rigoureusement ce
point de vue: le sens, la conscience dans l’information, n’a rien d’essentiel; ou plus exactement,
le sens d’une information n’est rien d’autre que Tensemble des actions qu elle declenche et
controle.” (Ruyer, 1954.)
14Here, I think, we need to make the distinction between the (1) meaning, (2) meaningfulness,
and (3) value of a message. A message to the decisionmaker reading keep doing what you’re
already doing is fully meaningful even though it does not entail a different action or necessarily
result in a level of goal achievement higher than would have occurred in its absence. (Of course,
a strict behaviorist would be unable to distinguish a meaningless message from a meaningful
message to maintain the status quo, but this is not a problem in a goal-directed model.)
592 RICHARD N. LANCLOIS

change in meaning implies a different selection from the range of states of

readiness. A meaningless message is one that makes no selection from the
range. An ambiguous message is one that could make more than one selection.
(MacKay, 1969, p. 24; emphasis original.)

MacKay offers the metaphor of a railroad switching yard in which the

configuration of tracks and switches stands ready to direct the trains passing
through it. By sending the right electronic signal—or, in older yards, by
inserting the correct key in a switch box—we can rearrange the configura¬
tion of tracks. The meaningfulness of a message thus depends on its form—
the shape of the key. And that meaning consists of the change the message
effects in the arrangement of the yard, the selection it makes from the set of
all possible configurations. (Notice that this example is somewhat less be¬
havioristic than that of the decision-making farmer, in that the reception of a
meaningful message does not, in this case, imply action. Although the selec¬
tion operation implies potential action—the shunting of a train in one direc¬
tion instead of another—it is meaningful even in the absence of action; it is a
property of the system itself.) This view of information and meaning seems
to me exceedingly suggestive and points to some implications we can gener¬
alize without, I think, much embarrassment.
The first implication is that meaning must always be defined in terms of
the system—or person—receiving the signal. Meaninglessness, MacKay
notes, “is a relative concept, and a precise definition of meaning would be
useless unless it automatically reminded us of this.” (MacKay, 1969, p. 86.)
The meaning of a message to a cybernetic system—a robot, a spacecraft, an
economic decision-system—depends on the system’s structure and the mes¬
sage’s form; and, while the human system is of a very different order, it
remains that meaning cannot be defined independent of that system—the
apprehending mind. “Meaning is always meaning to someone.” (MacKay,
1969, p. 36; emphasis original.)15 This may sound reasonable, but it is not an
entirely noncontroversial view, for it stands in opposition to numerous at¬
tempts to define meaning entirely in extrapersonal empirical and logical
terms.16
A correlative implication is that—perhaps surprisingly—this view tends
to blur rather than sharpen the distinction between semantic and nonseman-
tic information. If we speak of information as involving a selection operation
on the states of readiness of a system, then we can speak of the selective
information-content of a signal as somehow measuring the extent or mag-

lsAnd here, in an important sense, is where methodological individualism comes back into the
picture.
"’This is true of Wittgenstein in the Tractcitus logico-philosophicus and also of the later logical
empiricists. [Wittgenstein, 1922; Carnap, 1950; Bar-Hillel and Carnap, 19536; Bar-Hillel, 1964.]
I should also note that it was this attempt to eliminate the personal and (contra Mortazavian in
this volume) not the use of subjective probability that was the ultimate problem with the
positivist approach. It was a problem of too little subjectivism, not too much.
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 593

nitude of the selection operation performed. The expected value of informa¬

tion developed in the agricultural example is precisely this sort of measure.
How does it differ from the entropy measure of communications theory?

The communications engineer measures the selective information-content of

. . . signals, not in terms of the selective operation performed by the symbol on
the ensemble of states of readiness of the . . . receiver, but in terms of the
selective operation performed by the signal on the ensemble of signals. The
symbols are represented in this ensemble in the proportions in which they
normally occur, the most frequently used occupying the largest space and
being most easily selected. (MacKay, 1969, p. 75; emphasis added.)

Thus, the entropy measure of communication theory is not so much a mea¬

sure of nonsemantic information content as it is a measure of the semantic
(i.e., selective) information content of operations on one particular system,
albeit a system different from the one actually receiving the signal; it is a
value-of-information measure for a system in which value and probability
(or, rather, improbability) are identical.17
Another, more significant, aspect of this view—which we might call the
system-referential view of information—is, it seems to me, that it argues
strongly against the oil-flow model. Information is not homogeneous; mean¬
ing is a matter of form not of amount; and the value or significance of a
message depends as much on the preexisting form of the receiver as on the
message itself. More to the point, this view suggests that information is
stored as knowledge in a system not as oil is stored in a tank, but by virtue of
the change that information makes in the very organization of the system
itself. In a fundamental sense, knowledge and organization are identical.
I will postpone some of the more resonant implications of this last, rather
provocative, statement. For the moment, let me try to illustrate what such a
connection between knowledge and organization might mean.
We are all accustomed to thinking of memory as a function that takes
place in some isolated part of a system—a computer memory bank or a
human brain, for instance. But information is not stored in those places like
relics in an attic; information submitted to a system’s memory changes the
organization of that subsystem: The arrangement of magnetic elements in a

l7We can see this clearly by comparing the equations in footnotes 9 and 10. For the value-of-
information measure, we had

EVPI = p\Yl{w = H) + (1 - p)Ml[w = L),

where Alt is shorthand for the expression in brackets in footnote 9. For the entropy measure,
we had

H = —[p\og2p + (1 - p)log2(l - p)].

In the first case, the probabilities weight terms that measure the effect of a message on the
system; in the second case, the probabilities weight terms referring only to the selection of a
message from the set of messages.
594 RICHARD N. LANGLOIS

core memory, for example, is changed when data are stored. Furthermore,
the memory bank is not the only locus of memory in the system. In an
important sense, the entire organization of the system—hardware plus soft¬
ware, mind and body—contains functional knowledge to guide behavior. I
do not think this is a new or particularly controversial way of putting things.
At the risk of a charge of idiosyncrasy, let me go beyond this to suggest a
distinction between structural information and parametric information. The
former is information that operates on—that changes the basic structure of
the system; the latter is information that operates on parameters of the
system—on elements that adjust or calibrate the workings of the system
within the dictates of, but without altering, its underlying structure.
I apologize if this all sounds a little vague. But the nature of form and
structure is a problem that has animated philosophy since the Presocratics,
and I make no pretense of trying to solve it. At an intuitive level, the notion
of a system’s structure is fairly clear: A system has various fixed (or rela¬
tively slow changing) attributes that define its form, that set the scope,
ground rules, and boundary conditions for the system’s more variable as¬
pects. (This may have a physical correlative in hardware as distinct from
software, but it need not; even within the software of a computer program,
the mental representations of a human mind, or indeed any system in the
abstract, we can speak of an underlying structure.) If we restrict ourselves to
the mathematical realm, we can make the notions of structural and
parametric information more precise, if in the end perhaps no less
metaphoric.18
The agricultural decision-system again provides a clear example. Here,
the model’s structure lies in the form of the profit function; that function
specifies a parameter, weather conditions, that can be altered by reception
of an appropriate message. Our decisionmaker is able to obtain only
parametric information and thus to gain only parametric knowledge from a
signal. As the problem is formulated, no signal can change his profit function
or any of the basic givens of the situation he faces.
In the modern mathematical “economics of information [Hirshleifer and
Riley, 1979], the focus is exclusively on parametric information of this sort.
For a long time, the mainstream of economics had concentrated on “perfect-
information” models, where the decisionmakers were portrayed as having
full knowledge of all aspects of the decision-situations they faced. In newer

18Mesarovid offers a definition of systems structure that, as best I understand it, is consistent
with what I have in mind. The organization of a system consists in the systems relation that
maps one set of parts into another. This relation R “can be considered as defined by an abstract
relation and the specific values of the unspecified constituents, the so-called relational con¬
stituents; R = {T, £}, where T = systems structure, £ = set of relational constituents.”
(Mesarovid, 1964a, p. 10.) As Mesarovid further suggests (in Equation 9), these relational
constituents can take the form of parameters of the system. Thus, my distinction between
structural and parametric seems in accord with his view. Also see Mortazavian’s paper in this
volume.
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 595

models, the decisionmakers retain full structural knowledge of the problems

they face; but now there are certain key parameters—such as weather con¬
ditions—obscured from their vision. As we might guess, the myopic de¬
cisionmakers invariably choose less efficiently than their better informed
counterparts, a situation that has proven a boon to the legions of modern
economists who take pleasure in identifying causes of what they term mar¬
ket failure much as nineteenth-century paleontologists delighted in unearth¬
ing new species in fossil. Meanwhile, the importance of imperfect structural
knowledge—with its very different economic implications—has escaped
widespread attention.

KNOWLEDGE, STRUCTURE, AND ECONOMICS

It is fair to conclude, I think, that systems theory and the theory of knowl¬
edge and information (broadly defined) must ultimately be related in a funda¬
mental way. Both are concerned with form and organization. And it is little
wonder that communications theory, with its entropylike measure of infor¬
mation content, has held a singular fascination for systems theorists. An
organized entity is a nonrandom entity, one whose organization is unex¬
pected in some sense, and it is unexpectedness—negative entropy—that
communications theory measures.
Although, as we saw, the derivation of a scalar measure of organization is
possible only in the very special case with which communications theory is
concerned, it nonetheless remains that the knowledge content of a system
is closely bound up with that system’s organization—with its structure.
We have already seen two logics of organization identified by systems
theorists: causal systems and goal-directed systems. I would now like to
suggest that these two are not the whole story.
A causal system is an instance of what we might call a “mechanical”
structure. The movements of the parts are causally related to one another
within the dictates of a fixed structure. The centuries-old example is the
mechanical clockwork. Each gear moves by virtue of the force impressed on
it by a previous gear; and each carries out its function within the pattern
ordained by the designer. Once cut loose from its creator, the mechanical
system cannot increase its level of organization. Indeed, any change in
structure (other than those effected by the ministrations of the designer)
must lower the level of system organization, thus increasing entropy in some
sense.
Information and control are closely related concepts in systems theory. In
a strictly causal system, the only way to change behavior is by reprogram¬
ming the system; often, this can be accomplished by adjusting various con¬
trol variables to modify the system’s structure—much as we manipulate the
steering wheel and foot pedals to alter the behavior of an automobile. This is
596 RICHARD N. LANCLOIS

called open-loop control, in that the controlling information comes entirely

from outside the boundaries of the system.
The mechanical-control model has not been absent from economics. The
various models of stabilization policy, now happily somewhat out of fashion,
have long portrayed the economy as a system of aggregate variables (na¬
tional income, consumption, investment, etc.), which the government—
viewed as an entirely exogenous entity—could regulate by suitable manipu¬
lation of its control variables, for example, government expenditures.
As we have seen, goal-directed systems (or cybernetic systems, in the
narrow sense of the term) work somewhat differently. These systems also
have a fixed structure, but now there are certain manipulable variables that
can be altered by information generated within the system itself. The system
possesses a goal; and, by means of an information-feedback loop, it is able to
compare its situation with that goal and make appropriate adjustments to¬
ward it. Among control theorists, this is called closed-loop control.19
I have already suggested that there is a definite congruence between
causal systems and goal-directed ones, and while perhaps less horological
than the former, the latter are not necessarily less mechanical in the
everyday sense of the term. (An ordinary mechanical thermostat is a cyber¬
netic device.) But there remains a tendency to see cybernetic systems as
somehow more organic than causal systems. This is, 1 believe, largely be¬
cause many bodily systems also operate in a cybernetic fashion.-0
Rather than settling at a mechanical equilibrium point—like a clock run¬
ning down or gas molecules coming to terms with the surrounding tempera¬
ture—a cybernetic system achieves a condition of dynamic balance called
homeostasis. Like a mechanical system, though, a cybernetic system in
homeostasis is struggling to maintain a fixed level of organization. Any
change in the system’s state is, therefore, potentially dangerous, and pro¬
longed changes represent worrisome imbalances that threaten to destroy the
system. Indeed, many who take cybernetics seriously as a general meta¬
physics of order soon develop an attitude I tend to call thermodynamic
Manichaeism—the implicit belief that every trend, every economic or social

191 should also mention here the branch of systems theory called optimal-control theory. The
optimal-control theorist seeks the best way to control a system in order to achieve some goal.
For example, we might wish to calculate that trajectory of a rocket between two points that
minimizes flight time, fuel consumption, or some other objective. (And, as a matter of fact, it
was precisely such aerospace problems that formed much of the early subject matter of this
theory.) Mathematically, the optimal path is found using the calculus of variations (and some¬
thing called the Pontryagin Maximum Principle) or the (closely related) technique of dynamic
programming. Once found, this optimal trajectory can be imposed directly by open-loop control
or implemented through a feedback system to create goal direction and closed-loop control.
[See generally Bryson and Ho, 1968.] Optimal-control theory has found its way into economics,
especially in economic-growth models (not unrelated to the simple stabilization policy models
already mentioned) [Burmeister and Dobell, 1970, chapter 11] and in the theory of resource
extraction [Sweeney, 1977]. See also Kamien and Schwartz [1981] and Aoki [1976].
20Human biology was, in fact, an early inspiration for cybernetics. [Cannon, 1939.]
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 597

innovation is a potential victory for the forces of entropy and disorder over
the forces of homeostasis. [Langlois, 1981, chap. 2, esp. pp. 73-82.] And it is
thus probably no surprise that systems theorists [Mesarovic and Pestel,
1974] and cybernetically oriented biologists [Hardin, 1977] are prominent
among the new Malthusians who see dangerous social and ecological imbal¬
ances in the world’s future.
In economics, there has been some effort afoot, especially by so-called
post-Keynesians, to close the loop on the macroaggregate models of eco¬
nomic stabilization. These economists, as one observer correctly notes,
“. . . might wish to replace the Newtonian-clockwork model by something
they call a ‘cybernetic’ model, which may be an improvement (if it could
ever be devised), but a shift from mechanical statics to sophisticated
mechanical dynamics is no radical conceptual revolution.” (Kristol, 1981, p.
212.)
What is the alternative to the mechanical causal and cybernetic models?
The answer is best found by leaving the level of aggregate dynamics and
returning to consideration of the parts—the human agents. A human being—
at least in part or at times—is a goal-directed system. This is the basis of the
ideal type—homo economicus—that underlies much of mainstream eco¬
nomic modeling. Yet everyone this side of B. F. Skinner recognizes that the
flesh-and-blood human being does not operate with the stimulus/response
compulsiveness of a simple goal-directed cybernetic system. There is some¬
thing more, or perhaps different, at work.
One major school of thought holds, at least implicitly, that differences
between the human mind and a mechanical cybernetic system are ones of
degree rather than of kind. The mind is an immensely complex thing, and if
we could only construct a cybernetic system—that is to say, a digital com¬
puter, the apotheosis of the mechanical cybernetic system—with sufficient
complexity, we could largely replicate much of what the mind can do. Ad¬
herents to this view take heart from the old saying attributed to Marx that
quantitative change allowed to go on long enough inevitably becomes qual¬
itative change.
There are dissenters from this viewpoint, of course. Most notable among
these is Hubert Dreyfus, whose analysis continues to stir controversy in the
field of artificial intelligence. [Dreyfus, 1979.] In a significant sense, Drey¬
fus’s thesis rests on the system-referential view of knowledge already ar¬
ticulated. Human knowledge, he argues, is very much tied to the biologically
and culturally evolved structure of the human organism. What is meaningful
to a human is meaningful only in reference to that structure, and meaning
cannot be reduced to the system of explicit, extrapersonal, context-free
statements that a computer, by virtue of its own structure, must employ. It is
for this reason, Dreyfus argues, that no artificial intelligence program has yet
been (or could be) written in which the important functions of discriminating
meaning and significance are not preprogrammed by the human designer.
However this be resolved, it is clear that the human mind operates differ-
598 RICHARD N. LANGLOIS

ently from simple cybernetic systems. The human mind is an example of a

system in which information can result not merely in parametric but in
structural knowledge. Unlike our simple cybernetic decisionmaker, the
flesh-and-blood human is able, both spontaneously and as the result of sig¬
nals, to change the problem formulation, rearrange the structure of his or her
expectations, to alter his or her states of readiness. In Kantian terms, we
might say that the human system is able to “evolve categories of description
beyond those built into its design . . . .” (MacKay, 1969, p. 55.)
The lexicon of systems theory has a couple of words that sound as if they
were intended to articulate a conception of the metamechanistic or the
metacybernetic; these are open system and self-organizing system. The first
is of thermodynamic origin and refers to a system that is not isolated and
thus is able to exchange matter or information with an outside environment.
A self-organizing system—by one definition, at least [Mesarovic, 1964u, pp.
11_13]—is a system that can change its structure in response to the environ¬
ment.21 Clearly, a human system is both open and self-organizing. But I am
not entirely persuaded that these terms by themselves fully capture our
intuitive sense of the distinction between a mechanistic and nonmechanistic
(or metamechanistic) system. Openness and self-organization are certainly
necessary but perhaps not sufficient conditions to qualify a system for the
latter category. Indeed, these distinctions do not seem to rule out various
sorts of robots or things like “perceptrons” and “heuristic programs."
[Ibid.]22 Perhaps the best distinction, despite its greater generality, is that
between a “morphostatic” and a “morphogenetic" system. [Buckley, 1967,
pp. 58-59.]
Models in the mainstream of economics—including, as I have suggested,
those in the mathematical economics of information—portray the economic
agent as a passive cybernetic reactor, a morphostatic system that responds
to changes in data according to the logic of the economic problem pro¬
grammed into it by the economist. As Fritz Machlup has long and patiently
explained, this approach is not a statement about human psychology but a
perfectly reasonable and justifiable technique of analysis particularly for
the sorts of economic problems to which the basic tools of partial-
equilibrium comparative statics are normally applied. [Machlup, 1967.] But
even granting all this, there remains a wide range of problems that demand a
more active, morphogenetic ideal-type of the economic agent—one who can
acquire structural knowledge and change the economic problem he or she
faces.23

21 Ludwig von Bertalanffy’s definition of this term seems a bit different and comes nearer to the
distinction I am looking for. He describes a self-organizing system as one capable of “evolving
from a less to a more differentiated state.” (Bertalanffy, 1968, p. 68.)
22The reason for this, I believe, is that Mesarovic is willing to classify systems as self-organizing
if they change their structure within the dictates of a fixed higher level structure; that is, if they
“change their structure by using a relation from the set fl^.“ (Mesarovib, 1964«, p. 11.)
23Morphogenetic man has not been entirely neglected in economics. Under the title of entrepre-
SYSTEMS THEORY, KNOWLEDGE, AND THE SOCIAL SCIENCES 599

If economic agents are cybernetic reactors, they can adjust the economic
mechanism in light of changed circumstances, but they cannot thereby in¬
crease the organization (decrease the entropy) of that mechanism. And any
imperfection in the agent’s knowledge (that is, any lack of correspondence
between the economic problem perceived by the agents and the “true”
economic problem) can lead to a bad equilibrium, a market failure. But if
economic agents can alter the problem they face, if they can bring new
structural knowledge into the system, then elaboration of, and increased
differentiation in, the economic system becomes possible. Far from causing
disorder or chaos, apparent departures from homeostatic equilibrium can
actually result in an increase in system organization and a decrease in en¬
tropy. This is the phenomenon of “spontaneous order” [Hayek, 1967, p. 77]
or, in technical jargon, “deviation-amplifying mutual causal effects.”
[Maruyama, 1963.]24
Recognizing the existence and importance of morphogenetic processes
has a number of far-reaching implications for both systems theory and eco¬
nomics, but this is not the place to explore them, I am afraid. Instead, let me
close with some brief observations on relations between these two disci¬
plines.
Systems theorists of my acquaintance are sometimes inclined to the opin¬
ion that economics lacks an adequate systems perspective and its scientific
development would be rapidly enhanced by a complete subsumption of that
discipline into systems theory. This viewpoint is not without its ironies. In
an important sense, it was economists who were the first systems theorists.
Moreover, the founders of economics were concerned with a fully mor¬
phogenetic version of systems theory. [Hayek, 1967.] Adam Smith’s concep¬
tion of economic growth based on the increasing division of labor is very
much a theory about the evolution of economic structure from a less to a
more differentiated state. [A. Smith, 1776 and 1936.] And, indeed, it is now
widely recognized that the theory of evolution, known to us through the
biological theory of Darwin, was articulated at least a century earlier in the
social sciences.25
Of course, modern systems theory is also concerned with questions of
morphogenesis, but mathematical systems theory—in the guise of non¬
equilibrium thermodynamics [Prigogine, 1971] and topology [Thom,

neur, he has been studied by Joseph Schumpeter and more recently, Israel Kirzner, among
others. [Schumpeter, 1934; Kirzner, 1973.]
24 Another name for this class of phenomena is autopoiesis, an area that is apparently attracting
increasing interest among systems theorists. For a bibliographic introduction, see Zeleny
[1981].
25David Hume, who numbered economics and social thought in general among his philosophical
interests, adumbrated a genuinely Darwinian view of evolution in his Dialogues Concerning
Natural Religion, first published in 1779 but written in 1759, a hundred years before The Origin
of Species. [Hume, 1961, especially p. 478.]
600 RICHARD N. LANGLOIS

1975]26—is just beginning to grapple with these problems. In practice, the

systems theorists who wish to bring economics under their wing are the
control theorists and cyberneticians (in the narrow sense), the builders of
naive-holist macromodels of the economy. In this sense, then, a subsump¬
tion of economics into systems theory is a step backwards—a step back in
the direction of the French physiocrats, whose ideas Adam Smith tran¬
scended in founding the discipline of economics.27 The problem is not a lack
of high-powered systems mathematics in economics: Modern economists
have not been slow in the least to adopt whatever new mathematical tech¬
niques emerge from systems theory. Indeed, if there is any need to make
economics more system-theoretical, it is a need not to give it over to the
cybernetic modelers but to return the discipline to the more sophisticated
version of systems thought on which it was founded.

26This work by Thom, which presents the notion of catastrophe theory, has attracted a large
following in a number of disciplines and remains something of a cause celebre in mathematical
systems theory. It promises mathematical interpretations of some of the most ancient and
vexing problems of philosophy—the nature and origin of form and structure and the text
alternates enticing intimations of the profound with an utterly impenetrable formalism, thereby
applying a time-tested formula for attracting a cult following.
27On the similarities between physiocrats and certain modern systems models of the economy,
see Almarin Phillips [1955].
 SIGNALS, MEANING, AND
CONTROL IN SOCIAL SYSTEMS

Walter Buckley

There are three main topics in Langlois’s paper on which I would like to
comment. One is his philosophical critique of systems theory involving
“emergentism.” Second is his discussion of information and meaning in
relation to social science. Last is the question of control theory, again as it
may apply to social science.
I shall not dwell long on the first topic since, in my view, Langlois has
erected a straw man in arguing that systems theory, in its concern for the
whole and its emergent properties, tends to ignore the components. None of
the founding theorists or important contributors have held such a view. The
mistake perhaps derives from the emphasis given to emergent properties and
behaviors of the whole in the attempt to offset the strong reductionist ten¬
dencies in most of the sciences. This issue has been well reviewed by Anatol
Rapoport in his contrast of the traditional analytical methodology of physics
with the glaring need for synthetic methods in the biological and social
sciences. [Rapoport, 1968«.]
Langlois’s characterization of intelligent or sophisticated methodological
individualism wipes out any significant differences with emergentism and
does away with important issues that still need to be thrashed out in princi¬
ple. In practice, however, since we cannot begin to explain or predict the
properties or behaviors of complex dynamic wholes from a knowledge of the
parts, it seems wise to study both and try to understand how each interacts
with and influences the other.
In discussing communication theory, it is, of course, important to distin¬
guish and define signals, information, and meaning and distinguish as well
their relevance to simple systems as compared to complex systems that
decide or choose different behaviors. Soon after Shannon published his
theory of information, it became widely recognized that it was actually a
theory of signals, and he himself recognized that it did not deal with informa-
601
602 WALTER BUCKLEY

tion in the semantic sense. Langlois is correct to point to the mechanistic

implication of signal theory, though there is no need to deny that complex
systems such as computers are indeed information processors—as long as
that does not imply that the information has meaning for them. The example
that Langlois gives of “a slightly more sophisticated version of S/R” theory
doing duty as a goal-seeking system with the stimulus being the information
and having an “intermediate processing stage” gives too much to S/R
theory. A distinction that might be helpful here has been made by automata
theorists and others between simple systems with constant internal states
and systems with varying internal states. For the former, the outputs are a
function only of the inputs, and given a particular input at any time, the
output must always be the same. For the latter, the outputs are a function of
both the inputs and the internal state at a particular time, and, thus, the same
input may result in different outputs at different times or vice versa. All
systems capable of making a choice, or selecting different outcomes even if
the input is the same, depend on this feature—this means all living systems
and some that are man-made. Simple systems, including S/R systems, do not
“process information,” but only react to signals.
Signals are varying objects or physical processes and may or may not
convey information. Information is not an object or entity, but is inherently
relational, a mapping between two or more sets of events. When we seek
information from a statistical analysis of our research data, we are looking
for such a mapping. Information, if it properly maps to the internal states of
a system, can selectively trigger sets of behaviors. If the system has constant
internal states, it behaves predictably; if not, it is capable of novelty.
I was glad to see Langlois’s reference to the work of Donald MacKay,
which has been a unique contribution too little recognized and not suffici¬
ently built on. His notion of information changing the conditional probabili¬
ties underlying a person’s readiness to act in certain ways should help to
extend the application of modern communication theory to the psychologi¬
cal and social sciences. Given his definition of information as that which
does logical work on an organism to keep its field of purposive activity
matched to the requirements of its environment, he goes on to define the
meaning of an item of information as its selective function on the range of
the organism’s possible states of orientation. MacKay notes that this is also
a relational concept. On this basis, communication becomes a process
whereby some of this selective organizing work of one organism is induced
via symbols in another organism. This is a deeper way of saying that organ¬
isms can learn from the experiences of others without having to risk direct
confrontation with the environment themselves. In return, however, there is
the risk of false or incomplete information (or mapping).
Though Langlois has recognized the merit of MacKay’s work, some of
Langlois’s discussion is misleading or mistaken. It is not helpful to speak of
the meaning of a piece of information for what he calls causal or goal¬
seeking systems. (Here it is necessary to distinguish goal-seeking from pur-
SIGNALS, MEANING, AND CONTROL IN SOCIAL SYSTEMS 603

poseful systems. [Ackoff and Emery, 1972.] Modern weapons as well as

thermostats are goal-seeking.) Nor is it felicitous to speak of voltages being
meaningful and understood by such systems. And once again it is inept to
identify S/R systems with cybernetic systems.
More importantly, it is an oversimplification of MacKay’s concept of
meaning to define it “solely in terms of the action released” and thus iden¬
tify it with a conditioned, hence, mechanical response. Langlois comes
closer to the mark when he later points out that what identifies the realm of
meaning is the overall configuration of potential behaviors from which selec¬
tions might be made rather than any particular response in isolation. The
manner in which he presents the example of the agricultural decision-system
makes it more appropriate to a narrower signal or information theory. The
meaning of the input information about the possibility of rain is not simply
the acreage chosen for planting, but rather, it is the mapping between the
learned properties of rain and the range of possible consequences of rain for
the farmer’s particular goals and interests at the moment. If those goals,
interests, or other internal states change, the mapping—hence, the mean¬
ing—of rain may change. These various properties, consequences, goals,
and values interact in the farmer’s mind to lead to the choice strategy he
settles on. The meaning cannot be narrowly defined in terms of the final
action resulting from the decision process.
Also it is misleading to suggest that the meaning of a message does not
exist until the action occurs, and, hence, it is the action that generates the
meaning. There may be confusion here with the process whereby a meaning
comes to be generated in the first place. George Herbert Mead’s theory of
the social genesis of mind, self, and social institutions contributed greatly to
our understanding here. [Mead, 1934.] For Mead, meaning comes to be
established in social interactional settings through the association of ges¬
tures and symbols with particular acts. Thus, a symbol comes to mean the
tendency to act in a way called out by the object symbolized. For Mead, the
meaning of the symbol chair was the tendency to sit down in one. (This, too,
comes close to an overly simplified identification of meaning and an act it
might call out, representing a weakness in his views, deriving perhaps from
his behaviorist and pragmatist background and environment.)
Taking a more or less standard decision-model, we can see the many
different intermediate points at which input information may have a bearing
on—play a selective role in—determining some final behavioral outcome:

1. Input information may help select which of several problems we

should now make a decision about.
2. It may help select the set of potential strategies of action that might be
put to bear on the selected problem or goal.
3. It may help select a particular strategy from the set of potential ones.
4. Finally, it may help select an initial act in carrying out the strategy or
plan.
 604 WALTER BUCKLEY

An important implication of this viewpoint is that a set of signals or

information that is meaningful (that maps part of the world) for one person or
group may only be, literally, noise for another (as when I hear Chinese
spoken). Furthermore, people using the same language may talk past one
another when the words and phrases are mapped to very different social or
cultural worlds. Mead tried to show in great detail that the social bond,
which makes possible permanently organized social and cultural life, in¬
volves an empathic interpenetration of minds and perspectives on the world.
This is generated through a long socialization process based on symbolically
induced self-images, other images, and common meanings, which thus make
it possible to coordinate activities with others by taking the role of the other
and orienting behavior toward a collective generalized other. A better under¬
standing of the process is essential to an assessment of why it breaks down.
In turning to control theory, especially as applied to economics, Langlois
offers the rather startling thought that Adam Smith was a founding father of a
more advanced theory of cybernetic control and self-organizing systems.1
All of this requires a more precise discussion of control theory than Lang¬
lois, or I, can muster in a short space. I will try, however, to fill in some gaps
and resolve some ambiguities as I see them. At a minimum, not only must
open-loop and closed-loop control be defined more carefully, but the notion
of uncontrolled stabilizing system should be introduced.
If we conceptualize a control system as consisting, at minimum, of a
controlling element that outputs control variables to the controlled system, a
major problem of control is how the controlling system determines what
values of the control variables to output in order to keep the system on the
goal-track. This implies, also, that for a system to be able to control (or
regulate), it must have a reasonably well-specified goal.
In closed-loop control, such as our standard thermostat, an auto-pilot, or
a driver controlling a car down the middle of the lane, the information
required to determine the control variables is derived from the representa¬
tion of the goal-state in the system and from feedback information about the
system’s actual behavior—which, in turn, implies information about the
disturbances acting on it and causing it to deviate.
In open-loop control, this feedback information is lacking, possibly be¬
cause the system takes too long to respond to control or has not yet been
affected by expected disturbances but must prepare for them. The control
information must then come from direct assessment of the acting or impend¬
ing disturbances and previously learned or currently theorized knowledge of
what control actions are required to cancel the effects of the disturbances
and maintain the goal-track. An example of a tentative and still half-hearted
attempt at open-loop control is recent governmental activity aimed at in¬
troducing control measures to head off anticipated societal (and global) dis-

See Wassily Leontief s recent critique of academic economics and its nonempirical founda¬
tion. [Leontief, 1982.]
SIGNALS, MEANING, AND CONTROL IN SOCIAL SYSTEMS 605

turbances stemming from energy-resource depletion, pollution, population

growth in underdeveloped societies, and so forth. The effort is especially
disconcerting for Western societies, since there is an obvious inherent con¬
tradiction between the required societal and global control of relevant eco¬
nomic and other activities on the one hand and an economic ideology of no¬
control (laissez faire) derived from—yes, Adam Smith.
This brings us to the notion of uncontrolled but sometimes stabilizing
systems. [Alfred Kuhn, 1974.] As already stated, control (or regulation)
implies some clear representation in the system of a goal-state, whether
programmed in by adaptation and evolution or by a human being, or learned
from experience by a sentient being. Consider a simple ecological system of
rabbits and foxes and their interlocked niches. Given certain initial popula¬
tions of each in a certain stable environment, the population sizes of each
may (or may not) stabilize and remain relatively constant over long periods
as long as only relatively small disturbances occur. This stability is due to
mutual relations among the variables, but these are not error-bearing feed¬
back relations relative to some system goal. For there is no overall goal
represented anywhere in the system, although component organisms have
their own evolved goal-states (physiological needs). We speak of ecological
systems as maintaining a delicate balance, because they are not control
systems—not self-regulating. If internal or external environmental changes
begin to occur, the system will easily be disrupted if not permanently de¬
stroyed—as is occurring, for example, with most of the world’s rain forests.
The Adam Smith conception of an economic system translates into basi¬
cally the same as that previously outlined—an uncontrolled system that
sometimes stabilizes under rather ideal environmental conditions. Had
Smith acquired modern conceptions of cybernetics and control theory, he
could never have seriously introduced his notion of the “invisible hand” or
the market place as a control or regulating mechanism. Western economic
systems have demonstrated throughout history their delicate balance and
consequent vulnerability to internal or external change. They have come
under increasing attempts at public control but never sufficiently to override
the noise in the system generated by the independent goal-actions of many
actors. Modern collective decision theory has amply conceptualized and
demonstrated how the collective as well as individual goals of all may be
frustrated by the independent self-interested goal-behaviors of each. The
“tragedy of the commons” is built into classical economic theory, which is
currently reaping its dividends with special vengeance.
Langlois begins his paper with a critique of systems theory. I heartily
agree that there are many things wrong with the movement (and the people
involved in the Society for General Systems Research have recently been
strongly self-critical), though I do not think Langlois mentions the core
problem. Systems theory is widely recognized by most of its serious propo¬
nents as not a theory (except for the mathematician), but a set of conceptual
and methodological tools along with a number of guiding principles. It can-
606 WALTER BUCKLEY

not by itself provide a theory for any particular substantive area of study. To
be useful, it must be integrated into the conceptual and empirical materials
of a discipline. There are too few who have mastered it sufficiently well to
apply it knowledgeably to their social-scientific field of study, especially
empirical study. Most systems researchers are not social scientists and vice
versa. Consequently, many attempted applications of systems theory to
aspects of society are either poor social science or poor systems research. It
is asking a lot to expect someone to master another field of study such as
systems research, especially when the latter itself draws from a number of
disciplines. But some of its important concepts are being quietly assimilated
by various social sciences, though substantial progress may have to wait for
the development of institutes and programs designed to integrate systems
research and substantive studies.
It is my belief, however, that the promise of systems research for the
social sciences is still great and has much more to offer in the areas of
communication and social control than we have yet seen. The two need only
engage in more intensive and empirically relevant interaction.
 TOWARD A SYSTEM-BASED
UNIFIED SOCIAL SCIENCE

Robert D. Beam

We often hear that "one man can no longer cover a broad enough
field" and that "there is too much narrow specialization" . ... We
need a simpler, more unified approach to scientific problems, we
need men who practice science—not a particular science—in a
word, we need scientific generalists.

(Bode, Mosteller, Tukey, and Winsor, 1949, p. 553.)

Richard Langlois’s paper reflects the opinion that systems theory has failed
to bring about interdisciplinary synthesis of, and integrated education in, the
social sciences. I would like to present a quite different relation between
systems theory and the social sciences. In contrast to Langlois’s position, I
would like to suggest a means by which systems theory can “unify ... the
methodology of all sciences, . . . endow the social sciences with that natural-
science-like rigor,” and combat the “fragmentation and specialization” of
social-science knowledge. (Langlois in this volume.) It is my opinion that
systems theory has much to offer at the basic and most general level of social
science, because, and again in contrast to Langlois, I feel it has succeeded in
bringing “all disparate parts of inquiry under the sway of its organizing
force.” (Ibid.) Systems theory has provided a framework for social theory, a
framework that allows for unification of the behavioral and social sciences
on an analytic-deductive basis. Such a unified social science is built on
systems concepts from the ground up, and the chief objective of my com¬
ment is to provide a brief outline of this discipline as it relates to a selected
few of Langlois’s remarks.
I would first like to articulate a model that makes a clear distinction
between the servomechanism responding to environmental stimuli in a pre¬
programmed fashion and the human system as a complex controlled system
capable of effecting conscious learned responses to environmental stimuli.
This distinction is, I believe, one that Langlois also considers to be required.
A complex controlled human system seems to me to be analogous to what he
terms a metacybernetic system. This is a model that seeks to accomplish for
607
608 ROBERT D. BEAM

social science what the model of “economic man” accomplishes for eco¬
nomic science. In contrast to the cybernetic-reactor model of economic
man, this model enables the agent to alter the environment or his or her
relation to it.
Since all behavior of analytic relevance to social science is learned behav¬
ior, this model focuses only on behavior that is directed from the nervous
system toward some goal. The present model also assumes that all informa¬
tion that can be used to direct behavior is stored in the form of concepts
(images) to be retrieved from storage through perceptions, memory, or com¬
munications.
The simple cybernetic model includes three logically irreducible ingre¬
dients for purposive behavior selection: a detector that acquires and pro¬
cesses information about the system’s environment; a selector that chooses
responses on the basis of the system’s goal-state; and an effector that effec¬
tuates the overt behavioral response. The nature of control is the system’s
internal ability to detect deviance from a preferred state and correct its own
behavior so as to restore, or move back toward, the “preferred” position.
All purposive behavior requires negative feedback for both servomecha¬
nisms and organisms, including the human system.
But unlike other controlled systems, humans possess conscious aware¬
ness of, or information about, the system states of their detector, selector,
and effector (DSE) subsystems. The more complex model of the human
system contains sub-DSE controls for each of the main-level DSEs.
Mechanical man-made servomechanisms, even computers, do not possess
these types of sub-DSE controls.1 The complex cybernetic model presented
here provides a conceptually simple explanation of consciousness, even
though it is not yet firmly established where in the brain consciousness
resides or how it functions. Although the human biological system is respon¬
sible for effectuating overt behavior, it is the method of behavior selection
that is of interest to social science. Thus, it is the control system of humans,
as opposed to the biological system, that formulates and guides the execu¬
tion of behavior relevant to psychological and social analysis.
The terms detector, selector, and effector have strictly functional defini¬
tions. The detector encompasses all sensory, cognitive, and intuitive pro¬
cesses by which humans form, identify, and modify both real and imaginary
concepts about themselves and their environment. The selector spans the
entire range of emotions, pleasures, pains, fears, loves, desires, and so
forth—all things to which positive and negative valences can be attached, a
valence being a property of the nervous system that produces pleasant or
unpleasant feelings that reinforce approach responses or avoidance re¬
sponses, respectively. The effector includes all neural connections that di-

1 For a more complete presentation of the human system as a complex controlled system, see
Alfred Kuhn [1974, pp. 60-102; 1975, pp, 49-73], The contents of this paper draw heavily on the
model of unified social science, as developed in these two volumes.
TOWARD A SYSTEM-BASED UNIFIED SOCIAL SCIENCE 609

rect behaviors as diverse as splitting wood, singing an aria, or dancing the

tango. Nothing in this model precludes one DSE subsystem from modifying
the other two, as when detector modifies selector (we want what we believe
is possible, and we do not want what we believe is impossible), selector
modifies detector (we sometimes see what we want to see and hear what we
want to hear), or some other combination.

MEANING, INFORMATION, AND KNOWLEDGE IN THE DETECTOR

In this model, the detector deals with cognition and perception, which are
essentially the processes of learning and using patterns. Cognition, or pat¬
tern learning, involves the processes by which pattern concepts are orga¬
nized and stored in the detector. Perception involves the use of incoming
sensations as cues to identify (match) events in the environment with con¬
cept images (referents) of those events retrieved from storage. More
specifically, signal inputs from the environment activate sensory receptors
that modulate sensory nerves. This is the detection stage in which informa¬
tion received is without meaning. Information acquires meaning only after
the detector codes or organizes sensory and linguistic inputs into concepts
or patterns of concepts by categorizing the inputs by their similarities and
differences. The detector can organize information into its own concepts or
it can adopt the organized concepts of others through communication. The
detector forms new concepts by abstraction (inductively forms new “emer¬
gent” concepts through the essentially morphogenic process of random
variation and selective retention or intersection (acquires new concepts
by deduction and discrimination, usually in the form of semantic com¬
munications). Knowledge in the detector is thus organized information, and
this, I take it, is what Langlois means by his statement that knowledge and
organization are identical.
If cognition in the detector is the process of coding and storing informa¬
tion in conceptual bins, perception is the process of decoding or retrieving
coded information from storage. Perception is the process by which the
detector matches primary patterns of incoming sensations with secondary
patterns of previously stored images (knowledge) to infer information about
presently experienced reality that is not contained in either primary or sec¬
ondary form alone. Perception thus involves the receipt and use of primary
information as cues to identify or activate secondary information in the form
of previously stored concept-images. Matching uncoded sensory inputs with
previously coded (learned) and stored concepts is known as pattern recogni¬
tion. A signal input does not become a stimulus in the detector until it has
been decoded or identified, and strictly speaking, it is not the signal itself,
but the identification that constitutes the stimulus. A meaningful signal input
is one whose cues are sufficient to activate a stored channel of concept
images in the detector, or in Mac Kay’s terminology, to select from the range
610 ROBERT D. BEAM

of all possible “states of readiness.” MacKay’s metaphor of the railroad

switching yard aptly describes this decoding process in the detector.

MOTIVATION IN THE SELECTOR

The selector subsystem reflects the control system’s preferences, goals,

values, and inner tendencies. Learning occurs in this subsystem as well as in
the detector. The inborn (unlearned) components of this subsystem are its
primary motives or the pleasant or unpleasant valences that certain emo¬
tions or sensations activate. Primary motives are oriented toward physical
survival. The learned components of the selector are known as secondary
motives and are conditioned connections between a learned concept and an
inborn valence. We cannot appreciate or like economics until we have
formed some previous concept(s) of what economics is about. Pattern learn¬
ing thus involves learning a concept in the detector and (simultaneously)
attaching a valence to that concept in the selector. Valences about concepts
can be modified with experience or through additional concept learning.
Pattern using in the selector occurs whenever choices are made on the
basis of learned secondary motives. The strength of a secondary motive
depends on the degree of pleasantness or unpleasantness of the attached
valence. Whereas concepts constitute detector states, motives constitute
selector states that reinforce approach responses or avoidance responses to
concepts that set them off. Both the detector and the selector learn and use
primary and secondary levels of information and motives in selecting overt
behavior to be executed by the effector.

BEHAVIOR EXECUTION IN THE EFFECTOR

For the individual human, all decision processes are confined to the detector
and the selector. Effector processes are confined to the muscular execution
of previously selected behavior. Pattern learning in the effector involves
learning new performance skills, while pattern using involves their exercise.
Less emphasis is placed on the effector in understanding directed behavior
in humans, since social science is concerned more with the what and why of
behavior selection than the how.
A person’s behavior is guided toward his or her goals by decisions that
reflect the interrelated detector, selector, and effector states of his or her
control system. Decision theory deals with response selection in humans
under conditions of complexity, particularly with respect to the conscious
content of all DSE subsystems. The adaptive behavior process can be
viewed in this model as involving two stages, a performance stage and a
feedback stage. During the performance stage, the detector identifies an
opportunity function of perceived alternative responses consonant with the
TOWARD A SYSTEM-BASED UNIFIED SOCIAL SCIENCE 611

capabilities of the effector. The selector ranks these responses in a prefer¬

ence function according to their likelihood of achieving a given goal or set of
goals. A decision occurs when the most highly valued alternative is selected
from the opportunity function and carried out by the effector. A feedback or
learning stage follows in which the detector receives feedback information
about the success of the previous action. The selector compares the new
state of the environment with the goal-state and elects to continue or change
the response, and the effector receives new instructions from the selector
and carries them out but at a level different from that of the control system
itself. A single stream of behavior is manifested in which the feedback stage
of a previous action melds imperceptibly with the performance stage of the
next. Learning occurs when through experience or a change of goals, a given
stimulus-input of information elicits a different output-response than it did
before. Directed behavior in humans is the result of decisions that interrelate
perceived alternatives in the detector with secondary motives in the selec¬
tor.
This DSE model of behavior selection provides a “proper scientific expla¬
nation” of “willed” or goal-oriented behavior [Alfred Kuhn, 1981], an ex¬
planation that until the development of cybernetics had sharply divided the
goal-and-future-oriented soft sciences from the antecedent-cause hard sci¬
ences. Such an explanation occurs when the purposive action to be ex¬
plained, the explicandum, is deduced from pieces of information that consti¬
tute the explanation, the explicans. [Homans, 1967, p. 25; Brown, 1963, p.
122; Hempel, 1963, p. 50.] To illustrate, using Langlois’s decisionmaker:

Explicans
Premise 1 (detector) if: our decisionmaker perceives that planting acreage
Ax will yield a profit of $100,000.00 and that planting
acreage A2 will yield a profit of $75,000.00,
Premise 2 (selector) and if: his dominant motive is to maximize profits,

Explicandum
Conclusion (effector) then: he will plant acreage Ax.

This example illustrates why economics is the most theoretically precise

of all the social sciences. The model of economic man makes explicit as¬
sumptions about detector and selector states for each person. The detector
contains complete and accurate information. The selector’s goal-state is
unambiguous maximization of profit or utility. The behavioral response is
unequivocally prescribed as soon as alternatives are compared with the goal.
If we introduce uncertainty in the detector into the analysis, we may compli¬
cate the decision process, but not its basic logic of formulating the scope of
the decision and listing alternatives to be considered within the set. Viewed
as a cybernetic system, a decision’s components are the costs and values of
various alternatives, and the governor is the motive system of the decision-
612 ROBERT D. BEAM

maker. The selector will be unable to formulate a preference-ordering of

responses until expected costs and benefits of each have been assessed,
either through probability distributions or by buying the service of the clair¬
voyant. In either case, if the information needed to eliminate uncertainty in
the detector is costly, the decision to acquire it will be made only if the
expected value of this information exceeds its cost to the decisionmaker.

THE INTRASYSTEM/INTERSYSTEM AXES OF CONTROLLED SYSTEMS

All disciplines in the social sciences deal with types of social systems, yet
their specialized concepts bear little resemblance to one another. There
exist, however, some basic social mechanics that underlie all social science
in the same sense that principles of the lever, inclined plane, and pulley
underlie all mechanical contrivances. [Alfred Kuhn and Beam, 1982.] The
intrasystem/intersystem axes of controlled systems shown in Figure 1 reveal
basic system-based concepts of social interaction that underlie not only the
core disciplines of economics, political science, and sociology, but also such
related areas as small group theory, social psychology, management theory,
and organization theory. The rectangles in Figure 1 show two goal-oriented
systems, to wit, two individuals or two formal (controlled) organizations of
individuals. Within each system are listed the three logically irreducible
ingredients for purposive behavior: detector, selector, and effector. Behav¬
ior selection is separated into a decision stage and an effectuation stage with
feedback. An intrasystem view consists of analysis of the behavior of a given
system as a function of the DSE states of its control system. Since social
science deals with interactions among two or more goal-oriented systems, an
intersystem view is used to analyze interactions of their detector subsystems
(communication), their selector subsystems (transaction), and their effector
subsystems (organization).
Communication is the process of transferring patterns between systems.
It both reflects and affects their detector states. Transaction transfers things
of value between systems and reflects and affects their selector states. Or¬
ganization carries out joint behavior by two or more systems and reflects and
affects their effector states. Typically, systems theorists have analyzed link¬
ages (social interactions) among humans with the analytic tools in informa¬
tion and communication theory. [Cadwallader, 1959; Haberstroh, I960.]
Matter/energy interactions—for example, sexual union or physical com¬
bat—are typically categorized in a different conceptual bin. [Buckley, 1967,
p. 48.] The intersystem view of Figure 1 separates interactions analytically
into the science of pattern transfers and the science of value transfers across
controlled systems. Although considered the sine qua non of interaction,
communications are viewed in this model as the facilitators of transactions
that reflect the values people attach to their matter/energy transfers, and not
the matter/energy transfers themselves. Thus, communications and transac-
TOWARD A SYSTEM-BASED UNIFIED SOCIAL SCIENCE 613

Party A Party B

Figure 1. Intrasystem and intersystem axes of controlled systems.

tions are the interactions between humans that parallel the information and
matter/energy interactions between living systems. I am not at all certain
how Langlois would respond to the contention that a tight and coherent
model of social interaction should include both the science of communica¬
tion and the science of transaction,2 each of which can be briefly outlined as
follows. [Alfred Kuhn, 1974, chaps. 7-11.]

Communications

The Shannon-Weaver mathematical theory of communication is directed

primarily at measuring the accuracy with which symbols used in communi¬
cation can be transmitted. [Shannon and Weaver, 1949.] The theory is based
on a mathematical theory of information that postulates a source to generate
signals of a given variety and a receiver who can make use of this variety.
[G. A. Miller, 1953.] The theory assumes that both the source and receiver
possess similar concept-referents, which they attach to the signs they use to
communicate messages. The communication theorist thus assumes that the
signs transmitted during the communication process have the same meaning
to the sender and the receiver. Since output depends on input and noise,
accuracy in communication refers to the extent to which input to the com¬
munication channel resembles output from the channel. The theorist is not

2 A more extensive coverage of the propositions relating to these basic sciences can be found in
Alfred Kuhn [1974], chaps. 7-11.
614 ROBERT D. BEAM

concerned with the semantic level of communication, but with the amount of
information measured in binary digits that can be accurately transmitted
through the channel.
By contrast, the communication model in Figure 1 assumes that the chan¬
nel is capable of accurately transmitting information—signals—and is inter¬
ested instead in the accuracy with which (1) the coded pattern to be trans¬
ferred is clear in the detector of the sender, (2) the message is detected by
the receiver, and (3) the concept-referents that the transmitted signs select in
the receiver’s head correspond to patterns in the sender’s head. Analytic
interest is here directed to the effect of the communication on the detector
states of the interacting parties, and not on the channel capacity of the
transmission-medium or its levels of noise or redundancy.
I believe this view of semantic pattern-transfer is in accord with Mac-
Kay’s framework as Langlois describes it. It is an alternative to the oil-flow
model of information, which, like the systems-referential view, might argue
that information is not like “hardware products to be delivered through
obstacles with minimum damage at least cost.’’ (Nevitt, 1981, p. 595.) The
core of the semantic problem of communication for social science lies in the
proper selection of signs, such that they serve as cues sufficient to enable the
receiver’s detector to decode their meaning. The mathematical theory of
communication, with its emphasis on bit-counts, noise, entropy, redun¬
dancy, and so forth, has much to contribute to binary computers, telecom¬
munications, and other areas. But it has little to say about the manner in
which humans match signs with concept images or recombine these images
through sign communications to form new images. Although bit-count analy¬
sis can apply to information transmissions below the social level, it does not
determine how much meaning is transferred by a given sign. The meaning of
the sign depends on the concept-referent of that sign as it exists in the head
of the sender at the moment of transmission and within the context in which
the sign is used. The communication model in Figure 1 is concerned with the
transfer of patterns across detectors of separate individuals, and not with
transmission of information in bits. Hence, a discussion of bit-counts when
referring to pattern transfers is not directly relevant to the analysis.

Transactions

At the intrasystem level, an individual’s decisions reflect the comparison of

benefits and costs among alternatives, a selected behavior being one whose
expected benefits exceed its cost. When faced with two mutually exclusive
alternatives X and Y, the cost of acquiring Y is the forfeited value of X, and
the cost of acquiring X is the forfeited value of Y. At the intersystem level,
we deal with situations in which party A is in possession of X, party B holds
Y, and an exchange is contemplated by both parties. An exchange involves
two separate but mutually contingent decisions. A will give up X for Y if and
only if he values Y over X. B will give up Y for X if and only if she values X
TOWARD A SYSTEM-BASED UNIFIED SOCIAL SCIENCE 615

over Y. Whereas successful communications between parties require similar

detector-states with respect to the patterns exchanged, matter/energy ex¬
changes are successful to the extent that the parties’ selector states are
dissimilar (have reverse preference orderings) for the things exchanged.
Although the term exchange seems to fit the transfer of goods and services
between parties, the general principles that derive from such exchanges can
be extended to cover other events, such as blackmail, boycotts, hijackings,
wars, strikes, and peer-group pressures, and so forth, all of which lack a
matter/energy base and do not fit the formal conception of an exchange.
The more general term that includes all of these value-based interactions
is known as a transaction, the analysis of which relies heavily on the con¬
cepts of power (the ability to induce wanted behaviors in others regardless of
the overt terms of trade), and bargaining power (the ability to get good terms
for oneself in transactions with others). Power and bargaining power are
system variables that change according to the parties’ selector-states for the
things exchanged and mold the success and terms of a given transaction.
These same variables operate regardless of the context—social, political, or
economic—in which they occur. Such a broadened and systematized theory
of selector-based transactions between humans provides the conceptual
base for a science of power that has heretofore been scattered throughout
the literatures of game theory, organization theory, community power struc¬
tures, conflict theory, social psychology, international and domestic politics,
bilateral monopoly, collective bargaining, and social exchange theory.
Cybernetics has been defined as the science of communication and con¬
trol. [Wiener, 1948 and 1961.] With respect to the interaction between par¬
ties A and B in Figure 1, communication is necessary if A is to get B's
detector to know what A wants. The aspect of control, however, lies in
understanding why B complies with A’s demands, by examining the ways in
which A provides inducements to B's selector. [Alfred Kuhn, 1981, p. 587.]
Detector-based interactions are logically separated from selector-based in¬
teractions. Every real interaction involves both communication and transac¬
tion, and a tight and logically rigorous nomothetic science of social interac¬
tion can be achieved only if the two are separated analytically. This is
because no propositions about pure communication also apply to proposi¬
tions about pure transaction. Langlois appears to consider all interactions as
instances of communication only. However, in this model, communication
is viewed as the indispensable medium of interaction, while its content is
mostly transactional.

Organizations

Whenever a continuing and reasonably stable pattern emerges from the joint
effects of the separately decided behaviors of two or more parties, the rela¬
tion they form is called an organization. An organization is a social system of
interacting subsystems of human components. It is an emergent social phe-
616 ROBERT D. BEAM

nomenon in that it is a higher level system than its component members. The
whole of an organization is more than the sum of its parts, where sum is
taken to mean not their numerical addition, but their unorganized aggrega¬
tion. [Buckley, 1967, p. 48.]
Just as an analytic need arose to separate interactions into communica¬
tions and transactions, so there is also an analytic need to separate an
organization by its structure and its process. Its structure is described in
terms of its subsystems and their interrelated roles, a role being a set of DSE
states and actions of a subsystem, including its specified communications
and transactions with other subsystems. A role is thus a pattern-set of expec¬
tations about the behavior of the individual who occupies that role, and the
structure of roles is analogous to the blueprint of the organization. The
organization process describes the interactions of the role occupants them¬
selves, particularly the way in which their actual communications and trans¬
actions constitute the joint effects of the organization as a whole/ [A. Kuhn
and Beam, 1982, chaps. 8-9.]
As social systems, organizations span the analytic gamut from purely
controlled (formal) organizations to purely uncontrolled (informal) organiza¬
tions, with a continuum of mixtures (semiformal organizations) in between.
Formal organizations, such as governments or business enterprises, possess
a controlled subsystem that acts on behalf of the whole system and has the
capacity to communicate with, induce motivations of, and instruct its sub¬
system role-occupants to behave in the way it (the control system) desires.
[Ackoff, 1957-1958.] Formal organizations are thus oriented around the con¬
sciously coordinated behaviors of parts into behavior of the whole. Con¬
scious coordination plus whole-system goals distinguish formal organization
from its conceptual opposite, an informal organization. An informal organi¬
zation such as a pure market economy is an ecological social system that has
no control subsystem, no whole-system goals, and does not behave as a unit.
Each subsystem pursues its own self-oriented goals, and any coordination
that produces a joint effect of their interactions involves morphogenic pro¬
cesses that are wholly unplanned and undirected at the main-system level.
The main identifying criteria of informal organization are absence of con¬
scious coordination and presence of attention to subsystem goals. A semifor¬
mal organization such as the sociocultural system is informal in the sense
that it does not behave as a unit but formal in that selected subsystems that
do behave as units seek to modify their own behavior as well as the behav¬
iors of other subsystems in ways they believe are in the best interests of the
organization as a whole. Thus, to repeat, the main identifying characteristics
of semiformal organization are absence of conscious coordination and pres¬
ence of attention to whole-system goals by subsystems.

3 See the relation between an organization as an acting system and a pattern system, in Alfred
Kuhn and Robert Beam [1982], chaps. 8-9.
TOWARD A SYSTEM-BASED UNIFIED SOCIAL SCIENCE 617

THE INTEGRATION OF SOCIAL SCIENCE KNOWLEDGE

Every discipline in the social sciences is a set of lenses and a set of blinders.
Our academic institutions have been quite successful in providing special¬
ized lenses but have been less successful in removing the blinders. The
intrasystem/intersystem axes of controlled systems in Figure 1 are a tight,
logical way to unify the methodology of all social science disciplines. Re¬
gardless of the specific context in which it occurs, the study of behavior-
selection in humans can focus on the three irreducible components of goal-
oriented behavior: detector, selector, and effector, or DSE. Coordinated
action among humans is possible only by linking together their control sys¬
tems via their detectors (through communication), their selectors (through
transaction), their effectors (through organization), or some combination of
these. Communication, transaction, and organization are the most basic
building blocks of social analysis. When used as modular units, these gen¬
eral-purpose conceptual basics can be assembled into a vast variety of spe¬
cial-purpose configurations that lie scattered throughout the conventional
social science disciplines.
For example, supply-and-demand curves in economics can be derived
from the transaction model by adding parties to each side of an initial trans¬
action. [Alfred Kuhn, 1974; Beam, 1979.] Conflict and cooperation are
configurations of communication and transaction. Transaction has also been
successfully used as a framework for labor negotiations. [E. E. Herman and
A. Kuhn, 1981, chaps. 11-12.] The symbolic interaction process is a
configuration of communication, transaction, and perceptual-detector pro¬
cesses. Authority is a unique transactional interaction within the context of
formal organization. Decision theory is the study of complex DSE processes
under conditions of uncertainty. The theory of government, as modified by
conditions of sovereignty, all-purpose goals, involuntary membership, and
legitimacy, is a special-purpose configuration of the formal organization
model. [Alfred Kuhn, 1975, chap. 13.] Group decision-making is achieved
through communication (persuasive influence), transaction (trade-off or
compromise), or a particular transactional configuration known as dominant
coalition. In each case, the same few basic components are fitted together as
modular units to form special-purpose configurations previously viewed as
the exclusive domain of specialists in each of the social science disciplines.
The purpose of this approach is not to undercut the disciplines, but to
provide them with a common set of system-based analytic underpinnings.
The integration of social science knowledge is not to be achieved by setting
more and more areas of specialized knowledge side by side under some
thematic interdisciplinary umbrella. [Bowler, 1981, p. 214.] Integration is
achieved, rather, by developing more efficient conceptual structures that
allow the human mind to deal with larger amounts of information by means
of a smaller number of general concepts. It is achieved with a structure that
618 ROBERT D. BEAM

reaches across specialized vocabularies and methodology to bring all “dis¬

parate parts of inquiry under the sway of its organizing force” (Langlois) by
transforming the myriad fragments of social and behavioral science into a
coherent whole. This is the great promise that systems theory has long been
supposed to hold for the social sciences: The promise of a unified social
science that (to borrow a phrase from James G. Miller) “discerns the pattern
of a mosaic which lies hidden in the cluttered, colored marble chips of
today’s empirical facts.” (J. G. Miller, 1978, p. 42.)
 SYSTEMS, INFORMATION,
AND ECONOMICS

Richard H. Day

I have been greatly stimulated by Langlois’s remarks on systems and infor¬

mation in relation to economics. Rather than take issue with small points of
disagreement I would like to augment his ideas and relate them to a very
general, dynamic view of information in economic development. First, I try
to identify the economizing of scientific information that is the basis of
systems thinking. I then argue for a relativist view regarding the individ-
ualist/holist controversy. Next comes a brief illustration of Langlois’s equa¬
tion of knowledge and organization. I further try to make somewhat more
precise the idea of a morphogenetic agent, arguing that a nonanalytic faculty
of mind, creative intelligence, plays a fundamental role in generating struc¬
tural information. Finally, I suggest that if economic systems are globally
unstable, this faculty would be a necessary condition for the continuing
evolution of human culture.

SYSTEMS THINKING: SUCCESS OR FAILURE

No field should be judged by its failure to accomplish everything to which its

most ardent advocates aspire, for then we should have to judge every field to
be a failure—just as we would have to judge to be a failure everyone among
us whose grasp fell short of his or her reach. No, a field must be judged by
the ability of its contributors to identify principles and organize them into a
coherent system of thought that compels attention, invites use, and stimu¬
lates refinement, generalization, and application. From this point of view,
systems thinking has been a success.
If we look at the early pioneers alone (Lotke, Cannon, and Ashby are my
favorites), we find already firmly established the important fact that struc¬
ture and process, once carefully identified and described for a given range of
phenomena, are similar, perhaps identical, to those of quite a different range
of phenomena. This enables results developed in one field to have immediate
application to another.
619
620 RICHARD H. DAY

A striking example is the use of concepts of negative-feedback control in

engineering (servomechanisms), economics (behavioral adjustment pro¬
cesses), biology (homeostasis) and, indeed, systems dynamics, generally.
Here is a primal concept of great utility in a diversity of settings.
Other examples abound. Several years ago, I began considering the exis¬
tence of erratic fluctuations in deterministic economic processes; how this
happened is germane to our discussion. 1 was led to the work of Edward
Lorenz, a meteorologist, by Kenneth Cooke, a mathematician who was a
fellow participant at a small working conference on simulation in archeol¬
ogy. [Sabloff, 1981.] I had described my ideas on adaptive economics and
recursive programming, which I used as a framework for simulating eco¬
nomic development and technological change in modern agriculture and
industry. [Day, 1975; Day and Cigno, 1978.] But for this occasion, I illus¬
trated the ideas with models of the prehistoric transition from hunting to
agriculture and from feudal manorialism to a market-based agriculture.
[Day, 1981a.] Cooke observed that Lorenz's concept of deterministic, non¬
periodic flow could be applied to some of the equations in my models. A few
days later, I told Jess Benhabib, a colleague then at the University of South¬
ern California, of this work. He had already run across it quite indepen¬
dently in collaborating with another of our colleagues, Kazuo Nishimura, on
the existence and stability of cycles in economic growth. We began a collab¬
oration on the topic, in the course of which we made use of related literature
in biology, classical mechanics, and chemical thermodynamics, just to men¬
tion a few. [Robert May, 1976; Moser, 1973; Prigogine, 1971.]
The key idea that unifies research in these varied fields is that of the
complicated implications of nonlinear dynamic systems (difference or differ¬
ential equations). Others, too, emphasized this important principle. [Sam-
uelson, 1948; Forrester, 1962.] Its nature is still only partly understood and
the subject of intense investigation.
I was especially intrigued by results concerning a quadratic difference
equation that had arisen in some of my earlier studies. I had noticed the
complicated possibilities that equation possessed, but I had set the work
aside. [Day, 1967.] In the meantime, the same equation had been inves¬
tigated intensely in connection with totally different lines of work. [For a
review, see Day, 1982.] It was eventually found by Mitchell Feigenbaum to
exemplify properties common to a large class of nonlinear functions. [M. J.
Feigenbaum, 1980.]
These observations also shed some light on the nature of information.
Indeed, the structural-information content of phenomena may be described
by an abstract, dynamical system that seems to manifest salient features of
those phenomena. In this sense, the idea of a system is analogous to that of a
strategy or policy. A system represents all its possible implications in re¬
sponse to all conceivable applications, “parameterizations,” and calibra¬
tions just as a strategy represents all actions that respond to all conceivable
decision inputs. Systems thinking is, therefore, a way of expressing com-
SYSTEMS, INFORMATION, AND ECONOMICS 621

pactly as much as we know and can find out rigorously about classes of
disparate phenomena. In short, systems thinking is a means for economizing
thought.
No one should appreciate such a point of view more than economists. The
principle is economic. Its application is in no field better exemplified than in
economics. I have in mind Samuelson’s Foundations of Economics, which
did so much to identify and clarify the small number of principles that lay at
the heart of economic thought as it had evolved in its many branches.

INDIVIDUALISM, HOLISM, AND AGGREGATION

Langlois is correct in emphasizing that complete understanding of emergent

phenomena must involve attention to the individual parts that make up a
given whole. His examples, drawn from economics, are good ones, and it
would behoove systems scientists to devote more attention to insights from
that field. A certain amount of handslapping (if somewhat self-righteous)
may be salutary here.
But there is a nontrivial problem in the individualist point of view. There
are parts and there are parts, and parts within parts at various levels. It is
true that macroeconomics lacks even now an adequate microeconomic foun¬
dation, but we do macroeconomics anyway because we must. Governments
exist; they make policy, and there is a manifest need for reasonable policy.
Until really good microtheory of individual action and reaction can be devel¬
oped, less satisfactory macromodels will have to be relied on as primary
tools for carrying on debates on economic policy.
The problem goes deeper. Just as there are subatomic particles of which
atoms are composed, so the business firm is, in fact, an organization of
individual humans. It is not at all clear that an understanding of emergent
macroeconomic phenomena can be understood without reference to the
structure of relations among individuals within all the basic units of eco¬
nomic life, household, government, firm, bank, and so forth. And further,
can we really say that we understand the rationality of individuals until we
understand how and to what extent optimal choice emerges from the various
emotional drives and other diverse components of the mind? If economics
has made progress in the cause of the individualist methodology, it has
surely not made enough and only scratched the surface of this fundamental
problem.
My own opinion is that this problem can never really be solved. We can
only hack away at it: “What cannot possibly be completed remains never¬
theless a task,” said Karl Jaspers. (Jaspers, 1961, p. ix.) We must then
pursue understanding at many levels: Some must examine the parts (at a
given level) narrowly; others may investigate a system with parts defined in
some aggregate way in a more holistic fashion. We explore alternative aggre¬
gations, alternative specifications of detail. And at all times we must retain
622 RICHARD H. DAY

some uncertainty (and a healthy level of skepticism) about just what the
meaning is of abstract systems when used to represent what we understand
of phenomena.

THE MEANING OF INFORMATION

When Langlois considers the role of information in systems, he gets to the

root of the issue when he asserts the systems-referential view:

Information is not homogeneous; meaning is a matter of form not of amount;

and the value or significance of a message depends as much on the preexisting
form of the receiver as on the message itself. . . . [Information is stored as
knowledge in a system not as oil is stored in a tank, but by virtue of the change
that information makes in the very organization of the system itself. In a
fundamental sense, knowledge and organization are identical. (Langlois in his
paper in this volume.)

The McCulloch-Pitts model of the neuron and nerve nets provides a nice
illustration of this idea. [Shannon and McCarthy, 1956.] A nerve net is a
system of on-off threshold switches. An external input turns the input
switches on or off or leaves them unchanged. A sequence of messages
(pulses, signals) then flows from unit to unit altering the state of the system
by altering the state in each constituent element. Very simple examples of
such nets exhibit behavior analogous to memory and learning.
Whatever such a system may lack as a model of the brain, it may nonethe¬
less exhibit something profound about the nature of meaning in Langlois’s
systems-referential sense, how meaning emerges from a cascade of intercon¬
nected individuals capable of only elemental stimulus/response behavior and
where every meaning has an exact organizational analog.

STRUCTURAL AND PARAMETRIC INFORMATION:

THE MORPHOGENIC AGENT

The distinction Langlois makes between parametric information and struc¬

tural information is important. Let me elaborate and illustrate. Imagine a
mathematical system represented by difference or differential equations. Let
us suppose that a subset of equations represents states of an actor or agent.
A second subset represents the agent’s environmental states. The former
can be thought of as the agent’s policy. The complete system represents the
behavior of the agent-environment pair. Imagine that parameters of the
agent equations change. Then environmental states, as generated by the en¬
vironmental equations, which have remained fixed, will elicit modified be¬
havior. Parametric information has changed.
SYSTEMS, INFORMATION, AND ECONOMICS 623

Suppose that an actor can respond to environmental cues in accordance

with two quite different sets of equations, that is, two alternative policies.
He is now faced with a choice in addition to stimulus/response at the level of
elemental states. He must choose a policy. If for a given set of states a policy
switch occurs, structural information has changed.
Next, imagine that our agent senses environment (external) and agent
(internal) states and on the basis of these chooses a policy. Given this policy,
stimulus/response occurs. Suppose this choice can be represented by equa¬
tions, too. The overall system is expanded to include a morphogenetic com¬
ponent. Structure changes endogenously; form evolves. Elsewhere, I have
shown how models of this kind can represent essential features of economic
development and provide an endogenous theory of structural change. [Day,
1981a, b.)

CREATIVE INTELLIGENCE

Notice that in modeling the selection of policies by a new set of state-

transition equations, a structural change becomes formally like a parametric
change. This formal similarity disguises the point, however. There is a big
difference in nature between behavior of amoeba, which have quite fixed
patterns of S/R, and those of humans, who can, according to a highly
elaborated socio-politico-economic system, switch monetary and fiscal
policies from, say, Keynesian equations to Friedmanian ones.
Moreover, there seems to be a nonanalytic type of morphogenesis, the
essence of which is poorly represented by equations of any kind so far as I
know: Humans create structure. In shifting from one organization to an¬
other, we sometimes invent something new, something that did not exist
before. Why should we have this capability? No other creatures have it in
more than a rudimentary form.
Before suggesting the answer, let us emphasize how important creativity
is. Economists, focusing as they so often do on equilibrium and compara¬
tive-static theory and influenced as they are by politicians’ demands for the
analysis of immediate effects of policy, forget that the essence of life (and
very likely of stellar existence, too) is evolution: The progression of things
through a variety of forms; new forms emerge where none like them existed.
If conditions are favorable, they grow and flourish. Inevitably, they decline
and die out.
All of this suggests that given forms are globally unstable: Their equations
of motion ultimately carry them outside their bounds of homeostasis. Nature
handles this problem by variation and selection. Conscious individuals,
however, do not wish to be bound by this profligate, merciless process. If,
when their systems and the world about them begin to go awry, they could
only create something new, something that would set them working again so
to speak, then they would continue to exist. Nature need not kill her children
624 RICHARD H. DAY

if only they can create. To contrast this property with economic rationality
in the usual sense, I prefer to use the term creative intelligence. If it'did not
exist, we would not.

UNSTABLE SYSTEMS

From the long view, evolution, not “stationarity,” is the fundamental fact of
life. When individuals cannot adapt, they are killed; when organizations
cannot adapt, they are disassembled. When societies cannot adapt, they are
destroyed, swept aside, covered with sand, swallowed by jungle, or
trampled under by a new regime. Humans are destroying virtually all mam¬
mal species with whom they lack a symbiotic tie. Industrialized societies are
destroying the last vestiges of paleolithic man. The internal-combustion en¬
gine has driven away the horse, the oxygen process has eliminated the
Bessemer process. Everywhere and at every level, we see evidence that
economic systems are globally unstable and that in any given form they
eventually stop working. Just now, concern arises about the instability of the
contemporary world economy. The emergent phenomenon that becomes
most crucial for us is the next form of economic policy. From where does it
come?
It must come from the same attribute of mind that guided the subtle hand
of those stone-chippers of so many centuries ago who invented the
spearhead. The attribute that makes it possible to create an image, to set in
place a form in some way unlike any that came before. This attribute creates
a menu of possible ways of life from which policies can be selected and
behavior carried on. Once in existence itself and perpetuated within the
species, it always operates.
When something new is created, since structural information is changed,
its full implications are seldom perceived. The hunting band may have led
some innovative tribe to a flourishing culture, rich in food, ceremony, and
leisure. Eventually, however, when the last woolly mammoth was butch¬
ered, the last herd of giant bison driven over a cliff, a way of life came to an
end. The inventors of fine spears, bows and arrows unleashed a new spiral of
progress, but one that eventually also succumbed in unexpected ways.
The implications are that: (1) Intelligence creates new forms that at cru¬
cial times make possible a continuation of certain living groups; (2) with
changes in structure, a people may persist for a while; (3) the ultimate
consequences of adaption are unknown, and the new system is also (al¬
ways?) unstable, so that new forms must repeatedly be found. Mor¬
phogenetic man is thus an active participant in evolution, matter that has
come not only to reflect on itself, but also to design the new forms it will
occupy in its trajectory through space and time.
 INFORMATION ECONOMICS,
TEAM THEORY,
AND AGENCY THEORY
Richard Mattessich

This paper should be taken as a supplement rather than a commentary to the

preceding paper by Richard Langlois. Some of these pages were originally
written as a second part of my commentary on Hassan Mortazavian’s paper
in this volume. A request from the editors initiated its conversion into a
separate discussion dealing more directly with the relation between systems
theory and economics. I cannot, under the constraint of time and space, take
a thorough, critical position to Langlois’s point of view. But since his paper
does not deal with the special area of information economics and its exten¬
sions, one may with justification regard my contribution as complementary
to his. However, one point of criticism of Langlois’s view ought to be
expressed at this stage.

METHODOLOGICAL SYSTEMISM VERSUS

METHODOLOGICAL INDIVIDUALISM

I agree with Langlois when he asserts that “systems theory is not at all
methodological holism in any strong sense.’’ Indeed, in a recent comparison
between Bunge’s important work [Bunge, 1979a] and my own systems
methodology, I expressed this view:

Bunge is anxious of not appearing to be a holist. This is understandable in the

face of the absurdity of an absolute or extreme holistic Weltanschauung a la
Hegel. But how can one defend the systems approach (where the interrelations
between system and environment are so prominently emphasized) without
being a relative or moderate neo-holist? To escape this dilemma, Bunge distin¬
guishes three philosophic attitudes: On one extreme he sees atomism together
with ontological reductionism, which both of us, Bunge as well as myself,
reject. The middle ground is occupied by what Bunge calls systemics or sys-
temism, which I addressed (mainly for esthetic reasons) as relative holism or

Additional support from the Social Sciences and Humanities Research Council of Canada is
gratefully acknowledged.

625
626 RICHARD MATTESSICH

neo-holism. The other extreme is addressed by Bunge shortly as holism, which

I prefer to call absolute or extreme holism, the basic idea of which is again
rejected by both of us. But what is the essence behind systemism or relative
holism? I fully agree with Bunge (1979, pp. 1-3, 39-43) when he says that it has
both, a cognitive motive to discover similarities between systems, as well as a
practical motive to cope with the complexities of modern industrial societies.
Its philosophic justification is to be found in the fact that “both systemics
experts and ontologists are interested in the properties common to all systems
irrespective of their particular constitution, and both are intrigued by the pecu¬
liarities of extremely general theories, which are methodologically quite differ¬
ent from specific theories (p. 3). (Mattessich, 1982, pp. 53-54.)

But whether this is identical to saying, as Langlois does, that “systems

theory is in fact a form of intelligent methodological individualism” depends
very much on what one means by this expression. As pointed out by Lukes,
“methodological individualism acquires a range of different meanings in
accordance with how much of 'society’ is built into the explanatory ‘individ¬
ual’.” (Lukes, 1973, p. 601.) Most interpretations of methodological individ¬
ualism attribute to it the view that emergent properties of social institutions
are explainable solely in terms of the behavior of individuals making up
those institutions.1 A systems theorist might object to this view by pointing
out that just as the properties of water cannot be explained by the individual
properties of hydrogen and oxygen, so the behavioral properties of many
social institutions must be taken as basic and cannot be derived entirely
from the behavior of their individual constituents. For example, methodo¬
logical individualism, which dominated neoclassical economics for over a
century, was hardly in a position to anticipate the social costs and dreadful
ecological consequences of modern superindustrialism, precisely because
some of these costs and other consequences are emergent properties of this
particular, novel kind of system. In these environmental abuses lie the
reason why more holistic trends emerged and why the study of ecology and
the entire ecological movement became so prominent. The abuses also ex¬
plain why systems thinking, slumbering for over four decades—from its
inception by Aleksandr Bogdanov [1913 and 1922] to the 1950s or 1960s—
became the rage in many quarters and why neopositivism, not to be
identified with methodological individualism, has experienced such a de¬
cline. If, in the past three decades, the maxims of systemism had been
heeded as much as those of methodological individualism, mankind might
have been spared the disastrous ecological dilemma it is presently con¬
fronted with. In summary, to identify the systems approach with methodo¬
logical individualism, as commonly understood, means misunderstanding an
important aspect of the systems approach: namely, the acceptance of emer¬
gent properties as basic and irreducible.

'The definitions of methodological individualism offered in the two most important standard
works [Dray, 1967, pp. 53—58; Lukes, 1973, pp. 600-601] seem to corroborate this interpreta¬
tion.
 INFORMATION ECONOMICS, TEAM THEORY, AND AGENCY THEORY 627

As I pointed out elsewhere [Mattessich 1978, pp. 53-103, 247-248], tradi¬

tional science offers plenty of evidence of individual aspects of system
thought; in economics, Walras’s general-equilibrium analysis is an example
par excellence. In general, however, traditional economics disregarded such
important system-methodological insights as, for example, the distinction
between renewable and nonrenewable resources—a distinction crucial for a
more holistic theory of income as well as price and capital formation. It is
the merit of the systems approach to have awakened many areas and disci¬
plines to the need for a better understanding of emergent properties and,
hence, a fuller or more systematic treatment of the relation between system
and environment. And in this, methodological individualism seems to have
failed.

THE ADVENT OF INFORMATION ECONOMICS

According to Lamberton, “the economics of information and knowledge . . .

analyzes the processes by which information is produced, diffused, stored,
and used.” (Lamberton, 1971, p. 7.) The pioneers of this broadly conceived
economics of information and knowledge are Kenneth Boulding [1956a;
1966], Fritz Machlup [1962], Jacob Marschak [1954; 1964; 1974, vol. 2],
George Stigler [1961], and others. But information economics (IE) in the
more narrow sense2—particularly relevant to a rigorous systems approach—
originated with Marschak and was further developed by other authors, for
example, Marschak and Roy Radner [1972], Gerald Feltham [1972], Joel
Demski [1972 and 1980], and Demski and Feltham [1972; 1976; 1983]. In this
narrow sense, IE pivots on the evaluation of information systems (and thus,
by necessity, on the evaluation of particular information signals). In other
words, the following question becomes crucial: “Which information system
among alternatives ought to be chosen for a particular information purpose
or situation?” Although IE is far from being capable of answering these
questions in the complex environment of actual practice, important theoret¬
ical insights have been gained for simplified situations. The principal idea on
which IE rests is relatively simple; it emerges from statistical decision
theory as naturally as does a butterfly from the pupa’s cocoon. To reveal
this, a minor modification of the traditional decision model suffices. The
latter3 expresses the expected utility value E{a,) of action a,{i = 1, . . . , n) as
the sum total of all utility values u{ahSj) attributed to each action a, at state
Sj(j - 1, ,m) multiplied by the probability p(sj) with which each state Sj is
expected to occur. And the optimal action a0 to be chosen among the n

"It seems to be advantageous to make a distinction between information (as evidence) and
knowledge (as hypothesis) and, hence, between information economics and knowledge eco¬
nomics. [Mattessich, 1974a; 1978, pp. 226-233.]
m

3E(ai) = ^ u{ahSj) ■ p(sj) and E(a0) = max E(a,).

7=1 '
628 RICHARD MATTESSICH

alternatives results from selecting the highest among the n expected values
£■(«,), hence, from maximizing the latter over i - 1The modification
or conversion of the decision model into an information-economic model
now consists in making the previously hidden or implicit information signal
yk explicit by substituting the conditional probability p(iS/|;y*)—that is, proba¬
bility of event Sj after receiving message yk—for the previous probability
value pisj). Then the expectations of actions a, (or aok in case of the optimal
action, given signal yk) must also be formulated in terms of signal yk.4
Thus, the expected value of action a,, given signal yk, amounts to the sum
total of the products of the utility u(a„Sj) attributed to each action a, at state
sj(j = 1,,m) multiplied by the conditional probability p(s/|;y*) of state Sj
occurring, provided signal yk(k = 1, . . . , r) is received. The expected value
E(aok,yk) of the optimal action aok, given signal yk, is formed by an analogous
maximization choice. But note that first for each signal there may be a
different optimal action (for this reason, the action must now also bear the
subscript of the signal) and second, that aok need not be identical to a0,
otherwise a systematic information system would hardly be required.
But in IE the emphasis obviously cannot be on the optimal action of a
specific single signal yk but must be on an action optimal under all signal
possibilities. Furthermore, one is interested only in the excess or differential
utility that the information system yields over the expected value of the
optimal action E(a0), that is, compared with a situation employing an ordi¬
nary decision model. Finally, it has to be taken into consideration that the
use of the information system involves some costs C/, either by operating it
(say, in a market-research department) or by buying the information from
outside the firm (say, subscribing to a forecast service). Hence, the expected
net value £> of the information system I can be formulated as the summation
of each optimal expected action for each signal minus the expected value of
the optimal action E(ad), that is, without the employment of any information
system, minus the operating cost Cj of the information system.5 But since
there may be many alternative information systems (I = , N), the
expected net value E0 of the optimal information system must be expressed
as maximum among each of the expected net values Eh that is, for each of
the N information systems feasible.6 This requires that the calculation for Ek
be made for each of the N information systems.7

rn

*E(ajyk) = X u(a"sd ' Pi5j\Vk) and E{aok,yk) = max E(ahyk).

j= i

5E, = X E(aok,yk) ■ p(yk) - E(a0) - C,.

k= 1
6E0 = max £/.
/

71 have elsewhere offered an illustration of such a calculation for an agricultural weather¬

forecasting system. [Mattessich, 1978, pp. 219-224.]
INFORMATION ECONOMICS, TEAM THEORY, AND AGENCY THEORY 629

THE SIGNIFICANCE OF TEAM THEORY AND AGENCY THEORY

Of course, as valuable as the evaluation of information systems is for design¬

ers and users of such systems, it is merely one aspect of IE. The intercon¬
nection with system and organization theory (and thus with the supersys¬
tems served by information systems) manifests itself in many ways.
Marschak and Radner’s team theory is one such manifestation:

The economic theory of teams attacks a middle ground. We study the case in
which several persons perform various tasks including those of gathering and
communicating information and of making decisions; but they have common,
not divergent, interests and beliefs. Hence the optimality requirement is easily
defined, just as in the case of a single person. But the single person’s problem
of optimizing his information instrument and its use is replaced by that of
optimizing the allocation of tasks among the members of the team.

We have equated economical and efficient to denote an arrangement that is

most desirable (or, in the general case of organization, one that is viable),
under given feasibility constraints. These constraints include the limitations of
human capacities for communication and good decision-making (and analogous
limitations of inanimate instruments). While our concern is a practical, purpo¬
sive, prescriptive one, the general solutions we discuss would depend, when
applied to any concrete case, on data supplied by workers in descriptive
fields—the names of R. Cyert, J. March, and H. Simon come to mind!181—
psychological or sociological data along with those of natural technology. On
the other hand, our results may prove of some value to descriptive theorists of
human organization by pointing to those data, quantitative and otherwise, that
would be of most importance if one wanted to increase the efficiency of a given
organization. (Marschak and Radner, 1972, pp. 4-5.)

Another, more recent and perhaps more fashionable manifestation of the

interface of information economics with organizational and even contractual
aspects, is the broad area of agency theory. [Jensen and Meckling, 1976;
Hirshleifer and Riley, 1979; Baiman, 1980-1984; Feltham, 1984.] Origi¬
nally, this theory may have been regarded as merely one of the more recent
attempts in the perennial search for a more realistic theory of the firm.8 9 But
the widespread response it is presently receiving in several quarters of ap¬
plied science, such as accounting theory, finance, management science, and
so forth, seems to secure it a special position. Agency theory is concerned
with relations between control and information for the purpose of determin¬
ing the most preferable feasible contract between a principal (who could be
either the shareholders of a company or its single owner or a manager vis-a-
vis his or her subordinates or even an insurance company) and his or her
agent (management vis-a-vis the owners, a subordinate, or an insured) en-

8See, for example, Cyert and March [1963] and March and Simon [1958].
9For former attempts and the controversy surrounding them, see Machlup [1967; 1978, chap.
16, pp. 391-423.]
630 RICHARD MATTESSICH

tering into a mutual agency contract that delegates to the agent the manage¬
ment of some kind of entity in compensation for which he or she receives a
portion of the entity’s profit. Depending on the contract, the portion may
range from a fixed amount to a percentage of the profit, with possible penal¬
ties for imperfect fulfillment of the contract establishing some risk-sharing
between principal and agent. Both of them are supposed to work toward a
common goal, but since each of them is trying to maximize his or her own
profit or utility (possibly at the cost of the other party), a conflict of interest
is likely to emerge.
For this reason, an information system is required that informs principal
and agent about the entity’s total profit and supplies further information to
facilitate the agent’s managerial task and the principal’s control of the agent.
As long as the principal can observe the agent’s activities, no moral hazard
arises, and the optimal contract stipulates a fixed remuneration to the agent
with penalties whenever he or she is found to be remiss in his or her duties.
Often the principal’s monitoring of the agent’s activity is too expensive or
cumbersome. In this case, the ultimate profit shares of principal and agent
may not only depend on the total profit, but also on the particular differ¬
entiated information available to each of the two parties.
Applied to the situation of a newly formed company, this means that the
cost of raising equity capital (and the expected return) may be considerably
influenced by the kind of control-and-information system available to share¬
holders. Thus, it may be in the interest of promoters and managers to have
either a voluntary or institutionally enforced system of accounting and audit¬
ing regulations imposed on them, because the lack of directly monitoring
their activity otherwise results in higher shareholders’ risk. Such regulations
could lower the cost of raising capital and increase management’s own por¬
tion of the profit. Hence, the combination of agency theory and information
economics may well be relevant to such practical issues as the ongoing
legislative or quasilegislative activity of the Financial Accounting Standards
Board of the United States.
For these reasons, agency theory fits well into the general program of
information economics, since the value of individual information as well as
the entire information system plays a crucial role in this theory. The theory
also constitutes an obvious link between the information system and the
systemic or organizational structure of the entity involved. Here again, one
encounters rigorous mathematical presentations in the service of an empir¬
ical discipline with the ultimate aim of creating testable or at least refutable
hypotheses.10

10The pertinent literature is vast, and the bibliography arranged by Demski and Feltham [1981]
comes close to a thousand articles and books. But to be more specific, I list the following
relevant publications (beyond those previously indicated): Atkinson and Feltham [1980]; Arrow
[1974]; Barnea, Haugen, and Senbet [1981]; Dickson, Senn, and Chervany [1977]; Fama [1980];
Hilton [1981]; Holstrom [1979]; Itami [1977]; Mattessich [1975; 1983 forthcoming]; Ng and
Stoeckenius [1979], and Zmud [1979],
 ON THE RECEPTION OF NOISE
A Rejoinder

Richard N. Langlois

It is in many ways comforting to an author when the response his work

elicits is wildly divergent in perspective, focus, and tone. The comprehend¬
ing responses weigh against the misunderstanding ones; the supportive bal¬
ance the critical; the polite counteract the not-so-polite. And, at the very
least, the diverse set of responses my paper has called forth suggests that the
paper’s deficiencies were not entirely systematic ones.
What this diversity of response also illustrates, I believe, is that there are
indeed as many versions of systems theory as there are systems theorists. If
C. West Churchman is right, systems theory includes everything this side of
the I Ching.' Such diversity is really all I was trying to emphasize in my
evidently provocative suggestion that systems theory had failed in its sup¬
posed quest to unify all of the social sciences.
There are, of course, a few common themes of criticism among my com¬
mentators, and I will turn to these shortly. Let me begin with the happier—
but, as is usually the case, briefer—task of discussing those commentators
who have largely agreed with me.
Foremost among these is Richard Day, whose analysis I can endorse
almost without reservation. He explains very clearly a number of ideas I was
struggling to express. Of particular importance is Day’s treatment of the
form of nonanalytic morphogenesis that arises from creative intelligence;
this is a topic to which I will return before this rejoinder is over.
I also found much to agree with in Robert Beam’s piece. Unlike Walter
Buckley,2 he does seem to understand what the example of the “cybernetic

'Although Churchman’s piece is printed as a comment on Mortazavian, it also refers to my

paper; so it does seem appropriate to include Churchman peripherally in this rejoinder.
2 Since I do not deal with this aspect of Buckley’s comment elsewhere, perhaps a few lines are in
order here. I find little to disagree with in Buckley’s paragraphs dealing with information and
meaning—except his characterization of my own position. If one were to read Buckley and not
my original paper he or she would think I am some sort of behaviorist with a mechanistic
conception of meaning. In fact, my whole point in saying that meaning in very simple systems—
whether one wishes to call them stimulus/response, cybernetic, or whatever—is reducible to

631
632 RICHARD N. LANCLOIS

farmer” was all about, and his discussion is a successful if slightly jar¬
gonized expansion of what I was trying to say. Also, his treatment of the
important and often-neglected distinction between formal and informal or¬
ganizations is excellent, even if, with Friedrich Hayek, I would prefer to
avoid confusion and call the latter not an organization at all but an “order.”
[Hayek, 1973, pp. 36-40.]
Beam asks whether I would concur that models of social interaction
should involve both “the science of communication and the science of trans¬
action.” Absolutely. The economics of information, in particular, should be
precisely a study of both communication and transaction. I am a little trou¬
bled, though, with the idea that the analysis of transactions should rest
heavily on the concept of power. If anything, it seems to me, it should be the
other way around: Power, insofar as it appears as a social phenomenon, is
something that needs to be explained in terms of the structures of transac¬
tions involved. Economics has lately turned renewed attention to the idea of
the transaction as a unit of analysis, a trend that has opened up some inter¬
esting new areas of inquiry. [O. E. Williamson, 1979; 1980.]
Now, let me turn to those who found more fault with my presentation.
Both Buckley and Richard Mattessich are still bothered by an issue I was
trying to defuse: whether systems theory differs from methodological indi¬
vidualism in its desire to consider the whole greater than the sum of the
parts. I probably should have realized that my negative answer would raise
some hackles; there is much intellectual capital invested in the proposition
that systems theory is less reductionistic than methodological individualism
and more attentive to emergent phenomena. There may, in fact, be subtle
distinctions to be worked out in principle, but the important issues will
continue to lie elsewhere.
Systemism and methodological individualism are not (necessarily) identi¬
cal. (I had thought I was clear on this point.) But what distinguishes them is
not a debate about whether properties of the whole can be deduced solely
from properties of the parts; rather, it is a debate about what shall be consid¬
ered the parts. This way of putting things should raise even more hackles.
For the methodological individualist’s response—that in the social sciences
the only acceptable parts are the knowledge, expectations, intentions, and
actions of human beings—calls into serious question whole classes of fash-

action is to illustrate the extent to which it cannot be so reduced in more complicated systems.
Some of the confusion may have arisen from my portrayal—for heuristic reasons—of the
decision-making farmer as a real person. What I was actually trying to illustrate was the ideal
type of an economic agent as portrayed in a mathematical maximization model. In such a
model, meaning is indeed “the mapping between the learned properties of rain and the range of
possible consequences of rain for the farmer’s particular goals and interests at the moment”;
but in the model farmer, this is exactly a mapping of acreage into expected profit, because the
way in which I have constructed the model rules out any changes in internal states or values and
goals. I heartily agree that this is too narrow a definition of meaning in economics: That is one of
the central points of my paper.
RECEPTION OF NOISE 633

ionable systems models in which physical quantities impinge directly on one

another as if without the intervention of human minds.
One way to see the issue is to recognize, as Mattessich and Buckley
apparently do not, that prediction and explanation are not identical, at least
not in the social sciences. While it is certainly true that one cannot predict
the behavior of water solely from the properties of hydrogen and oxygen
considered separately, it is nonetheless true that, contra Mattessich,3 one
can—in fact must—explain water with reference to hydrogen and oxygen.4
That is what chemists always do; there is no other way to explain water.
Similarly, the social sciences are concerned almost exclusively with phe¬
nomena that are the unintended (emergent) byproducts of human intention.
But those phenomena, once recognized, can and must be explained in terms
of human intention even if their behavior could not have been predicted from
considering the individual’s intention alone.5
This has some fairly radical implications. Casting the problem in these
terms makes it clear that far from being less reductionistic than methodolog¬
ical individualism, systems theory is often in danger of being far more reduc¬
tionistic. Although the methodological individualist believes that social
wholes must be explained (though not predicted) in terms of human actions,
he or she does not believe that such wholes should be explained in terms of
more basic or more material components such as energy, dollars, or even
homogenized “labor value.”
As a consequence, methodological individualism is skeptical of the many
system models that take this materialistic and—yes—reductionistic ap¬
proach. Good examples of such models are the various energy-analysis mod¬
els that attempt to draw conclusions about the allocation of resources and
the value of various commodities solely by considering something called
“embodied energy content.” (Costanza, 1980.)6
Mattessich suggests that methodological individualism somehow missed
the “important system-methodological distinction” between renewable and
nonrenewable resources. In the first place, system theory per se is a very
general theory and has nothing whatever to say about whether in a particular
application one should distinguish between the renewable and the non-

3 Buckley also seems to confuse explanation and prediction in this way when he writes that “we
cannot begin to explain or predict the properties or behaviors of complex dynamic wholes from
a knowledge of the parts. ...”
4For a good discussion of exactly this hydrogen and oxygen example, see Dennis C. Phillips
[1976, pp. 14-15, 32-33],
5It is in this explanatory sense that the social sciences, as Buckley suggests, should be synthetic
rather than analytic. But Anatol Rapoport was not the first to argue this position. In words first
published in the 1940s, Hayek wrote that: “While the method of the natural sciences is . . .
analytic, the method of the social sciences is better described as compositive or synthetic.”
(Hayek, 1941 and 1979, pp. 66-67.)
61 have referred to Robert Costanza’s article as an example because it was certainly considered
system-theoretic in some circles, as evidenced by the fact that it was reprinted in the General
Systems Yearbook series.
634 RICHARD N. LANGLOIS

renewable. In order to claim the necessity of this distinction, one would

have to invoke some ancillary methodological doctrine. For example,
methodological individualism added to systems theory might point to the
importance of various distinctions about the knowledge possessed by social
agents. By analogy, it is not systems theory by itself but something like
methodological resourcism or methodological ecologism that Mattessich is
implicitly invoking. And, again, this is not more holistic than individualism
but less, since it calls for the reduction of social phenomena to natural
resources, that is, for the explanation of social phenomena in terms of natu¬
ral resources.
In the second place, methodological individualists are by no means unable
to deal with emergent phenomena like “the social costs and dreadful ecolog¬
ical consequences of modern superindustrialism.” Economics has a well-
developed theory of external costs—in fact, I am aware of no other field that
has a similar theory. [See, for example, Baumol and Oates, 1975.] Econom¬
ics tackles these emergence problems head on—it simply comes up on occa¬
sion with answers different from those the methodological resourcist would
like. Indeed, some economists might even go so far as to question what
Mattessich seems to view as an established fact if not as a self-evident
truth—the existence of a “disastrous ecological dilemma” facing mankind.
In a certain sense, I do agree that the methodological doctrine one
adheres to colors the way one sees a problem or a phenomenon. As Buckley
put it, what is meaningful to one person—say, Chinese—may be noise to
another who has not studied Chinese. Similarly, someone who studies social
phenomena with a particular set of parts in mind will very likely see the
behavior of other parts as mere noise. The danger is least when the parts are
human beings, since, if one builds the models right, whatever is important to
human beings—be it resources, pollution, or whatever—should find its way
into the models through them. But if the parts are something like resource
flows, then there is serious danger of neglecting human elements altogether:
For while humans may care about resources, resources care very little about
humans.
And this is why economic and other individualistic models are frequently
far less apocalyptic than their materialist counterparts: They take into ac¬
count human ability to adapt and innovate, a capacity that Day described so
well. Humanist models have a resource that resource-minded models invari¬
ably overlook.

The impact of man in the evolutionary process arises because of the capacity of
his images—that is, the knowledge present in his mind—to grow by a kind of
internal breeder reaction: the imagination. It is this which has given the human
nervous system such a fantastic social-evolutionary potential, a potential of
which we have probably hardly used up one per cent in the brief history of the
human race. (Boulding, 1964, p. 141.)
RECEPTION OF NOISE 635

Which brings us to Adam Smith, another source of raised hackles among

my commentators. The notion that Smith anticipated modern cybernetics is
scarcely novel to me. The biologist Garrett Hardin—who made popular the
idea of a “tragedy of the commons” [Hardin, 1968]—also suggested that

long before Claude Bernard, Clerk Maxwell, Walter B. Cannon, or Norbert

Wiener developed cybernetics, Adam Smith has just as clearly used the idea in
The Wealth of Nations. The “invisible hand” that regulated prices to a nicety
is clearly this idea. In a free market, says Smith in effect, prices are regulated
by negative feedback. (Hardin, 1961, p. 54.)

In fact, what I was trying to say is something rather different. What Smith
was getting at in his conception of an “invisible hand” is not the idea that
given an appropriate set of legal institutions, the market will produce homeo¬
stasis governed by negative feedback. What Smith argued was that given an
appropriate set of institutions, the free exchanges of individuals will lead to
the production of wealth—to continual and cumulative change in organiza¬
tion and technology. [Bohm, 1982.] For Smith, the unintended conse¬
quences of individual action within what he called “the system of natural
liberty” would be in the nature of evolutionary morphogenesis. [Adam
Smith, 1776 and 1936, p. 651.]
As I tried to emphasize in my paper, it is important to get straight which
kind of a system one thinks the economy is. If one does not conceive of it as
an evolving order but as a cybernetic system in need of central direction,
then one is likely to interpret change and innovation not as morphogenesis
but as chaos and disorder.
But the choice of which lens one uses to view the economy (or society
more generally) is not arbitrary. As a historical matter, Buckley is dead
wrong when he says that Western economic systems have been generally
feeble and unstable. It may or may not be true that the economicopolitical
system is unstable. The free institutions on which the market economy is
based are the exception not the rule in history. But when those institutions
were more or less firmly in place—as in Western Europe and America during
most of the last 200 years—the result, as even Marx recognized, was the
greatest outpouring of human creative intelligence and energy in the history
of mankind. There were certainly recessions and depressions (some of
which were arguably caused or aggravated by attempts at central control),
but it is a perversity of perspective to call the system feeble or unstable.7

71 was pleased in this context to see Churchman point out the fallacy of seeing world poverty
and starvation as basically a food problem. The problem is an institutional one—a problem of
setting up in Third World countries the kinds of social and economic institutions conducive to
the production of wealth. But I was distressed at his explanation in terms of superp°wer
politics, although the difficulty may have arisen from the brevity with which he expressed
himself. As stated, his explanation is an instance of what Karl Popper called “the conspiracy
636 RICHARD N. LANGLOIS

The tragedy of the commons can indeed be a significant phenomenon in

free economic systems. But far from being endemic in the institutions Smith
advocated, the tragedy of the commons results precisely when those institu¬
tions are absent.* * * * * * * 8 Economists have recognized clearly for at least 20
that tragedy-of-the-commons problems and so-called externality problems
more generally can always be traced to problems in the existence and
definition of property rights. [Coase, I960.] The implication is that such
externalities by no means imply the necessity of central direction or control;
they may in most cases simply call for the creation, clarification, or
redefinition of property rights.
But Buckley’s point is not merely that a more-or-less free economy—with
or without adequately defined systems of property rights—will produce bad
outcomes (like too much pollution or too much grazing of rangeland). Buck¬
ley’s claim is that such an economy is unstable. It is in putting the matter this
way that Buckley betrays the attitude I called thermodynamic Manichaeism.
He likens the economy to a predator/prey system, which he describes as
uncontrolled. This, he says, explains the increasing attempts at central eco¬
nomic and social control, which, however, have been largely frustrated by
the noise of the independent goal-actions of individuals. I am afraid I find
this explanation of economic reality nothing short of chilling.

theory of society,” a mode of social explanation more primitive than most forms of theism.
[Popper, 1966, pp. 94-95.] The reason this mode of explanation fails, as Popper notes, is that we
live in a complex world in which the prevalence of unintended consequences makes power a
complicated affair and a generally implausible explanation. It is not clear that countries would
all know what to do—or could do it if they knew—in the absence of superpowers. Neither is it
clear that superpowers can prevent nations so determined from setting up wealth-producing (or
at least starvation-preventing) systems. There are a number of countries (Taiwan and South
Korea) that are certainly the focus of superpower attention but have nonetheless been able to
set up fairly successful wealth-producing institutions. And there are dozens of forgotten coun¬
tries entirely unable to change their behavior patterns in the face of starvation.
8The political order Smith advocated was never literally laissez faire; it was, in fact, based on a
common-law framework involving property rights and a strong conception of justice. His claim
was that when these institutions are in place, people left to themselves will produce good social
outcomes. In more system-theoretic terms. Smith was arguing that good social outcomes would
be the emergent result of individual action within the constraints of certain boundary condi¬
tions. If the catch-phrase laissez faire meant anything, it meant that the economic system should
not be viewed as requiring external central control in the manner of an aircraft or spaceship.
This is the control-theoretic content of the old slogan “a government of laws and not of men.”
Moreover, Smith’s advocacy of the “system of natural liberty” was based on its justice proper¬
ties; that it led to good economic outcomes was for Smith simply a happy coincidence.
[Buchanan, 1979.] The justice properties of Smith’s system, incidentally, are closely related to
those implied in the Kantian imperative cited by Churchman. I was very pleased to see this
reference to Kant, but I was also a little surprised to see this imperative discussed favorably in
an article extolling rationalist planning—since, to most philosophers, the Kantian imperative is
taken as implying individual rights that, if not exactly inconsistent with rationalist planning, are
nonetheless conceived of precisely as strong constraints to, or protections against, the central¬
ized implementation of rationalist plans. [Nozick, 1974; Fried, 1978; Dworkin, 1978.]
RECEPTION OF NOISE 637

The predator-prey uncontrolled system is simply the wrong lens through

which to view a more-or-less free economic society. Aspects of the system
may behave in the predator/prey fashion; in particular, competition among
business firms (or, more generally, among ideas or patterns of economic
behavior) may usefully be cast in terms of ecological niches. But while any
particular firm (or idea or pattern of behavior) may fail to adapt to change,
the economic system as a whole is not therefore any more unstable or
nonadaptive than is the whole process of biological evolution when a partic¬
ular species fails to survive a change. Systems-dynamic models of the preda¬
tor/prey sort are often useful, but they do not accurately capture the full
logic of an evolutionary system. [Boulding, 1978.]
That someone should see the economy or society as an unstable or uncon¬
trolled system is so frightening precisely because—as Buckley unabashedly
puts it—the free choices of individuals do not appear to him as consistent
with order; they are mere noise. I doubt that I need elaborate on the political
implications of such a perception.
All of this illustrates, once again, the extent to which one’s methodolog¬
ical starting-point influences the results of the final analysis. If one reacts to
the words and phrases of spoken Chinese as if they were mere noise, it is
quite probably because one has not studied Chinese. And if one sees the
independent goal-actions of men as mere noise, it may well be for the analo¬
gous reason.
EPILOGUE
 SEMANTIC QUIRKS IN
STUDIES OF INFORMATION

Fritz Machlup

Semantic and methodological discussions are to the ears of some scientists

what symphonic and chamber music is to the tone-deaf. They do not hear
anything that makes sense or seems to matter. Composers have not at¬
tempted to dedicate tone poems to the tone-deaf; not-so-wise students of the
philosophy of science have not given up their ambition to write for scientists
who disdain what they regard as “splitting hairs.” There are probably de¬
grees of methodophobia; as an optimist, I continue to believe that a lively
exposition can succeed in convincing some doubters that there may, after
all, be something to these quibbles about quirks.
The quirks on which I intend to quibble in these pages are semantic but
also methodological. A discussion, for example, of the meanings of “sci¬
ence” and “scientific” cannot help being a melange of semantic decisions
about methodological positions. In general, whereas purely historical se¬
mantics may do without methodology, analytical semantics of scientific
terms will usually contain a large dose of methodology.
My remarks are organized under three main headings: What They Mean
by Information; What They Mean by Science; and What They Mean by
Computing.

WHAT THEY MEAN BY INFORMATION

The pronoun “They” in this heading does not refer just to the members of
the diverse cultures interested in the study of information but it refers to all
who use the word. If a scholar adopts a word from everyday life to denote

This manuscript was unfinished at the time of Fritz Machlup’s death; it was prepared for
publication by Una Mansfield, to whom he had handed over the completed parts. Missing
subsections are indicated by an Editor’s Note at the point where they would have appeared in
the section “What They Mean By Science.” Fortunately, the subjects missing from this text
have been treated elsewhere by Machlup, and scholars interested in knowing more of his views
should consult the volume Knowledge and Knowledge Production. [Machlup, 1980.]

641
642 FRITZ MACHLUP

some specific “designata” different from the common meanings, the latter
are not thereby extinguished. They remain in the dictionary and we have no
right to disregard them.

Dictionary Definitions

The original meaning of the word ‘'information” in several modern lan¬

guages comes from Latin, where informare means ‘‘to put into form”; thus,
according to the Oxford English Dictionary, the verb ‘‘to inform” means ‘‘to
form (the mind, character, etc.) esp. by imparting learning or instruction”;
but more frequently, ‘‘to impart knowledge of some particular fact or occur¬
rence to;1 to tell (one) of something.” The noun ‘‘information” has essen¬
tially two traditional meanings: (1) ‘‘the action of informing; the action of
telling or [the] fact of being told of something,” and (2) “That of which one is
apprised or told; intelligence, news.” Any meanings other than (1) the telling
of something or (2) that which is being told are either analogies and
metaphors or concoctions resulting from the condoned appropriation of a
word for something that had not been meant by earlier users.
Many definitions of information are the result of limiting the something
that is being told to something previously unknown to the person or persons
getting informed or previously known with less confidence. Accordingly,
information may become equivalent to news, directions, instructions, ad¬
vice, confirmation, or reconfirmation. Such restrictions to novelty, surprise,
or more assuredness, however, are quite arbitrary; this becomes clear as
soon as one realizes that besides new information, there may be old informa¬
tion, repeated information, and expected information, and therefore no logi¬
cal justification for excluding these by dictatorial or persuasive definitions.

Information and Knowledge

Teachers of information science and computer science have sometimes

found it suitable to their purposes to instruct students to observe a hierar¬
chy—in ascending order—within the triad: data, information, and knowl¬
edge. Historical semantics gives only qualified support to such proposals.
A close and firm link between information and knowledge has always
existed, and most dictionaries define information as a certain kind of knowl¬
edge. Some have the word information denote “a transfer of knowledge” or
“a form of knowledge” or “a piece of knowledge.” Distinctions between
information and knowledge have been proposed chiefly on three scores: (1)
Information is piecemeal, fragmented, particular, whereas knowledge is

1 Lest the slightly archaic word “to impart” is misunderstood, its definition is “to give a part or
share of; to bestow, communicate as knowledge or information; to make known, tell, relate
. . . .” Hence, it always refers to an activity, not just an opportunity of being observed. Learn¬
ing about something by looking should not be confused with learning by being told.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 643

structured, coherent, and often universal. (2) Information is timely, transi¬

tory, perhaps even ephemeral, whereas knowledge is of enduring
significance. (3) Information is a flow of messages, whereas knowledge is a
stock, largely resulting from the flow, in the sense that the “input” of infor¬
mation may affect the stock of knowledge by adding to it, restructuring it, or
changing it in any way (though, conceivably, information may leave knowl¬
edge unchanged).2 These distinctions are mutually compatible, though they
often relate to different aspects of the cognitive processes or states involved.
None of the distinctions relates to practical usefulness; neither knowledge
nor information needs to be useful or valuable to merit its designation.
Indeed, the fact that people speak of “useless information” and “useless
knowledge” indicates that usefulness is no definitional criterion. [Machlup,
1982, pp. 8-11.] Nor is it a requirement of normal language use that informa¬
tion be correct and knowledge be true. Indeed, we all have learned that
information may be misleading or downright incorrect or false; and philoso¬
phers of science have taught us that we must never accept empirical knowl¬
edge as definitely true. For a few decades linguistic philosophers dominated
the philosophy departments of some great universities and preached the
“justified true belief” definition of knowledge. [Machlup, 1980, pp. 37-38,
43, 114-120.] This definitional requirement clashed with the more sensible
verdict of other philosophers that “all human knowledge is uncertain, inex¬
act, and partial.” (Russell, 1948, p. 507.) To assume that scientific knowl¬
edge is always true is to violate a truly scientific attitude.
Another difference between information and knowledge may deserve em-

2This third distinction has been most interestingly elaborated by Boulding: “. . . we cannot
regard knowledge as simply the accumulation of information in a stockpile, even though all
messages that are received by the brain may leave some sort of deposit there. Knowledge must
itself be regarded as a structure, a very complex and frequently quite loose pattern, . . . with its
parts connected in various ways by ties of varying degrees of strength. Messages are continually
shot into this structure; some of them pass right through its interstices (. . .) without effecting
any perceptible change in it. Sometimes messages ‘stick’ to the structure and become part of it.
. . . One of the most interesting questions in educational theory and practice is under what
conditions does information ‘stick’ in this way, and under what conditions does it fail to ‘take
hold’ of the structure of knowledge. . . . Occasionally, however, a message which is inconsis¬
tent with the basic pattern of the mental structure, but which is of such a nature that it cannot be
disbelieved hits the structure, which is then forced to undergo a complete reorganization.”
(Boulding, 1955, pp. 103-104.)
Manfred Kochen made a statement about “[w]ays to integrate the growing body of informa¬
tion into a commonly shared fabric of knowledge.” (Kochen, 1970, p. 44.) This may sound as if
it were the same as Boulding’s view but really is not. Kochen evidently thinks of a social “body
[stock] of information” and a social “fabric of knowledge,” as if the sum of individual minds
could be hypostatized into an “aggregate mind” of the whole society, or perhaps an “electronic
encyclopedia” that includes all fragmented and obsolete piecemeal messages that once con¬
veyed timely information of transitory relevance. There is no growing body of information, for
what is not integrated into knowledge is thrown out or forgotten. There is, of course, a growing
flow of information and some of it is put into storage in computer systems. Whether the eventual
output is information or knowledge depends on the individual users.
644 FRITZ MACHLUP

phasis: Information is acquired by being told, whereas knowledge can be

acquired by thinking. In a way, this notion is implied in the third of the listed
distinctions, with the restructuring of knowledge induced by new informa¬
tion. The point, however, is that the rethinking may be only unconsciously
related to information previously received or it may be entirely unrelated to
received information. Any kind of experience—accidental impressions, ob¬
servations, and even “inner experience” not induced by stimuli received
from the environment—may initiate cognitive processes leading to changes
in a person’s knowledge. Thus, new knowledge can be acquired without new
information being received. (That this statement refers to subjective knowl¬
edge goes without saying; but there is no such thing as objective knowledge
that was not previously somebody’s subjective knowledge.)
Going back to the three earlier distinctions, one may ask whether the
featured differences can be identified at the time information is received or
only after some thinking about it. When school children are told that there
will be no school tomorrow, this news item is immediately cast as informa¬
tion and will ordinarily not be characterized as new “knowledge,” although,
of course, the children “know” it. When they are told that 7 x 8 = 56, this
will usually be characterized as an integral part of a coherent and enduring
knowledge structure. But the borderline is sometimes blurred. When people
are told about an epidemic disease, a new virus, and available vaccination,
they receive both information and knowledge. When a new discovery or a
new theory is announced in newspapers and news broadcasts, this will be
information to most recipients but new knowledge to specialists in some
fields of inquiry. Still another possibility is that what someone first regards
as a fragmented piece of information of only transitory relevance will even¬
tually, perhaps much later, turn out to have initiated a significant change in
his or her knowledge structure.
To sum up, information in the sense of telling and being told is always
different from knowledge in the sense of knowing: The former is a process,
the latter a state.3 Information in the sense of that which is being told may be
the same as knowledge in the sense of that which is known, but need not
be the same. That even in everyday parlance people sense a difference can
be seen from the fact that in railroad stations, airports, department stores,
and large public buildings we expect to find a booth or counter marked
Information but never one marked Knowledge. Similarly, in our modern
economy we find many firms selling information services, not knowledge

3In an earlier discussion of the “Uses, Value, and Benefits of Knowledge,” I used the analogy
of a transport and delivery service in relation to the object transported and delivered: “To use
information—a process, mind you—is to listen, to look at, to read; in short, it is its reception
and, if possible, the full or partial understanding by the recipient. The use of the knowledge
conveyed is something else. The act of delivering is one thing, the object delivered is another
.... Use of mode of transportation and use of the transported object are different things.
Likewise, use of a mode of information should not be confused with the use of the message or
knowledge conveyed.” (Machlup, 19796, pp. 63 and 65.)
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 645

services. On the other hand, we would frown on education programs that fill
the student’s head with loads of information: We want them to disseminate
knowledge of enduring value and to develop a taste or thirst for more knowl¬
edge, not just information.4

Information and Observation

Many a scientist uses observation and information as equivalent terms. It is

easy to understand that an observer may feel that his observation, or the
object of his observation, “tells” him something; most of these objects,
however, are dead, and neither able nor willing to tell anything to anybody.
To speak of information in this case is just a metaphor. If the object of
observation is a human artifact, genuine information may be involved if, for
example, an old inscription invites deciphering. The inscription was evi¬
dently designed to inform; hence, an attempt to read and interpret it is part
of a process of information. Information takes at least two persons: one who
tells (by speaking, writing, imprinting, pointing, signaling) and one who
listens, reads, watches.
The metaphoric use of “interrogation” and “information” for “ex¬
perimentation” and “observation” has been frequently proposed. For ex¬
ample: “I see no reason why what is learned by direct observation of the
physical environment should not be regarded as information just as that
which is learned by observing the marks on a document .... The primary
source of scientific information is nature itself.” (Brookes, 1974, pp. 142—
143. Emphasis in the original.) I, for one, see strong reasons for rejecting a
confusion between “observing” and “reading” or “being told.”
The reference to reading invites comment on another source of confusion:
Readings from instruments in a scientist’s laboratory, the reading of data
from his records of observation, and the reading of a report he has prepared
on his work, findings, or conclusions, are three different things. Only the last
of these three instances of reading constitutes information in the basic sense
of telling about something. The first, the instrument readings, are only
metaphoric information: They are part of the observation of pointers in
apparatuses designed on the basis of preconceived theories; neither the
instruments nor the pointers are “telling” anything, though their reading
may furnish valuable clues to the observer and analyst. The second instance,
the reading of the data, may be part of a process of information, namely, if
the data were developed by others, as, for example, when technicians tell
the investigator what they have observed. If, however, the scientist reads
his own data, his own notes, this reading is part of his work as observer and
interpreter. It would be misleading to say that he tells something to himself,
or that the data tell him anything: In actual fact, he merely looks, thinks,

4The last paragraph reproduces what I wrote in chapter 1 of volume 3 of my projected ten-
volume series. [Machlup, 1984.]
646 FRITZ MACHLUP

rethinks, interprets, infers. Only when he tells his fellow scientists about his
work—by word of mouth, by a written report, or by a published paper—is
genuine information (in the original sense of the word) being conveyed.
To use the words information and observation as synonyms is bad lin¬
guistic practice; it leads to confusion between different sources of knowl¬
edge and is therefore epistemologically and methodologically unsound. The
fact that eminent scientists have sometimes been semantically imperceptive
and have metaphorically used the term information in lieu of observation is
no good reason for emulating them.5

Information and Data

The use and misuse of the term data is due, in part, to linguistic ignorance.
Many users do not know that this is a Latin word: dare means “to give”;
datum, “the given” (singular); and data, “the givens” (plural). Data are the
things given to the analyst, investigator, or problem-solver; they may be
numbers, words, sentences, records, assumptions—just anything given, no
matter in what form and of what origin. This used to be well known to
scholars in most fields: Some wanted the word data to refer to facts, espe¬
cially to instrument-readings; others to assumptions. Scholars with a hy-
pothetico-deductive bent wanted data to mean the given set of assumptions;
those with an empirical bent wanted data to mean the records, or protocol
statements, representing the findings of observation, qualitative or quantita¬
tive. With this background of historical semantics, a reader of recent
definitions of, or statements about, data cannot help being appalled.
There are writers who insist that data consist entirely of numbers.6 That
this restrictiveness is (fortunately) exceptional can be inferred from the fact
that the majority qualifies the noun and speaks of numerical data and statisti¬
cal data where the givens are sets of numbers. Such numbers are rarely
given in nature, but instead have to be produced by employing sophisticated
techniques using instruments that sometimes are built on the basis of rather
complex theories. A good deal of prior knowledge has gone into the con¬
struction of measuring devices that are to furnish the numerical data wanted.
Yet, we often read of “raw data” being used to produce information and

5Among the metaphoric users is Claude Bernard: “We must observe without any preconceived
idea; the observer’s mind must be passive, must hold its peace; it listens to nature and writes at
its dictation." (Bernard, 1865 and 1927, p. 22. Emphasis added.) The statement is wrong in
more than one respect, but especially in the phrase I emphasized.
As a curiosum I may cite a phrase that compounds the confusion by speaking of “macro-
scopically observable information,” where “information” evidently stands for “phenomena.”
The same writer “informs” us wrongly that “entropy is related to ‘missing information’ ”; what
is missing is the possibility of observation. [Shannon and Weaver, 1949.]
6“The corpus of information . . . consists of two types of information—non-data and data.
Non-data is non-numeric .... Data, on the other hand, is numeric, highly formatted and results
from analysis.” (Dolan, 1969, p. 41.)
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 647

eventually knowledge.7 Perhaps raw data are only relatively raw, in that
they are inputs for the production of more highly fabricated kinds of infor¬
mation.8 Many writers prefer to see data themselves as a type of informa¬
tion, while others want information to be a type of data.9
One can probably find quotations supporting all possible combinations of
the three terms or of the concepts they are supposed to denote.10 Each is said
to be a specific type of each of the others, or an input for producing each of
the others, or an output of processing each of the others. This semantic
muddle, however, need not cause any serious trouble, because the argu¬
ments in which data, whatever they are, play a central role are relatively
simple: Data entry, data storage, data retrieval, data processing, data ser¬
vices, and all the rest, refer simply to things fed into a computer. These
things, now data from the point of view of the programmers, operators, and
users of the computer, need not be data in any other sense. Drafts of a
manuscript for a learned monograph (or of a mystery story, for that matter)
may have been typed into the computer; or the subject index for a textbook;
a bibliography of writings on the history of French painting; detailed statis¬
tics of the gross national income of the United States from 1940 to 1980;
expert knowledge for the diagnosis of diseases of various kinds; graphs and
images of all sorts; or what not. Most of these inputs, now in the memory of
the computer system, are very far from being raw data in the sense of
empirical scientific analysis. Whether they are pieces of fragmented informa¬
tion—say, the football scores of American college teams for Saturday,
November 13—or whether they are elaborate compendia of systematic
knowledge in some discipline, for the users of these materials accessible in
the computer memory they are data stored and data retrieved.
Of course, these materials are not stored in the form in which they were
typed, with words and numbers; instead, they are encoded by means of
digital data representation. Data transmitted from a computer are “repre¬
sented by the presence or absence of an electrical impulse (representing a bit
‘on’ or a bit ‘off’)-” (Shelly and Cashman, 1980, p. 86.) Thus, there may be a
slight source of semantic confusion between data and data representation,

7Thus we read about “transformations that result in data becoming information and knowl¬
edge.” (Kent, 1982, p. 315.)
8For example, information is defined as “the result of processing of data, usually formalized
processing.” (Hayes, 1969, p. 218.)
9Manfred Kochen cites approvingly a statement he attributes to West Churchman: “Informa¬
tion ... is essentially raw data. Knowledge is interpreted data.” (Kochen, 1970, p. 48.) Thus
both information and knowledge are said to be data; and information is said to be raw, though in
fact it may be thoroughly processed and well-done.
10 “In more elaborate versions of the . . . model the analyst can request the synthesist to
synthesize a new CKS [contingent knowledge structure] . . . that may change the effective
information that the analyst obtains from the raw input data. . . . The contingent knowledge
structure (CKS) is a data structure which represents the understander’s knowledge of the state
of the world.” (R. J. Bobrow and Brown, 1975, pp. 115-117.)
648 FRITZ MACHLUP

but this can easily be avoided by using terms such as impulse or signal, or
even symbols.11
Why all this has caused so many writers in information science to stumble
over their attempts to define data and information in terms of one another is
probably explained by their ambition to sound more sophisticated than the
problems warrant. There is no need to establish either a hierarchy or a
temporal sequence in discussing data and information. Apart from computer
systems, the two words may be equivalents—say, “we have the informa¬
tion” or “we have the data needed for a particular inquiry”—or they may be
otherwise related—say, “our data do not furnish the information wanted in
this case.” On the other hand, in a computer system, data are what has been
fed into the memory of the system and is now available for processing.
This oversimplified presentation on “what they mean by data” has disre¬
garded some differences of possible importance. Take, for example, three
possible outputs: (1) a printout that gives us exactly what has been fed into
the memory of the computer, the same words and numbers in just the order
in which they have been stored; (2) a new arrangement of the data, after a
programmed process of sorting and re-ordering of the stored materials
(chronological or alphabetical ordering, or selecting on the basis of detailed
instructions); and (3) an output different from the stored data as a result of an
analysis made by the computer in accordance with involved instructions
specified, step by step, in a highly sophisticated program.12
Since the writers on computer and information sciences are arguing about
the question whether the results of such high-degree processing should still
be called data or should be referred to as information, 1 submit that—in this
case—the name does not make a bit of difference for any reasonable pur¬
pose. From the point of view of the computer operator, one may still say the
data are in the memory and the output is something else. From the point of
view of the user of the output, data and information are equivalent terms.
The people selling management information systems (MIS) feel better if they
call the output of their system information, that is, something of a higher
order. Some people in management-decision theory prefer to say that the
MIS output is not yet all that is needed for a decision, that more analysis

"“Sometimes a distinction is made between the mechanistic representation of the symbols,

which is called ‘data,’ and the meaning attributed to the symbols, which is called ‘informa¬
tion’.” (Teichroew, 1978, p. 658.)
12 Examples of the two higher degrees of data-processing may be helpful: (1) Input: reservations
by passengers of an airline made on particular dates for particular flights on particular days;
Output: an alphabetized list of all passengers who have reserved space on a particular flight, or a
list of the reservations a particular passenger has made for consecutive flights. (2) Input:
numerical series of business information considered relevant for analyses leading to reports
helpful to management decisions about the operations of the business firm; (MIS Information)
Output: sales report showing fluctuations in the demand for product model X over a
period of Y years.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 649

and, especially, more judgment is needed, and hence that the output sup¬
plied to the decisionmakers is still in the “lower” category of data. I repeat
that this quibbling is of no consequence.
As a relief from the dreariness of this discussion, I want to end it with a
quotation from the paper by Jesse Shera in this volume. He exclaimed:
“Data, data everywhere and not a thought to think.”

Decisions, Actions, Uncertainty

With this business of the semantic or operational links between data and
information out of the way, we may return to our survey of meanings of the
word information. We shall find that numerous proposals for restrictive
definitions have been made and, for better or worse (more often, for worse),
widely adopted.
There are those who link, by definition, information to decision-making
and action.13 This is one of the many instances in which a word that has a
wide meaning is appropriated for use with a very narrow meaning. For these
definers, information, to be information, has to have value, has to be used
for decision-making, and has to be designed to lead to action. I submit that
more than 90 per cent of all information received during a day (week, month,
year) by people in all walks of life is not related to any decisions or impend¬
ing actions. Just think of the sports sections of newspapers and broadcasts,
or of the general news reports day after day! Most recipients of such infor¬
mation do not entertain any thought that it may help them in making deci¬
sions for practical actions. Even on the highest level of thinking, information
need not be in the service of pragmatics.
Another restriction built into proposed definitions of information is that it
must reduce uncertainty on the part of those getting informed. This issue is
complicated by the fact that uncertainty means different things in the discus¬
sion of physical phenomena, processes, and measurements on the one hand,
and in the discussion of human behavior and cognitive processes, on the
other. Since I am still talking about information in the sense of telling some¬
thing to somebody, I may at this point confine myself to uncertainty as it
bears on persons receiving information and to the definitional contention
that information is not information unless it reduces uncertainty. This re¬
striction by definition has been imposed by eminent representatives of the
behavioral and social sciences, especially psychologists and economists.
[Garner, 1962; Arrow, 1979.] For the moment I need not say more than that
this definition is at variance with the common usage of the word. Countless

13 “Information is data of value in decision-making.” (Yovits and Abilock, 1974, p. 163.) The
perpetrators of this definition proceed to the statement that: “The information contained in a
decision state is related to the mean square variance of the expected values of the courses of
action.” (Ibid., p. 166.)
650 FRITZ MACHLUP

numbers of messages that convey information in the ordinary sense are

received by people without any effect on their uncertainty; and some news
items may even raise uncertainty in several respects.

Linguistics and Language Philosophy

Students of linguistics use the term information not always for what one
person tells another by some form of explicit message but rather for ex¬
plaining the relations among words in a sentence. In this formulation, infor¬
mation, in a very special sense, is something implicit in verbal expression; it
is not an action, process, or piece of knowledge, but it is what identifies the
meaning of words. This usage has the interesting result that meaning is
defined as information whereas, ordinarily, information is viewed as convey¬
ing meaning and as having a “meaning-content.” This apparently contradic¬
tory interrelation is duplicated in another instance: A message is often said
to contain information, and information is said to consist of messages. One
way to overcome such inconsistencies may be to distinguish information
from representations of information and from transmission of information.
In order to avoid confusion between the transmission of physical signals
and the meaning-content of messages or statements, Bar-Hillel and Carnap
adopted the term semantic information. As if there could be such a thing as
nonsemantic information! Yet, it was probably necessary to use that glaring
pleonasm in conferences with representatives of the statistical theory of
signal communication who had appropriated the word information to denote
“signal apart from meaning.” Bar-Hillel and Carnap, emulating the com¬
munication theorists’ efforts at measurement, proceeded to search for a unit
of semantic information; after analyzing the notion of an atomic statement
they proposed a unit of measurement, called semantic content-element.
[Bar-Hillel and Carnap, 19536, pp. 148-149.] In any case, for these philoso¬
phers of language and inductive (probabilistic) inference, information is
defined in terms of semantic content-elements of verbal statements.14
For many logicians and philosophers, information consists of statements,
and the amount of information contained in a statement is determined by the
relative number of excluded alternatives. If a statement predicates one thing
and thereby denies or excludes many alternatives, it is said to constitute
more information than a statement that excludes only a few other possible
states of the world. These considerations may make good sense in scientific
testing, where alternative outcomes are not too numerous; but in primitive
pieces of news the number of alternatives excluded by the statement may be
indefinite. It would depend on the expectations the recipient of the informa-

14Inductive inference is used by Carnap in contradistinction to deductive inference; and induc¬

tive probability is distinguished from statistical probability.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 651

tion entertained before he or she received it; the number of alternative

possibilities may well be unlimited.
The notions of uncertainty and of probability are, of course, related, and
both play important roles in various definitions or characterizations of infor¬
mation. “Two alternative notions of semantic information are reduction in
uncertainty and change in belief. . . . Information is defined in terms of
probabilities . . . .” (Jamison, 1970, pp. 28-29.) With two notions of seman¬
tic information and three views of probability—statistical, logical, and sub¬
jectivist—we are confronted with “six alternative theories of information.”
(Ibid., p. 29.) The quoted author, Dean Jamison, is inclined toward “a
definition of information that is adequate from a subjective point of view.”
(Ibid., p. 53.)

Cognitive Information Summarized

All meanings and definitions just reviewed were of information in the general
sense of something being told to somebody, where this somebody was sup¬
posed to grasp what was being told. In other words, it was meaningful
(semantic, cognitive) information; either it was the act (or process) of telling
or it was that which was being told. Some definitions, however, were restric¬
tive, refusing the designation information to some types of (what others call)
information. Thus, to list the restrictions most often proposed, information
is not recognized as information according to (any particular) proposed
definition unless (1) it is about something previously unknown to the recipi¬
ent; or (2) it is about something previously less assuredly known to the
recipient; or (3) it affects the stock or structure of the recipient’s knowledge;
or (4) it consists only of raw data, not yet interpreted; or (5) it is useful to the
recipient in some way; or (6) it is used in decision-making by the recipient; or
(7) it bears on actions contemplated, considered, or actually taken by the
recipient; or (8) it reduces the recipient’s uncertainty; or (9) it helps identify
the contextual meanings of words in sentences; or (10) it excludes some
alternatives to what is predicated in a statement; or (11) it changes some
belief held by the recipient, particularly with respect to the distribution of
probabilities in the recipient’s view.
This is surely not an exhaustive list of alternative definitions of informa¬
tion directed to a human mind; but it suffices to show that many analysts
prefer to make things true by definition and to save words to modify a more
broadly defined term. For example, instead of saying new information or
unexpected information, some choose to omit the adjective and decree that
novelty or unexpectedness is a necessary characteristic of information.
We now turn to metaphoric uses of the word. We shall, in the next
subsections, discuss uses of the term information in connection with de¬
scriptions or models of processes or phenomena pertinent to living humans
or, more generally, “living systems.”
652 FRITZ MACHLUP

information Within Living Systems: The Nervous System

Not all writers on nervous systems deem it necessary to use the term infor¬
mation in describing what goes on when sensory impulses or signals are
conveyed to the central nervous system or when motor controls are exer¬
cised by it over muscles or other organs. In scanning some of the texts and
handbooks of neurophysiology I have found that some writers can do quite
well without the term information; perhaps they have decided that it is better
to avoid the (often misleading) metaphor.15 Other writers, however, seem to
have concluded that the nervous system can be more easily explained if
communication systems such as telegraph and telephone are taken as
analogies.16 From analogy to metaphor is only a short step and users of a
metaphor may become so used to it that they adopt the term as the genuine
designation of the phenomena or processes they wish to describe or explain.
It is not always clear whether the writers on sensory physiology, psychophys¬
ics, and cognate subjects want the word information to refer to its content
or only to the process of transmission. When neurophysiologists describe
nerve-communication in terms of firings from one neuron to the next, they
can easily avoid references to “information being passed on.” However,
they have a much harder time when they want to describe tactile sensations
from mechanical stimulation being picked up by receptors in the skin and
sent on to reach a central neuron. Is the pin-prick or the tickle a stimulus, an
impulse, or perhaps a piece of information encoded by cutaneous receptors
and decoded by central neurons? To judge from the literature, it seems to be
difficult to dispense with the use of the term information.17

15 “Neurophysiology deals with functional aspects of the nervous system, with transmission of
nerve impulses, motor control, reflexes, and even perception, emotion, and mentation.” (Ochs,
1965, p. 9.) “[T]he means of propagation of the nerve signal itself is electrical in nature.” (Ibid.,
p. 11.)
16“The nerve fibre can be compared, from the point of view of its function, with a cable carrying
information.” (Zimmermann, 1978, p. 68.) This writer proceeds to discuss information trans¬
mission in the nervous system and to employ the term information not only for the transmission
process but also for measuring the amount of information in the sense used in the mathematical
theory of communication.
17 Analysts of the sensibility of the skin distinguish among several types of “receptors of cutane¬
ous sensations” and report on mechano-reception (the skin’s sensibility to touch and pressure),
thermo-reception (the skin’s sensibility to heat and cold), and nociception (the skin's sensibility
to pain). [R. F. Schmidt, 1978, p. 81.] The process of transmission from the receptor to the
receiving end of the neural string—yes, the receptors in these instances are the senders—is
similar for different sensations. The question of just what is being transmitted is difficult to
answer; one can understand the analysts’ terminological promiscuity in the exposition of vari¬
ous sensibilities, when they rather indiscriminately dally with impulses, stimuli, sensations,
inputs, signals, and information. In his discussion of vibration receptors in the skin, Robert
Schmidt states that “[t]his receptor can transmit no information about the depth of skin indenta¬
tion.” (Schmidt, 1978, p. 91. Emphasis added.) But “mechano-receptors . . . can signal the
position of a joint....” (Ibid., p. 96. Emphasis added.) However, “signals” deriving “from the
central motor systems . . . appear to send a ‘memo’ (efference copy . . .) of their signals to the
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 653

Perhaps we should not worry so much about proper word use, and accept
the fact that information has become an all-purpose weasel-word. Even a
writer as careful and lucid as Michael Arbib, in a passage in this volume,
manages to use the word information in three senses: as a process, as a
current (specific) content, and as the accumulated content of several previ¬
ous messages.18 Lest the reader get the impression that the metaphoric use
of the term information in neurophysiology began only with the emergence
and dramatic growth of the theory of information, I should mention that
many earlier instances can be found in the literature.19
Brain research is another subdiscipline in which some writers are fond of
using the term information while others seem to avoid it. We find this differ¬
ence in linguistic preference in different reports on the split brain and the
“interhemispheric communication” between the two halves of the brain.
[Sperry, 1965, 1982; Gazzaniga, 1970, 1981.] Thus, we read that after sever¬
ing the connections between the hemispheres “information learned by one
half-brain did not transfer to the other.” (Gazzaniga, 1981, p. 517.) The two
halves were no longer on speaking terms, so to speak.
Before I proceed to the next group of habitual users of the term informa¬
tion in biology, the geneticists, I may point to an interesting difference in
metaphoric use. Writers on neurophysiology are essentially concerned with
what once was discussed in terms of stimulus and response and is now
discussed in terms of signals and information; the analogy here refers to brief
ad hoc messages, telling in a few spoken words, as it were. On the other
hand, writers on genetics, with their models of a genetic code, refer to long
sets of long-term instructions, specifications with a blueprint for future per¬
formance, hence, telling in a long written scroll. Thus the metaphor “infor¬
mation” alludes to “spoken messages” within the nervous system, but to

muscles, to interact centrally with the sensory input from proprioceptors. These efference
copies give advance information about the intended muscle activity and the movements that
will result.” (Ibid., p. 99. Emphases added.) Even more metaphorically, “Pain . . . informs us
of threats to our bodies, for it is activated by noxious (tissue-damaging) stimuli.” (Ibid., p. 111.
Emphasis added.)
18In his paper in Section 2 of this volume, Arbib has two sentences where all three senses are
alluded to: “Information picked up modifies the perceiver’s anticipations of certain kinds of
information that, thus modified, direct further exploration and prepare the perceiver for more
information.” Arbib admits, however, that we do not know yet “how the organism can be
committed to an overall action by a population of neurons, none of which has global information
about which action is appropriate.” Global information, we assume, refers to accumulated
experiences (perhaps knowledge).
19 “We know that when the eye sees, all the consequent information is transmitted to the brain
by means of electrical vibrations in the channel of the optic nerve .... The impulses which
flow in the arm nerves of a typist convey to her fingers the translated information which reaches
her eye or ear, in order that the fingers may be caused to strike the proper key.” (Bush, 1945, p.
58.) I understand that more recent descriptions would not be in terms of electrical vibrations,
but the point in offering the quotations was to show the early use of the word information for
neurophysiological signals.
654 FRITZ MACHLUP

“written directions’’ within the genetic system. (I probably merit a severe

scolding for this “outrageous flirting with the potentially disastrous conse¬
quences of false analogies,” against which George Miller warns in such
delightful language in his paper on “Informavores” in this volume.) In any
case, I have found little, if any, resistance to the metaphoric use of the word
information in either neurophysiology or in molecular biology. And even
information scientists condone the practice.20

Information Within Living Systems: The Genetic System

Responsibility for introducing into literature on genetics expressions like

“information in writing” rests with the biologists. Biochemists discovered
the genetic code, a kind of written language with equivalents for letters,
words, and sentences (and even punctuation marks), formed by “nucleotide
base symbols or characters.” (Nirenberg, 1963, p. 80.)21 According to
George W. Beadle, there are five billion nucleotide bases (of four different
types) in the nuclear chromatin of a human cell. The so-called codons (each
consisting of three bases) “could convey more than 300,000,000 words of
written prose.” This corresponds to 600,000 pages of 500 words each, or to
1,000 volumes of books with 600 pages each. “That is a set of genetic
specifications for making one of us out of an egg cell . . . .” (Beadle, 1963a,
pp. 3-4.) In the formulation by James Watson, “the fertilized egg contains
all the information necessary for the growth and development of an adult
. . . animal.” (Watson, 1965, pp. 10-11.)

20For example, after talking about the documentary sense of information, Bertram C. Brookes,
a distinguished writer in information science, made this pronouncement: “Clearly, they [the
geneticists] are not using ‘information’ in its documentary sense and it is a far cry from docu¬
mentary ‘marking’ and ‘parking’ to the biochemical processes of a living cell. But while the
information analogy is helpful to microbiologists I see no reason why it should not be used.”
(Brookes, 1974, p. 146.)
21 The genetic code is made up by four nucleotide bases (symbols, characters, letters), conven¬
tionally denoted by the letters A, T, G, and C. Pairs (or doublets) of letters would allow coding
of only 42, or 16, “words,” which would be insufficient to specify 20 amino acids. Sequences of
three letters (triplets) allow coding of 43, or 64 words, or codons, and thus permit a good deal of
redundancy. The “sequence of nucleotides in DNA carries the genetic information that orders
(codes) the sequence of amino acids in proteins.” (Watson, 1965, p. 99.) “The molecule’s
message written in nucleic acid code” may represent structural genes, controlling “how
molecules of specific amino acids are combined to form a particular protein, programming the
pattern of spatial organization of that protein.” Other messages written in the same code
represent regulatory genes, programming “the pattern of temporal organization, determining
when syntheses of protein molecules shall occur by controlling rates of synthetic processes.”
(J. G. Miller, 1978, p. 221. Emphases in the original.)
The allusion, in the text above, to punctuation marks refers to termination or the end of a
chain of triplets. “Termination is not a random process, but is highly controlled by special ‘full-
stop’ or termination codons. Recent studies have shown that certain triplets or codons (called
‘nonsense codons’) in the mRNA [messenger RNA] chain automatically bring about termina¬
tion of the peptide chain at that point. The nonsense codons are UAA, UAG, and UGA. When
the ribosome reaches a nonsense codon, the bond between the final amino acid and the tRNA
[transfer RNA] molecule to which it is attached is hydrolyzed.” (Baker and Allen, 1982, p. 561.)
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 655

It may be worthwhile seeing how researchers in molecular biology, cy¬

togenetics, and phylogenetics (evolutionary biology) have come to deal with
different kinds and forms of information in several different contexts. Be¬
sides the information “stored” in the genes, or “imprinted” in the DNA by
sequences of nucleotides, there is the “transcription,” or “copying,” of
DNA information into an RNA “transcript,” and then the “translation” of nu¬
cleic-acid information into protein information. There is also translation of
the information encoded in the message in the cytoplasm, a “backward flow
of genetic information” from RNA to DNA, resulting in “DNA copies of
messenger RNAs, which have become inserted in the genome.” [The quoted
phrases are from “Research News” published in Science, vol. 212 (1981),
p. 313, and vol. 216 (1982), p. 969.] I wonder whether, in these and similar
formulations, information as a content of meaning in terms of action or effect
is sufficiently distinguished from physical representations of information. In
addition to the phenomena and processes involving the written genetic code,
there are other kinds of information in the geneticists’ scenario, for example,
a hypothesis of a “cross-talk” between adjacent chromosomes, and hypo¬
theses of “extrinsic generic instructions,” or “information” from the envi¬
ronment, that may account for phylogenetic patterns in evolution. [Science,
vol. 214 (1981), p. 1334, and vol. 217 (1982), p. 1239.]

Information Within Social Systems: Society, Polity, Economy

If we were to follow the sequential order prescribed by James G. Miller in

his book on Living Systems, we would have to move up from the cell and the
organism to the group and the organization before we came to society. [J. G.
Miller, 1978.] For our purposes—reviewing what is meant by the term infor¬
mation—Miller’s sequence would not be useful, because it plays down or
conceals the one issue that matters most: The difference between informa¬
tion in a metaphoric sense where no minds and no cognitive processes are
involved, and information in the original and traditional sense where mean¬
ingful perceptions and thoughts reach a mind that receives and interprets
them. Not the organism but the individual plays this role in an appropriate
model of received information, no matter whether the individual is regarded
as standing alone or as thinking, acting, and reacting as a member of a group,
an organization, or a society.22 These points were made and elaborated by
Talcott Parsons in a review article on Miller’s book. [Parsons, 1979.] As a
matter of fact, these issues have been clear to all social scientists who
understood the principles of methodological individualism—and therefore

22For James Miller, “A society is a large, living, concrete system with organizations and lower
levels of living systems as subsystems and components.” Miller explicitly rejects Toynbee’s
view of looking at society as “a product of the relations between individuals,” and likewise the
definition by Parsons, Shils, and others, of a social system as “interactive relationships of a
plurality of individual actors.” (J. G. Miller, 1978, p. 747.) Here is one of Miller’s explicit
statements on information (or perhaps a proposed definition): “Information is the patterning of
matter-energy in systems.” (Ibid., p. 1030.)
656 FRITZ MACHLUP

did not confuse it with political individualism—and who rejected the myths
of romantic holism—not to be confused with the holistic aspect of general
system theory. Some of the papers in Sections 8 and 9 of this volume deal
with the problems in question and generate heat and light in the process.
(The reader may enjoy both these outputs.)
What the word information means to individual members of groups, or¬
ganizations, and society, and also to analysts of these social systems, is not
different from what the dictionary definitions say. On all these levels, infor¬
mation has or is a meaningful content. In the analysis of information flows,
of the possible relations between information and decision, of possible ef¬
fects of information upon mass action, and of several other problems, spe¬
cialists in the study of organizations may emphasize other aspects than do
the students of economic society or political society; some writers may find
it expedient to make different specifications with regard to degrees of uncer¬
tainty or changes in beliefs, or with regard to revisions of expectations or
even of goals; still, the fundamental notion of information is the same in all
social sciences.
There have been writers who were fascinated by notions of social infor¬
mation and social knowledge, in the senses of new information acquired or
total knowledge possessed by society—where society was supposed to be
something quite different from the individuals that compose it.23 The basic
ideas of system-thinking seem to foster a belief in the reality of a whole apart
from its components and their interactive relations. If information is some¬
thing that reaches a mind, or several or many minds, it ought to be clear that
the whole does not have a mind of its own and can neither receive nor
process the information that has reached its members. To be sure, some
special decisionmakers, say, members of legislatures, administrations,
public agencies, and so forth, may have the right and the power to act upon
received information (collectively) issuing laws, decrees, and regulatory or¬
ders controlling, constraining, inducing, or prohibiting certain kinds of ac-

23Norbert Wiener has in his Cybernetics a chapter on “Information, Language, and Society,”
in which he tries to differentiate the “intelligence of society” or “community” from the intelli¬
gence of its members. [Wiener, 1948 and 1961, p. 162.] He states that there is “no necessary
relation in either direction between the amount of racial, or tribal, or community information
and the information available to the individual.” (Ibid., p. 158.) Not all references, however, to
social and individual knowledge are naive or fallacious. On sensible distinctions between social
and individual knowledge, proposed by Bertrand Russell, Alfred Schutz, Thomas Luckmann,
and others, see my book on Knowledge and Knowledge Production. [Machlup, 1980, pp. 161 —
162, 167-173.] However, the contributions of system theorists to the flows of information into
the system and among its components are, by and large, rather confusing. Thus, for example,
James Miller speaks of a “special form [of information flow], the flow of money, which is one
sort of information.” (J. G. Miller, 1978, p. 1027.) Miller may possibly be thinking
of prices emerging in commodity markets and interpreted as giving signals or information to
producers and consumers. But this has nothing to do with the circulation of money; and it surely
does not support the notion that flows of money are flows of information. It is not a useful
analogy—just a fallacy.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 657

tions. Still, both the controllers and the controlled are individuals, even if
the controllers may claim to be acting on behalf of society. Moreover, mass
actions by individuals not entitled or empowered to legislate or to govern
may well exercise powerful influences on governors and controllers. The
titles and constitutional functions assigned to particular components of a
social system do not necessarily determine their effectiveness; the control¬
ling forces may be widely diffused, independent of any official organization
chart.

Information Within Man-Made Systems: Machines

Interactions among parts of man-made apparatuses are sometimes referred

to as intercommunication or information flows. Let us recall that Norbert
Wiener gave to his book Cybernetics the subtitle “Control and Communica¬
tion in the Animal and the Machine.’’ He proposed the term feedback for the
“chain of the transmission and return of information” and showed that
“there are . . . feedback chains in which no human element intervenes.” His
first examples for a “purely mechanical feedback system” were the thermo¬
stat and the steam engine. “The information fed back to the control center
tends to oppose the departure of the controlled from the controlling quan¬
tity.” (Wiener, 1948 and 1961, pp. 96-97.)
The similarity of the working of homeostatic feedbacks in physiological
and in man-made systems makes the use of the term information in both
living and nonliving systems acceptable as long as one does not forget that
the term is used as a metaphor. Real information can come only from an
informant. Information without an informant—without a person who tells
something—is information in an only metaphoric sense.24
I wonder why so many scientists have found it permissible to overlook,
disregard, or minimize the differences between information and those types
of causation that have nothing to do with signs, meanings, and mental pro¬
cesses. Among possible reasons may be the use of algebraic expressions in
lieu of literary statements. If Y = f(X), writers and readers of such a func¬
tion are led to shut out any questions of whether A is a mechanical, electri¬
cal, anorganic-chemical, organic-chemical effector or a sign conveying to
some mind or minds a meaningful message that may influence the recipients
in their considerations, decisions, and actions. Colin Cherry regarded the
difference in the responses of man and of machine as essential: quasi-

24One of the few specialists who was sufficiently perceptive to see the difference was Colin
Cherry: “All communication proceeds by means of signs, with which one organism affects the
behavior [or state] of another. ... In certain cases it is meaningful also to speak of communica¬
tion between one machine and another as, for example, the control signals which pass between
a guided missile and a ground radar.” Cherry proceeds to consider the question how to “distin¬
guish between communication proper, by the use of spoken language or similar empirical signs,
and other forms of causation.” (Cherry, 1957, p. 219.)
658 FRITZ MACHLUP

voluntary versus semiautomatic. “If I push a man into the lake, he inevit¬
ably goes in; if I tell him to jump in, he may do one of a thousand things.”
(Cherry, 1957, p. 220.) To be sure, the question whether X was a push or a
sign may be of little interest to a scientist who wants to explain the state Y of
the man, dead or alive, but surely wet. However, to a scientist not satisfied
with algebra it will make a difference whether X was a physical force or
rather a sequence of signs in gestures or words. Indeed, some scientists may
wish to inquire what kind of meaningful information has induced our man to
jump in the lake: a command, an advice, a threat of worse consequences, a
bet, an offer of a bribe, a report of a sad event? If any of these messages and
any acts of physical force can be indiscriminately called information, this
term has lost much of its usefulness.
Despite this grave charge, I concede that, with the degree of discernment
that one should expect from any academically trained person, the probabil¬
ity of confusion caused by metaphoric uses of the term is not intolerably
great.

Information as an Alias for Signal Transmission

We now come to the last group of users of the term information; it is a rather
special case: They use information and amount of information in a sense that
has so little to do with any traditional or metaphoric meanings of the word
that one can only wonder why the scientific community has allowed it to
continue. In the Prologue to this volume, especially in the subsection on
“Information Theory,” it was pointed out why the misnaming has been
infelicitous, misleading, and disserviceable, and also why frequent attempts
to correct this state of terminological affairs have been unsuccessful. At the
present juncture I merely wish to state just what the information theorists (or
rather the expositors of the mathematical theory of communication or signal
transmission) meant when they said information.25
Information in the sense used in (narrow) information theory refers
neither to the process of telling something to somebody nor to the something
that is being told; indeed, it has nothing to do with meaning, not even in a
metaphoric sense. Instead, it is the statistical probability of a sign or signal
being selected from a given set of signs or signals. This probability is of
significance for the design of communication systems, for determining the
optimal size of a communication channel, for measuring its capacity for
signal transmission, for appraising the efficiency of a system (existing or
proposed), and for other purposes. Expressions such as “information in the
technical sense,” “statistical sense,” “mathematical sense,” or “engineer¬
ing sense,” have become customary in order to distinguish this concept from

25 As Donald MacKay writes in his paper in Section 7 of this volume, “[Shannon’s] theory did
not define the concept of information at all.”
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 659

information in a meaningful or semantic sense. Kenneth Boulding speaks of

“Bell Telephone or B.T. information” and, like hundreds before him, dis¬
misses the notion as irrelevant for the social and behavioral sciences. (See
his paper in this volume.)
It is difficult, perhaps impossible, to formulate a clear definition of so-
called information in the statistical sense. It surely is not information nor is it
signals transmitted. As a matter of fact, few writers have attempted a
definition; what they set out to define was a “measure of information,” or a
“rate of information,” or a “statistical relation between signs.” (Cherry,
1957, p. 226.) In Cherry’s words, “Shannon’s measure of the selective infor¬
mation rate of signals in terms of their statistical rarity” is at best one of
several things one may want to measure. (Ibid., p. 229.) The adjective selec¬
tive modifying the noun information can remind us of the close relation to
probability theory but it does not inform us of the meaning of the thus
modified noun. Moreover the terms measure, rate, and relation must refer to
some specified object or objects, and it is not clear just what the object,
misnamed information, is meant to be. We still have to be told what is being
measured.
One might, paraphrasing several information theorists, try to define infor¬
mation in the technical sense as “the coding and transmission of signals
contained in a given repertoire [or set].” Although this would make sense, it
is not what Shannon’s formula expresses. The reference to probability
theory, of which this kind of information theory is an application, is too
indirect; the word “selection” does better in this respect. The old definition
by Ralph Hartley, saying in effect that information is the successive selec¬
tion of signs or words from a given list, is among the clearest; he rejected all
meaning as a “mere subjective factor,” and insisted that “it is the signs that
we transmit, or physical signals; we do not transmit their meaning.” (Hart¬
ley, 1928, quoted from Cherry, 1957, p. 43.) More descriptive expressions
were suggested in the introductory essay of this volume: activating impulses
are transmitted through the communication channel, designed to make the
receiver pick out the intended signs from a given ensemble (repertoire,
alphabet). Activation rate is measurable and easily correlated with the no¬
tion of channel capacity. Activation and amount of activation are meant
when communication theorists speak of information and amount of infor¬
mation.
That we should stress the (measurable) physical signals rather than their
meanings is reasonable for communication engineers designing telegraph,
telephone, and other transmission channels. At the level of personal oral
communication, we would, to be consistent with their terminology, have to
define information as vibrations of the air, measurable by amplitudes and
frequencies; and at the level of hand-written communication, we would have
to define information in terms of ink-marks or pencil-strokes. This would be
possible, of course, but it would be a wasteful use of the word information; it
would not be what people meant by it.
660 FRITZ MACHLUP

Information as Negative Entropy

In discussing the unfortunate use of information as an alias for signal trans¬

mission or activation, I have not mentioned the term entropy, often em¬
ployed to illuminate (or obscure) information theory. The relation between
the terms merits separate discussion, but unfortunately space limitations
preclude such discussion here.
To be sure, there are those who consider entropy and information as
semantically related terms. Thus the physicist Gilbert N. Lewis stated that
“gain in entropy always means loss of information, and nothing more.”
(Lewis, 1930, p. 573.) And Warren Weaver, in his interpretation of Shan¬
non’s theory, said first that “information is measured by entropy”—which
would presume that they are different things—then went on to say that
“entropy is related to ‘missing information’,” and finally that “entropy and
information” were equivalent concepts. [Shannon and Weaver, 1949.] Still,
although one can find instances in which information and entropy are made
equivalent terms, the relationship is more interesting from the point of view
of methodology.

A Compendium of Meanings

At the beginning of this discourse on “What They Mean by Information” we

looked into dictionary definitions; they all referred to telling something or to
the something that is being told. Information is addressed to human minds
and is received by human minds, though the recipient need not always be
chosen by the informant or transmitter. Pieces of information carry mean¬
ings and are interpreted by cognitive processes, but not necessarily by all
intermediaries. A thoughtful analysis of this basic sense of information
would probably make the informants’ intention to inform a criterion of infor¬
mation.
Starting from this fundamental concept we have noted several restrictive
definitions of information in the basic sense: Restrictions were imposed by
users of the term for purposes of their theoretical tasks. The requirement of
truth or correctness should exclude false or incorrect messages; the require¬
ment of value or usefulness should exclude messages not helpful in decisions
and actions; the requirement of novelty should exclude repeated or redun¬
dant messages; the requirement of surprise should exclude messages that the
recipient expected; the requirement of uncertainty-reduction should exclude
messages that leave the recipient’s state of uncertainty unchanged or in¬
creased; and so forth. No exhaustive enumeration of persuasive or dictato¬
rial restrictions is here intended. Specialists in several disciplines—
psychology, economics, decision theory, linguistics, and others—have
imposed restrictions when it suited their purposes and have thereby made it
harder for nonspecialists to understand what was meant.
Information is used in some metaphoric senses if, though some character¬
istics of the basic concept are missing, other essentials are present and
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 661

suggest transmission of meaningful messages from transmitters to recipi¬

ents. The major metaphoric uses occur in the description of the operation of
living systems. Writers on the nervous system, on brain research, and on the
genetic system have been the chief employers of the metaphor. The charac¬
teristics of information in the basic sense that are missing from information
in any of the metaphoric senses are, chiefly, the involvement of a mind and
of cognitive processes. These characteristics can help the social scientist to
distinguish between genuine and metaphoric information when he analyzes
the operation of social systems. If his analysis includes individuals, transmit¬
ting and receiving signals and messages and making up their minds before
they act or refrain from acting, he is concerned with information in the basic
sense. However, if he resorts to systems of nondeliberating organisms or to
a society as a whole, which cannot have a mind of its own, he is at best
indulging in metaphoric uses of information, more likely with alleged forces
that are not information in any sense. What makes the metaphoric uses of
the word information acceptable is that, although the processes involved are
physical (electrical, chemical) and biological (neurological, molecular)
rather than cognitive, one can think of nerve cells “telling” something and
of genes “telling” various other cells of the body what to do and when to do
it.
This fiction can hardly be sustained in the case of man-made apparatuses
and machines. To use the word information for actions and interactions,
feedforwards and feedbacks, of the different parts of a nonliving (though still
goal-directed) system, comes close to a caricature. Too many cyberneticists
have accepted the antropomorphic expression that machine-part A “tells”
something to machine-part B. The word “information” in this sense stands
for “effectuation.” No intention, no mind, no cognitive action is involved.
The use of the word information where only observation and analysis are
involved, is just a confusion. Those who believe that observation of physical
reality or “consultation” of data “tell” us anything have misunderstood the
basic lessons of methodology.
Finally, to speak of information when information theorists explain their
system is a sad misuse of language; they explicitly abstract from a meaning-
content of the signals the transmission of which they describe. Their system
does not care about telling anything, directing or advising anybody, arousing
anybody’s interest, or inducing any decisions or emotions. Appropriate
words to use in the context would be signal transmission, actuation, or
activating impulses. The use of the word information in this sense has led to
unending confusion and should no longer be condoned.

WHAT THEY MEAN BY SCIENCE

This second piece about semantic quirks should not take nearly as much
space as the previous one. Not that the word science has been used in fewer
662 FRITZ MACHLUP

diverse meanings than the word information; but the differences in the
meanings of science can be more easily explained.

Dictionary Definitions

The Latin noun scientia, derived from the verb scio, scire (to know), means
knowledge (both knowing-what and knowing-how), awareness, cognition,
insight, skill, and science in the widest sense. Some writers, Caesar among
them, spoke of scientia atque usus militum, which alludes to a possible
differentiation between theoretical knowledge and practical experience in
military matters, but a broad sense of “knowledge” was the more common
meaning of scientia.
The Oxford English Dictionary gives the following meanings: (1) The state
of fact of knowing; knowledge or cognizance. (2) Knowledge acquired by
study, b. Trained skill. (3) A particular branch of knowledge or study; a
recognized department of learning. (4) A branch of study which is concerned
either with a connected body of demonstrated truths or with observed facts
systematically classified . . . under general laws. (5) The kind of knowledge
or intellectual activity .... What is taught in the Schools, or may be learned
by study, b. In modern use often ‘Natural and Physical Science’, c. The
portions of philosophy, logic, etc., included in the course of study in the
Oxford School of Literae Humaniores.
Note the division of Entry (5) into (a) any discipline taught, (b) natural and
physical sciences, and (c) humanistic studies, where (b) is given as the
modern use and (c) as having become obsolete in the twentieth century. This
list reflects a history of semantic change but does not do justice to it: It is a
history of arbitrary restrictions imposed by successive schools of thought,
each trying to exclude the teachings of its opponents as unscientific. I have
elsewhere presented a brief sketch of this semantic development, but I think
that a still more concise statement here will be helpful. [Machlup, 1980, pp.
62-70.]
The sequence of exclusions has some similarities to a game of musical
chairs. For the playful the game may be good fun; in actual fact, however, it
has caused, and continues to cause, a waste of scientists’ time as each tries
to secure for himself a safe chair—instead of doing more important work.

Excluding Empirical Knowledge

The first exclusion was of empirical knowledge or any kind of knowledge

that could be doubted or had to be tested by experience. Science was sup¬
posed to be perfect knowledge, derived from unquestionable principles; it
was abstract knowledge demonstrable with mathematical certainty. This
concept of science goes back to Aristotle, but was dominant until the seven¬
teenth century, remaining strong for the next two centuries, and was still
preferred by some exceptional philosophers in the twentieth century.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 663

Limiting references to a minimum, I cite only three: The French philoso¬

pher Descartes held in 1644 that “any knowledge that can be questioned
ought not to be called ‘science’.” The British astronomer Sir John Herschel,
in 1831, formulated the definition of science as abstract logically demonstra¬
ble knowledge. And the Italian Benedetto Croce is one of those who held out
for this classical position when he wrote, in 1913: “But we cannot rest
satisfied with asserting the right of Logic to be recognized as a science: we
must make the further demand that Philosophy alone—and not empirical
sciences—be admitted as science in the strict sense of the word.” (Croce,
1913, p. 201.)
In this statement, however, philosophy was given its modern meaning, for
until the nineteenth century philosophy was the name for what today are the
empirical sciences: “Astronomers and physicists did not feel degraded by
being called natural philosophers rather than scientists; nor did historians,
political economists and other students of human society take offense at
being called moral philosophers.” (Machlup, 1980, p. 64.)

Excluding Nonsystematic Knowledge

The next stage in the semantic development was really one of liberalization,
in that the exclusion of empirical knowledge was lifted to the extent that
such knowledge was systematized according to stated principles. The Ger¬
man philosopher Immanuel Kant was among the leaders in making science
less exclusive; he still reserved the designation “science proper” for bodies
of knowledge formulated in propositions of apodictic certainty, that is,
purely analytic propositions. And metaphysics was still “the science of the
first principles of human cognition.” (Kant, 1848, p. 581.) But what had
previously been natural philosophy and moral philosophy became two cate¬
gories of empirical sciences, along with the formal sciences.

Excluding All But the Natural Sciences

The next turn in the sequence of exclusions took place in mid-nineteenth

century England and affected the semantic development in all English-
speaking countries. The natural scientists, only recently admitted to the
society of scientists, “threw out” those who had previously excluded
them—the rationalists and all who relied on speculative arguments. First the
natural scientists meant to expel only those professing metaphysics, but
soon they went further and excluded all but the laboratory scientists, the
experimentalists. Thus, the designation science was to be reserved for the
natural sciences to the exclusion of social sciences, history, philosophy,
logic, ethics, literature—in short, all the scholars who do most of their work
at a desk, in a library.
One of the consequences of this radical change of word-meaning was a
language barrier between English-speaking countries and the rest of the
664 FRITZ MACHLUP

world. None of the equivalents of the word science in languages other than
English has that particular restrictive meaning. The academies of sciences
everywhere but in the English-speaking world retained their divisions for
historical sciences, juridical sciences, philosophical sciences, and so forth;
how can one translate these words into English without contradiction in
terms? (The phrase “arts and sciences” is often called in for the purpose.)
And how should “scientists” in the restrictive sense be translated into
French, German, Italian, Japanese, or Russian? (It can be done, of course,
by adding some additional words to modify the noun.) Far more serious than
the task of translation have been the countless confusions when partners in
learned discussion erroneously confounded the exclusive designation with
the wider meanings of science.26

Excluding Laymen's and Practitioners' Knowledge

On the continent of Europe and in other non-English-speaking countries the

restriction of the meaning of “science” to natural sciences or experimental
sciences has never been accepted. However, since definitions are inherently
restrictive, they must exclude something: If the genus proximum of science
is knowledge, the differentiae specificae are (1) the systematic organization
of that knowledge on the basis of stated principles, (2) the acquisition of that
knowledge in arduous study and honest research, and (3) the use of that
knowledge for the sake of knowing, disseminating, or adding to it in further
research. The last two specific criteria are usually stated negatively in the
form of two exclusions: of laymen’s knowledge (which is not acquired in
arduous study) and of practitioners’ knowledge (which is used for a practical
purpose other than in teaching, writing, and researching).
No other exclusions besides those of unsystematic knowledge, laymen’s
knowledge, and practitioners’ knowledge are promulgated by this cos¬
mopolitan sense of science; no restrictions apply with regard to subject
matter, research technique, modus operandi, or form of presentation. Refer¬
ences to this meaning of science can be culled from encyclopedias and
dictionaries in virtually all languages except English. [Machlup, 1980, pp.
67-70.]

26The title of Max Weber’s essay “Wissenschaft als Beruf” was rendered in English as “Sci¬
ence as a Vocation.” Since Weber meant by science all disciplines or branches of learning, the
translator should have rendered it as “academic disciplines” or “science and other scholar¬
ship,” or some other awkward phrase. Titles of editions of the scientific papers of eminent
lawyers, art historians, or theologians in Europe could not be translated into English before
checking whether the papers were in the authors’ own fields or perhaps on subjects of natural
science. After all, in Germany or France a lawyer’s paper on a legal issue prepared for publica¬
tion in a law journal is a scientific paper; if it were a brief for the court’s consideration, it would
be of direct practical usefulness and hence “nonscientific.” Erudition and purpose—not the
subject and not the method—determine what is science in cosmopolitan usage.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 665

Excluding Philosophy

A second round of the elimination contest between science and philosophy

began at the end of the nineteenth century but, almost hilariously, after a
complete semantic turn-around. The earlier exclusion of philosophy had
been the exclusion of empirical disciplines, such as physics and biology,
anthropology, and politics, then known as natural and moral philosophy,
respectively. Now these philosophies had become the sciences, and the
universal systems of analytical propositions, based on axioms and logical
implications, had become philosophy. As if in a comedy of errors in which
characters exchange their clothes and their names, we find scientists throw¬
ing out philosophers—only that in the first act the evicted ones were the
searchers for truth by experience, in the second act they were searchers for
truth by ratiocination.
Not that everybody agreed with the expulsion of philosophy. From the
supreme position in science—philosophia prima in Francis Bacon’s
classification of the sciences, and the “science of the sciences” in the
schemes of the academies—philosophy was to be outcast as a nonscience!
Not unexpectedly, the philosophers protested most loudly. “We cannot
refuse the name science to logic ... ”, said Morris Cohen. (Cohen, 1931, p.
89.) Benedetto Croce’s strong claim on behalf of philosophy was cited
above. Rudolf Carnap reserved to the “formal sciences” and to “knowledge
based on common sense” prominent places in the Logical Foundations for
the Unity of Science. (Carnap, 1939 and 1969, p. 145.)
Anglo-American scientists, on the other hand, seem determined not to
tolerate the company of philosophers, and even Anglo-American philoso¬
phers seem to justify segregation, and even discrimination. Bertrand Russell
is often quoted among those who draw a firm dividing line between science
and philosophy: “[PJhilosophy consists of speculations about matters where
exact knowledge is not yet possible . . . and for that reason questions are
perpetually passing over from philosophy into science as science ad¬
vances.” Moreover, philosophy keeps us “aware of how much that seemed
like knowledge isn’t knowledge.” (Russell, 1960, pp. 11-12.)
Most contributors to this volume appear to share the view that philosophy
should be excluded from the concept of science. Hassan Mortazavian shows
this tenet even in the title of his paper, “System Theory versus System
Philosophy”; and in its text he elaborates by saying that “philosophy of
systems—a branch of philosophy dealing with problems centered around the
notion of system. . . must be distinguished from system theory, which is the
science of systems.”
The separation of philosophy from science has been deeply deplored by
some of our great scientists. Thus, Louis de Broglie, the physicist, wrote:
“In the nineteenth century there came into being a separation between
scientists and philosophers. The scientists looked with a certain suspicion
666 FRITZ MACHLUP

upon the philosophical speculations .... The philosophers, in turn, were no

longer interested in the special sciences because their results seemed too
narrow. This separation, however, has been harmful to both philosophers
and scientists.” (de Broglie, 1941, pp. 26-27.) Philipp Frank, although he
could not, as a teacher at Harvard, insist that philosophy was a science, did
write about ‘‘science as a fragment of philosophy, [and] how ‘science’ can
become philosophy.” (Frank, 1957, pp. 28, 32.) In a section on speculative
science he vigorously denied that philosophy is ‘‘concerned with hypotheses
of a more speculative nature than those found in science . . . , since all
hypotheses are speculative. No distinction can be made between scientific
and speculative hypotheses.” (Ibid., p. 37.)

Excluding Nontested and Nontestable Knowledge

Some may think that the exclusion of philosophy is tantamount to the exclu¬
sion of nontestable knowledge. This is not so. The concept of science that
excludes philosophy may still include many hypotheses and propositions of
a hypothetico-deductive character as well as models containing ‘‘inter¬
mediate” variables that are nonmeasurable and nonobservable. A concept of
science that excludes nontested and nontestable knowledge-claims is, there¬
fore, more restrictive than the one that excludes philosophy.

[Editor’s Note (U.M.): This subsection was unfinished at the time of Professor Machlup’s
death. Also missing from the manuscript were the following subsections listed in his outline for
inclusion at this point: Excluding Nonempirical Knowledge; Excluding Nontheoretical, Non-
nomological Knowledge; Excluding Knowledge of Man and Society; Excluding Ideographic
and Historical Knowledge; Excluding Literary Scholarship in the Humanities; Excluding
Evaluative and Prescriptive Knowledge; and Excluding Nonquantitative Knowledge. The
manuscript continued with the subsection, Excluding Art.]

Excluding Art

Occasional remarks by teachers and writers about some discourse or en¬

deavor being ‘‘not science but merely art” or ‘‘as yet only art but gradually
developing into science” cause me to comment briefly on this issue.27 I have
done so on earlier occasions and can therefore be very brief. [Machlup,
1969, pp. 106-108; 1978, pp. 432-434; 1980, pp. 90-92.]
Most attempts to separate science from art (chiefly practical arts, not fine
arts) suffer from insufficient attention to nuances both in science and in art.
Only Jeremy Bentham’s distinction can avoid problems of making arbitrary
divisions in the spectrum from “pure thought” to “actual performance,”
since Bentham decided to contrast thinking and doing, or “thought” and

27One example: “If computer programming is to become an important part of computer re¬
search and development, a transition of programming from an art to a disciplined science must
be effected.” (Bauer, Juncosa, and Perlis, 1959, pp. 121-122.)
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 667

“action.” (Bentham, 1816, p. 9.) Even with these categories there are
difficulties, in that the scientist too does something—experimenting, writ¬
ing, teaching—and the performer ordinarily thinks when using the special
knowledge needed for the attainment of a task.
Another distinction was proposed by John Stuart Mill: “Science is a
collection of truths; art, a body of rules, or directions for conduct.” (John
Stuart Mill, 1836 and 1844, p. 124.) This seems clear until we realize that
rules for conduct could easily be translated into general propositions of
cause and effect, and vice versa. As we look for criteria to distinguish
science and art, we will find it necessary to subdivide each. The distinction
between basic and applied science is unavoidable; as to art, we ought to
realize that (1) general teaching of rules for performance, (2) precepts for
concrete situations, (3) directives for immediate action, and (4) execution of
such directives are all different things. [Machlup, 1980, p. 92.] In this divi¬
sion of the spectrum, the line that separates applied science from art as a
body of general rules for possible performance is by no means hard and fast.
This is not necessarily a disadvantage; sometimes it is better not to impose
arbitrary demarcation lines where differentiation is a matter of judgment.
In his Turing Award lecture in 1974, Donald Knuth addressed himself to
the question of “Computer Programming as an Art.” He clearly disas¬
sociated himself from the opinion “that there is something undesirable about
an area of human activity that is classified as an ‘art’; it has to be a science
before it has any real stature.” (Knuth, 1974a, p. 667.) One of his theses was
that programming was also “an art form, in an aesthetic sense.” Program¬
ming “can be like composing poetry or music,” it can give aesthetic plea¬
sure to the programmer as well as to the user, it can be “a beautiful thing.”
(Ibid., pp. 670, 672.) Of course, programming is also an art in the technical
sense, in that it takes know-how and skill to write a good program for a task
to be executed by the computer. On the other hand, the required knowing-
how is based on a goodly measure of scientific knowing-that, and the pro¬
gram is designed to assist in finding other pieces of knowing-that. Even the
purest scientist needs to have various skills; to comprehend, to reason, to
see previously overlooked problems, to ask the right questions, to devise
relevant experiments, to interpret the findings intelligently, and so forth. In
other words, the scientific researcher needs a lot of know-how—and this is
art. Undoubtedly, the mixture of know-how, know-that, and know-what-to-
find-out is different for different tasks and different researchers; there may
be tasks for which routine skills are the chief ingredients. On the whole,
assessments of the share of art and the share of science in research are
questionable. To some extent, such assessments enter into job descriptions
and the choice of staff titles, for example, in the distinction between profes¬
sional and technical personnel; but I doubt that this is what people have in
mind when they proclaim that some activity is only art, not science.
Moreover, I cannot help, when I hear such a pronouncement, asking the
rude question “So what?” If programming, modeling, experimenting, fore-
668 FRITZ MACHLUP

casting, and similar activities are art, not science, what bearing does this
have on the activity so demoted? Perhaps the assessor has some dictum in
mind like “science is true” whereas art is not aiming at truth. The answer,
then, would be that the mentioned activities are engaged in the service of
finding, or approaching, the truth, even if no scientist can ever be sure to
have captured it.

Excluding Technology

Among dictionary meanings of the word technology are “applied science”

and “industrial science.” More formally, technology may be defined as the
science of techniques and their uses, (originally) in the transformation of
materials into products but (lately) also in any kind of operation, even purely
mental activity. Thus, technology is a practical science in the sense that its
theoretical and descriptive propositions can be applied in practical pur¬
suits.28 The idea that technology is formulated in the imperative mood, like a
cookbook, is obsolete; most imperative, normative, or prescriptive state¬
ments can be reformulated as declarative sentences.29
It follows that science and technology are not opposites; the relation is
that of genus and species. The misconception that they are opposites may be
due to historical reasons. Institutes of technology in continental Europe
have long been separated from traditional universities. This separation was
not because of a distinction between scientist and professional; indeed, the
ancient professional schools—law, medicine, and theology—training the
learned professions, were regarded as the superior faculties of the univer¬
sities. The different status of technology was perhaps due to the fact that
students of technology did not need, and rarely had, knowledge of Latin and
no one so “uneducated” could be admitted to a university. Even the second¬
ary schools preparing students for higher education were separated—in lo¬
cation as well as ideology: A classical curriculum was required for those
preparing for university, a modern-realistic curriculum for those preparing
for higher schools of technology. By now much of this is past history but the
ideological separation seems to survive.

28The concept of “practical science” goes back to Aristotle and held an important place in the
system of four “orders of science” distinguished by Thomas Aquinas—natural philosophy,
rational philosophy, moral philosophy, and practical sciences. Aquinas used “mechanical arts”
as a synonym for “practical sciences.” As examples he listed medicine, shipbuilding, and
strategy. [Machlup, 1982, pp. 26-27.]
29This statement seems to contradict the contention of Pearson and Slamecka in their paper to
the effect that “one of the distinguishing features of science and technology concerns the
essential role played in science by declarative sentences and in technology by imperative
sentences.” This point holds equally for a handbook of medicine: Handbooks and reference
books in medicine and in technology (computer technology, fibre technology, mining technol¬
ogy, and so forth) are not actually written in the imperative mood. Only their pragmatic applica¬
tions are closer to those in the foundation sciences, and to emphasize this, some philosophers
have resorted to the cookbook analogy.
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 669

Some writers on the relation between science and technology like to

stress the place of science as a prerequisite of technology, referring to the
fact that technologists and engineers, to do their job well, must know enough
physics, chemistry, and a few other disciplines to use or apply these bodies
of knowledge. This would make science a source of supply for the produc¬
tion of technology—another rather naive conception, which disregards the
important intellectual feedback from technology to various other disciplines
including the basic natural sciences. The essential point is that the universe
of learning includes many disciplines, all coherent and systematically or¬
dered arrangements of knowledge—and hence sciences—and many over¬
lapping or interdependent. Included in this class is the science of tech¬
nology.30
Is there a difference between technology and engineering? Again the con¬
trast between science as knowledge and art as performance is relied upon as
a distinguishing characteristic. Engineering is the activity of designing and
constructing artifacts, often works of public utility; and technology is the
general knowledge required for that activity. Technology, according to the
dictionary, is the “discourse or treatise” dealing with the known pos¬
sibilities of transforming available inputs into desired outputs. Thus, the
distinction between technology and engineering makes it even clearer that
technology is a branch of science—the most practical branch perhaps—
developed with ultimate “action-orientation” or “mission-orientation.”31

WHAT THEY MEAN BY COMPUTING

Some people are very proud of the continuing changes of their language as
old words acquire new meanings. They take this as a sign of progress: Theirs
is a living language, which flexibly adapts itself to changes in the environ¬
ment, particularly in technology. They are not disturbed by the fact that
some of the new meanings of old words are largely due to ignorance of the
original meanings or to laziness, avoiding the task of coining new words that

30A subtle comment on the triple relationship among science, philosophy, and technology may
be quoted here: “The union between science and philosophy was possible only during a period
of separation between science and technology. Modem science was born when technology
became scientific. The union of science and technology was responsible for the separation
between science and philosophy.” (Philipp Frank, 1957, p. 29.)
31 In discussing the alleged juxtaposition of science, on the one hand, and practical arts, technol¬
ogy, and engineering, on the other, one may ask, as Saul Gorn does in his lead paper in this
volume, why the words academic and practical are so often used as mutual insults. The reason,
I submit, is that people often like to explain their career choices by appeals to higher principles:
The academic looks down on the practical man who is obtuse to the eternal values of the search
for learning for the sake of learning; the man of practice looks down on the long-haired “college
prof” who wastes his time on merely “academic” matters and is oblivious to practical affairs.
Both groups may be happy in their “superior wisdom.”
670 FRITZ MACHLUP

would express what is actually meant.32 The word “computing” acquired a

new meaning when it was realized that the electronic computer could do
other things than what it had originally been designed for.

Dictionary Definitions

The word “to compute” is a direct derivation from the Latin computare,
where it has the same meaning as in English: “[T]o determine by computa¬
tion; to reckon, count; to take account of; [and] to make computation.” Its
synonym is “to calculate.”
The computing engine or computing machine, later called the computer,
was designed for number manipulation by means of symbol manipulation.
Charles Babbage (1792-1871), realizing that his machine could do more than
numerical calculations, called his device an analytical engine.33 The elec¬
tronic machines developed in the twentieth century, though capable of ma¬
nipulating symbols of any sort, were nevertheless designated as computers,
since the primary use for which they were intended was numerical computa¬
tion.

Multipurpose Machines

It did not take long for nonnumerical tasks to develop, to grow, and to
overtake the numerical ones. Some enterprising people might have come
forth with a new name for that miraculous machine that could do so many
things besides computing—but this did not happen or, at least, any new
names suggested were not accepted. So the noun “computer” stuck even
after it was generally known that the computer was capable of doing scores
of things other than computing. As a matter of fact, many computer experts
began to extend the meaning of the verb “computing” to make it comprise
virtually everything that can be done with a computer. This semantic expan¬
sion has caused some, though not serious, obstacles in professional and
scientific communication.

Now a Multipurpose Word

The logic of the semantic change is a little primitive: If a computer is a

device for computation, then anything that is done by a computer should be

32 Among the stock examples of a long time ago are manufacture (meaning made by hand, now
used for the opposite, made by machine) and manuscript (meaning written by hand, now
including the opposite, a typescript).
33 “Meanwhile ... he [Babbage] was at work on . . . his Analytical Engine, a grander, bigger,
all-purpose calculating machine. It would not only be capable of arithmetic calculations, but it
would also be capable of analysis and of tabulating any function whatever.” (McCorduck, 1979,
pp. 25-26.)
SEMANTIC QUIRKS IN STUDIES OF INFORMATION 671

called computation. It is easy to understand why more conservative seman-

ticists object to this peculiar logic. For it leads to possible confusion by
intelligent readers who have not yet been brainwashed; for example, when
they read a sentence or clause to the effect that “mental processes are
computational processes,” they are most likely to think of processes of
numerical computation—but would be wrong.
Several writers in this volume take it for granted that computing means
symbol manipulation. Allen Newell says that when the artificial intelligence
community emphasizes symbol manipulation, they do not mean nonnumer-
ical processing but only that prominence of many nonnumerical tasks
“make[s] the characterization of computers as number manipulators no
longer ring true.” Avron Barr’s paper carries the title “Artificial Intelli¬
gence: Cognition as Computation”; he does not even mention the possibility
that readers may associate computation with numbers. To the writers on
cognitive science it is quite obvious that computational psychology and
computational linguistics do not deal with numerical computations but only
with programs for and processes by the computer.
The traditionalist, who does not like semantic changes that could with just
a little care be avoided, will regret the expansion of the intension of comput¬
ing and computation. Are those who want to use these words in their original
meanings now forced to add number or numerical as modifiers to indicate
what they mean? They feel robbed by the appropriation of their terms by a
“gang of wordlifters.”34
Incidentally, the expansion of the meaning of the word computing did not
go all the way to having it include everything that can be done with a
computer. The most notable exceptions refer to processes where symbol
manipulation seems more prominent: Word-processing has not yet become
text computation; composing of music has not yet become music computa¬
tion; composing of camera-ready copy for the printer has not become com¬
putational composition. That may still come. The expansion that has already
taken place has reached the point of no return. Readers will have to accept
the notion of nonnumerical computation.

34When I requested Zenon Pylyshyn’s view on this semantic question, he replied: “Although
the term does have its roots in early calculating machines, my impression is that in psychology,
philosophy of mind, and certainly computer science it is viewed as an entirely neutral term
which designates what computers do—i.e., manipulate symbols. The public, of course, still
equates that with calculation over numbers, but in this case it is the public’s idea which will
change, rather than the term.” (Pylyshyn in private correspondence, August 20, 1982.)
 CUMULATIVE LIST OF REFERENCES

Abelson, Robert P., “Computer Simulation of‘Hot’ Cognition,” in Tompkins, Sil¬

van S., and Messick, Samuel, eds., Computer Simulation and Personality: Fron¬
tier of Psychological Theory (New York: Wiley, 1963), pp. 277-298.
Abelson, Robert P., “The Structure of Belief Systems,” in Schank, Roger C., and
Colby, Kenneth M., eds., Computer Models of Thought and Language (San
Francisco: W. H. Freeman, 1973), pp. 287-339.
Ackoff, Russell L., “Towards a Behavioral Theory of Communication,” in Man¬
agement Science, vol. 4 (1957-1958), pp. 218-234.
Ackoff, Russell L., “General Systems Theory and Systems Research—Contrasting
Conceptions of System Science,” in General Systems, vol. 8 (1963), pp. 117—
124.
Ackoff, Russell L., “Towards a System of System Concepts,” in Management
Science, vol. 17 (July 1971), pp. 661-671.
Ackoff, Russell L., and Emery, Fred E., On Purposeful Systems (Chicago: Aldine-
Atherton, 1972).
ACM Curriculum Committee on Computer Science, “An Undergraduate Program
in Computer Science—Preliminary Recommendations.” A Report from the
ACM Curriculum Committee on Computer Science, in Communications of the
ACM, vol. 8 (Sept. 1965), pp. 543-552.
ACM Curriculum Committee on Computer Science, “Curriculum 68: Recom¬
mendations for Academic Programs in Computer Science,” in Communications
of the ACM, vol. 11 (Mar. 1968), pp. 151-197.
ADA Reference Manual (Washington, D.C.: U.S. Department of Defense, July,
1980; rev., summer 1982).
Akaike, Hirotugu, “Stochastic Theory of Minimal Realization,” in IEEE Transac¬
tions on Automatic Control, vol. AC-19 (Dec. 1974), pp. 667-674.
Amarel, Saul, “On Representations of Problems of Reasoning about Actions,” in
Michie, Donald, ed., Machine Intelligence 3 (Edinburgh: Edinburgh University
Press, 1968), pp. 131-172.
Amarel, Saul, “Computer Science,” in Encyclopedia of Computer Science (New
York: Petrocelli/Charter, 1976), pp. 314-318.
Amari, Shun-ichi, “A Method of Statistical Neurodynamics,” in Kybernetik, vol. 14
(1974), pp. 201-215.
Amari, Shun-ichi, and Arbib, Michael A., eds., Competition and Cooperation in
Neural Nets (New York: Springer-Verlag, 1982).
AmpLre, Andre-Marie, Essai sur la philosophie des sciences, ou Exposition analy-
tique d’une classification naturelle de toutes les connaissances humaines (Paris:
Bachelier, 1834; seconde partie [ed. Jean Jacques Ampere], Paris, 1838, 1843;
another ed., Bruxelles: Culture et Civilisation, 1966).

673
674 CUMULATIVE LIST OF REFERENCES

Anderson, John R., Language, Memory, and Thought (Hillsdale, N. J.: Lawrence
Erlbaum, 1976).
Anderson, John R., and Bower, Gordon H., Human Associative Memory (Wash¬
ington, D.C.: V. H. Winston, 1973).
Aoki, Masanao, Optimal Control and Systems Theory in Dynamic Economic Analy¬
sis (New York: North-Holland, 1976).
Arbib, Michael A., “A Partial Survey of Cybernetics in Eastern Europe and the
Soviet Union,” in Behavioral Science, vol. 11 (1966), pp. 193-216.
Arbib, Michael A., The Metaphorical Brain: An Introduction to Cybernetics as
Artificial Intelligence and Brain Theory (New York: Wiley-Interscience, 1972).
Arbib, Michael A., “Artificial Intelligence and Brain Theory: Unities and Diver¬
sities,” in Annals of Biomedical Engineering, vol. 3 (Sept. 1975a), pp. 238-274.
Arbib, Michael A., “Cybernetics After 25 Years: A Personal View of System
Theory and Brain Theory,” in IEEE Transactions on Systems, Man, and Cyber¬
netics, vol. SMC-5 (1975b), pp. 359-363.
Arbib, Michael A., “Perceptual Structures and Distributed Motor Control,” in
Brooks, Vernon B., ed., Handbook of Physiology: Section 1: The Nervous
System, Vol. II: Motor Control (Bethesda, Md.: The American Physiological
Society, 1981), pp. 1449-1480.
Arbib, Michael A., “Modeling Neural Mechanisms of Visuomotor Coordination in
Frog and Toad,” in Amari, Shun-ichi, and Arbib, Michael A., eds., Competition
and Cooperation in Neural Nets (New York: Springer-Verlag, 1982), pp. 342-
370.
Arbib, Michael A.; Caplan, David; and Marshall, John C.,eds., Neural Models of
Language Processes (New York: Academic Press, 1982).
Arbib, Michael A., and Manes, Ernest G., “Machines in a Category,” in Journal of
Pure and Applied Algebra, vol. 19 (Dec. 1980), pp. 9-20.
Arden, Bruce W., ed., What Can Be Automated? The Computer Science and En¬
gineering Study (COSERS) (Cambridge, Mass.: MIT Press, 1980a).
Arden, Bruce W., “COSERS Overview,” in Arden, Bruce W., ed.. What Can Be
Automated? The Computer Science and Engineering Study (COSERS) (Cam¬
bridge, Mass.: MIT Press, 1980b), pp. 3-31.
Aristotle, The Nicomachean Ethics of Aristotle, with English notes by J. S. Brewer
(Oxford, England: Henry Slatter, 1836; another ed., New York: G. P. Putnam
Sons, 1926). [1st ed., 1513]
Armer, Paul, “Attitudes Toward Intelligent Machines,” in Feigenbaum, Edward
A., and Feldman, Julian, eds., Computers and Thought (New York: McGraw-
Hill, 1963), pp. 389-405.
Arnol’d, V. I., “Lectures on Bifurcations in Versal Families,” in Russian Mathe¬
matical Surveys, vol. 27 (Sept./Oct. 1972), pp. 54-123.
Arrow, Kenneth J., “Economic Equilibrium,” in Sills, David L., ed., International
Encyclopedia of the Social Sciences, vol. 4 (New York: Macmillan and Free
Press, 1968), pp. 376-388.
Arrow, Kenneth J., The Limits of Organization (New York: W. W. Norton, 1974).
Arrow, Kenneth J., “The Economics of Information,” in Dertouzos, Michael L.,
and Moses, Joel, eds., The Computer Age: A Twenty-Year View (Cambridge,
Mass.: MIT Press, 1979), pp. 306-317.
CUMULATIVE LIST OF REFERENCES 675

Ashby, W. Ross, Design for a Brain: The Origin of Adaptive Behaviour (New York:
Wiley, 1952; 2nd ed., rev., 1960).
Ashby, W. Ross, An Introduction to Cybernetics (London: Chapman and Hall, 1956).
Asratyan, G. E., and Simonov, P., How Reliable is the Brain? (Moscow: Mir, 1975).
Atkinson, Anthony A., and Feltham, Gerald A., “Information in Capital Markets:
An Agency Theory Perspective,” Working Paper, Faculty of Commerce and
Business Administration, University of British Columbia, Vancouver, B.C.,
1980.
Attneave, Fred, Applications of Information Theory to Psychology: A Summary of
Basic Concepts, Methods, and Results (New York: Holt, 1959).
Augustinus, Aurelius, St. Augustine’s Confessions (Cambridge, Mass.: Harvard
University Press, 1950-1951).
Backus, John, et al., “The FORTRAN Automatic Coding System,” in Proceedings
of the Western Joint Computer Conference, sponsored by the Institute of Radio
Engineers, the American Institute of Electrical Engineers, and the Association
for Computing Machinery and held [in Los Angeles] Feb. 26-28, 1957; pp. 188—
198.
Backus, John, “Can Programming Be Liberated From the von Neumann Style? A
Functional Style and Its Algebra of Programs” [the 1977 ACM Turing Award
Lecture], in Communications of the ACM, vol. 21 (Aug. 1978), pp. 613-641.
Baiman, Stanley, “Agency Research in Managerial Accounting: A Survey,” in
Journal of Accounting Literature, vol. 1 (spring 1980), pp. 154-213; reprinted in
Mattessich, Richard, et al., Modern Accounting Research: History, Survey, and
Guide (Vancouver, B.C.: Canadian Certified General Accountants Research
Foundation, forthcoming 1984).
Baker, Jeffrey J. W., and Allen, Garland E., The Study of Biology (Reading,
Mass.: Addison-Wesley, 4th ed., 1982).
Ballard, Dana H., and Brown, Christopher M., Computer Vision (Englewood
Cliffs, N.J.: Prentice-Hall, 1982).
Baralt-Torrijos, Jorge, “A Programmatic Interpretation of Combinatory Logic”
(Ph.D. diss., Georgia Institute of Technology, 1973).
Bar-Hillel, Yehoshua, Language and Information (Reading, Mass.: Addison-
Wesley, 1964).
Bar-Hillel, Yehoshua, and Carnap, Rudolf, “Semantic Information,” in Jackson
of Burnley, Willis Jackson, Baron, ed., Communication Theory. Papers read at a
symposium on Applications of Communication Theory, held [at the Institution
of Electrical Engineers, London] Sept. 22-26, 1952 (London: Butterworths,
1953a), pp. 503-512.
Bar-Hillel, Yehoshua, and Carnap, Rudolf, “Semantic Information,” in The Brit¬
ish Journal for the Philosophy of Science, vol. 4 (Aug. 19536), pp. 147-157.
Barker, Roger G., and Wright, Herbert F., The Midwest and Its Children: The
Psychological Ecology of an American Town (Evanston, Ill.: Row, Peterson,
1954; New York: Harper and Row, 1955).
Barnea, Amir; Haugen, Robert A.; and Senbet, Lemma W., “Market Imperfec¬
tions, Agency Problems, and Capital Structure: A Review,” in Financial Man¬
agement, vol. 10 (summer 1981), pp. 7-22.
Barr, Avron; Bennett, James S.; and Clancey, William J., Transfer of Expertise:
676 CUMULATIVE LIST OF REFERENCES

A Theme for Al Research (Stanford University, Heuristic Programming Project,

Working Paper No. HPP-79-11, 1979).
Barr, Avron, and Feigenbaum, Edward A., eds., The Handbook of Artificial Intelli¬
gence, vols. 1, 2 (Los Altos, Calif.: William Kaufmann, 1981).
Bartlett, Frederic C., Remembering: A Study in Experimental and Social Psychol¬
ogy (Cambridge: Cambridge University Press, 1932).
Barzun, Jacques, Darwin, Marx, Wagner (Garden City, N.Y.: Anchor Doubleday,
2nd ed., 1958).
Bateson, Gregory, Steps to an Ecology of Mind (New York: Ballantine Books,
1972).
Bauer, Walter F.; Juncosa, Mario L.; and Perlis, Alan J., “ACM Publication
Policies and Plans,” in Communications of the ACM, vol. 6 (Apr. 1959), pp.
121-122.
Baumol, William J., and Marcus, Matityahu, Economics of Academic Libraries
[prepared for the Council on Library Resources by Mathematica Inc.] (Washing¬
ton, D.C.: American Council on Education, 1973).
Baumol, William J., and Oates, Wallace, The Theory of Environmental Policy
(Englewood Cliffs, N.J.: Prentice-Hall, 1975).
Beadle, George Wells, Genetics and Modern Biology (Philadelphia: American
Philosophical Society, 1963a).
Beadle, George Wells, The New Biology and the Nature of Man [the Dewey F.
Fayerburg Memorial Lecture, 1963], in Phoenix (Ann Arbor: University of
Michigan, 1963b), vol. 1.
Beam, Robert D., “Testing the Integrated Social Science Hypothesis: An Economic
Approach” (Ph.D. diss.. University of Cincinnati, 1979).
Becker, Joseph D., “Reflections on the Formal Description of Behavior,” in Bob-
row, Daniel G., and Collins, Allan, eds., Representation and Understanding:
Studies in Cognitive Science (New York: Academic Press, 1975), pp. 83-102.
Beer, Stafford, Brain of the Firm (London: Allen Lane, 1972).
Beer, Stafford, “Progress in Cybernetics and Systems,” in Trappl, Robert, and
Hanika, Francis de Paula, eds., Progress in Cybernetics and Systems Research,
Vol. II: Socio-Economic Systems, Cognition and Learning, Systems Education,
Organization and Management. Papers presented at a symposium organized by
the Austrian Society for Cybernetic Studies and held [in Vienna] Apr. 1974 (New
York: Halsted Press, 1975), introduction.
Bell Laboratories, “UNIX Time-Sharing System,” in Bell System Technical Jour¬
nal, vol. 57, no. 6 (July-Aug. 1978), pp. 1897-2312, part 2.
Bellman, Richard, Introduction to the Mathematical Theory of Control Processes,
2 vols. (New York: Academic Press, 1971a).
Bellman, Richard, Mathematics, Systems and Society: An Informal Essay (Stock¬
holm: Swedish Natural Science Research Council [NFR], report no. 2, 1971/?).
Bensoussan, Alain, Filtrage optima! des systemes lineaires (Paris: Dunod, 1971).
Bensoussan, Alain; Delfour, M.; and Mitter, S. K., Control of Infinite Dimen¬
sional Systems (Cambridge, Mass.: MIT Press, 1982).
Bensoussan, Alain, and Lions, Jacques Applications des inequations
Louis,
variationnelles en controle stochastiques (Paris: Dunod, 1978).
CUMULATIVE LIST OF REFERENCES 677

Bensoussan, Alain, and Lions, Jacques Louis, Controle impulsionnel et inequations

quasi variationnelle (Paris: Dunod, 1981).
Bentham, Jeremy, Introduction to the Principles of Morals and Legislation (Oxford:
Oxford University Press, 1789).
Bentham, Jeremy, Chrestomathia (London: Payne and Foss, 1816).
Berger, J., “A Comment on Information Theory and Central Dogma of Molecular
Biology,” in Journal of Theoretical Biology, vol. 65 (Mar. 1977), pp. 393-395.
Berlekamp, Elwyn R., Key Papers in the Development of Coding Theory (New
York: IEEE Press, 1974).
Berlekamp, Elwyn R., ‘‘The Technology of Error-Correcting Codes,” in Proceed¬
ings of the IEEE, vol. 68 (May 1980), pp. 564-593.
Berlinski, David,On Systems Analysis: An Essay Concerning the Limitations of
Some Mathematical Methods in the Social, Political and Biological Sciences
(Cambridge, Mass.: MIT Press, 1976).
Bernard, Claude, Introduction to the Study of Experimental Medicine (first pub¬
lished in French, 1865; English trans. H. D. Greene, London: Macmillan, 1927).
Bernard, Claude, Leqons sur les phenomenes de la vie communs aux animaux et
aux vegetaux (Paris: J. B. Bailliere et fils, 1878-1879), 2 vols.
Bernstein, Jeremy, “Profiles: Marvin Minsky,” in The New Yorker, Dec. 14, 1981,
pp. 50-126.
Bernstein, NikolaiA., The Coordination and Regulation of Movement (New York:
Pergamon Press, 1967). Translated from the Russian.
Bertalanffy, Ludwig von, Kritische Theorie der Formbildung (Berlin, Borntraeger,
1928).
Bertalanffy, Ludwig von, Modern Theories of Development: An Introduction to
Theoretical Biology (London: Oxford University Press, 1933; New York: Har¬
per Torchbooks, 1962). Translation and adaptation by J. H. Woodger of
Kritische Theorie der Formbildung, Berlin: Borntraeger, 1928.
Bertalanffy, Ludwig von, “Zu einer allgemeinen Systemlehre,” Blatter fur
deutsche Philosophic, vol. 18, nos. 3/4 (1945). Reprinted in Biologia Generalis,
vol. 19 (1949), pp. 114-129.
Bertlanffy, Ludwig von, “An Outline of General System Theory,” in The British
Journal for the Philosophy of Science, vol. 1 (May 1950), pp. 134-165.
Bertalanffy, Ludwig von, Robots, Men, and Minds (New York: Braziller, 1967).
Bertalanffy, Ludwig von, General System Theory: Foundations, Development,
Applications (New York: Braziller, 1968).
Berwick, RobertC., and Weinberg, A., The Grammatical Basis of Linguistic Per¬
formance (Cambridge, Mass.: MIT Press, 1983).
Bever, Thomas G., “The Cognitive Basis for Linguistic Structures,” in Hayes, John
R., ed., Cognition and the Development of Language (New York: Wiley, 1970),
pp. 279-352.
Bever, Thomas G., “Broca and Lashley were Right: Cerebral Dominance is an
Accident of Growth,” in Caplan, David, ed., Biological Studies of Mental Pro¬
cesses, proceedings of a Conference on Maturational Factors in Cognitive Devel¬
opment and the Biology of Language, held [at MIT] June 1978 (Cambridge,
Mass.: MIT Press, 1980a), pp. 186-230.
678 CUMULATIVE LIST OF REFERENCES

Bever, Thomas G., “Normal Acquisition Processes Explain the Critical Period for
Language Learning,” in Diller, Karl C., ed.. Individual Differences and Univer¬
sal in Language Learning Aptitude (Rowley, Mass.: Newbury House, 1980/?),
pp. 176-198.
Bever, Thomas G., and Hurtig, R. R., “Detection of a Non-Linguistic Stimulus is
Poorest at the End of a Clause,” in Journal of Psycholinguists Research, vol. 4
(1975), pp. 1-7.
Bhagavad-Gita, The, trans. Christopher Isherwood (London: Phoenix House, 1947).
Binet, Alfred, and Simon, Theodore, The Development of Intelligence in Children:
The Binet-Simon Scale, trans. Elisabeth S. Kite (Baltimore: Williams and Wil¬
kins, 1916).
Bitzer, Donald, “The Wide World of Computer-Based Education,” in Rubinoff,
Morris, and Yovits, Marshall C., eds., Advances in Computers, vol. 15 (New
York: Academic Press, 1976), pp. 239-284.
Black, Max, “The Definition of Scientific Method,” in Stauffer, Robert C., ed..
Science and Civilization (Madison: University of Wisconsin Press, 1949), pp.
67-95.
Blaivas, Alexander, “Visual Analysis,” in Mathematical Biosciences, vol. 28
(1977), pp. 101-149.
Bloomfield, Leonard, An Introduction to the Study of Language (New York: Holt,
1914).
Bloomfield, Leonard, Language (New York: Holt, 1933).
Bloomfield, Leonard, “Linguistic Aspects of Science,” in International Encyclope¬
dia of Unified Science, vol. 1, no. 4 (Chicago: University of Chicago Press,
1939).
Blumenthal, Arthur L., The Process of Cognition (Englewood Cliffs, N.J.: Pren¬
tice-Hall, 1977).
Bobrow, Daniel G., “Dimensions of Representation,” in Bobrow, Daniel G., and
Collins, Allan, eds., Representation and Understanding: Studies in Cognitive
Science (New York: Academic Press, 1975), pp. 1-34.
Bobrow, Daniel G., ed., Special Issue on Non-Monotonic Logic, Artificial Intelli¬
gence, vol. 13 (Apr. 1980), pp. 1-172.
Bobrow, Daniel G., and Collins, Allan, eds,.. Representation and Understanding:
Studies in Cognitive Science (New York: Academic Press, 1975).
Bobrow, Daniel G., and Raphael, Bertram, “New Programming Languages for
Artificial Intelligence Research,” in ACM Computing Surveys, vol. 6 (Sept.
1974), pp. 153-174.
Bobrow, Robert J., and Brown, John Seely, “System Understanding: Synthesis,
Analysis, and Contingent Knowledge in Specialized Understanding Systems,”
in Bobrow, Daniel G., and Collins, Allan, eds.. Representation and Understand¬
ing: Studies in Cognitive Science (New York: Academic Press, 1975), pp. 103—
129.
Bode, Hendrik; Mosteller, Frederick; Tukey, John; and Winsor, Charles, “The
Education of a Scientific Generalist,” in Science, vol. 109 (June 1949), pp. 553—
558.
Boden, Margaret A., Purposive Explanation in Psychology (Cambridge, Mass.:
Harvard University Press, 1972).
CUMULATIVE LIST OF REFERENCES 679

Boden, Margaret A., Artificial Intelligence and Natural Man (New York: Basic
Books, 1977).
Boden, Margaret A., Minds and Mechanisms: Philosophical Psychology and Com¬
putational Models (Ithaca, N.Y.: Cornell University Press, 1981a).
Boden, Margaret A., “The Case for a Cognitive Biology,” in Boden, Margaret A.,
Minds and Mechanisms: Philosophical Psychology and Computational Models
(Ithaca, N.Y.: Cornell University Press, 19816), pp. 89-112.
Bogdanov [Malinovskii], Aleksandr Aleksandrovich, Vseobshchaia organization-
naia nauka, tektologiia [Tektology: The Universal Science of Organization],
3 vols. (St. Petersburg: Izdatelstvo, 1913; 2nd ed., Moscow: Izdatelstvo,
1922).

Bogdanov [Malinovskii], Aleksandr Aleksandrovich, Allgemeine Organisations-

lehre—Tektologie (Berlin: Organisationsverlagsgesellschaft GmbH, 1926).
Translation from Russian to German of vols. 1 and 2 of Vseobshchaia or-
ganizationnaia nauka, tektologiia, 1913 and 1922.
Bogdanov [Malinovskii], Aleksandr Aleksandrovich, Essays in Tektology, trans.
and ed. George Gorelik (Seaside Calif.: Intersystems Publications, 1980).
Boguslaw, Robert, The New Utopians: A Study of System Design and Social
Change (Englewood Cliffs, N.J.: Prentice-Hall, 1965).
Bohm, Stephan, “Das Biespiel der ‘unsichtbaren’ Hand,” in Acham, Karl, ed.,
Gesellschaftliche Prozesse (Graz, Austria: Academia, 1982).
Boltzmann, Ludwig von, “Weitere Studien iiber das Warmegleichgewicht unter
Gasmolekiilen,” in Kaiserliche Akademie der Wissenschaften [Vienna] Sitz-
ungsberichte, II Abteil 66 (1872), pp. 275-370.
Boltzmann, Ludwig von, “Uber die Beziehung zwischen dem zweiten Hauptsatze
der mechanischen Warmetheorie und der Wahrscheinlichkeitsrechnung respek-
tive den Satzen iiber das Warmgleichgewicht” [“On the Relation between the
Second Law of Thermodynamics and Probability Theory and the Laws of Ther¬
mal Equilibrium”], in Kaiserliche Akademie der Wissenschaften [Vienna] Sitz-
ungsberichte, II Abteil 76 (1877), pp. 373-435.
Boltzmann, Ludwig von, “Der zweite Hauptsatz der mechanischen Warme¬
theorie,” a paper written in 1886, reproduced in Boltzmann, Ludwig, Populdre
Schriften (Leipzig: Johann Ambrosius Barth, 1905a), pp. 25-50.
Boltzmann, Ludwig von, “Uber statistische Mechanik,” a paper delivered at a
congress in St. Louis, 1904, reproduced in Boltzmann, Ludwig, Populdre Schrif¬
ten (Leipzig: Johann Ambrosius Barth, 19056), pp. 345-363.
Bongard, Mikhail Moiseevich, Pattern Recognition, ed. Joseph K. Hawkins,
and trans. Theodore Cheron (Rochelle Park, N.J.: Hayden, Spartan Book,
1970).
Bookstein, Abraham, “The Bibliometric Distributions,” in The Library Quarterly,
vol. 46 (Oct. 1976), pp. 416-423.
Boole, George, An Investigation of the Laws of Thought, on which are Founded the
Mathematical Theories of Logic and Probabilities (London: Walton and Ma-
berly, 1854. Republished as vol. 2 of George Boole’s Collected Logical Works,
Chicago: Open Court, 1916; New York: Dover ed., 1961).
Boulding, Kenneth E., “Notes on the Information Concept,” in Explorations
[Toronto], vol. 6 (1955), pp. 103-112.
680 CUMULATIVE LIST OF REFERENCES

Boulding, Kenneth E., The Image: Knowledge in Life and Society (Ann Arbor:
University of Michigan Press, 1956a).
Boulding, Kenneth E., “General Systems Theory—The Skeleton of Science,” in
Management Science, vol. 2 (Apr. 1956b), pp. 197-208.
Boulding, Kenneth E., The Meaning of the Twentieth Century (New York: Harper,
1964).
Boulding, Kenneth E., “The Economics of Knowledge and the Knowledge of
Economics,” in American Economic Review, vol. 56, no. 2 (1966), pp. 1-13.
Boulding, Kenneth E., Eco-dynamics: A New Theory of Societal Evolution (Bev¬
erly Hills, Calif.: Sage Publications, 1978).
Bowler, T. Downing, General Systems Thinking: Its Scope and Applicability (New
York: Elsevier/North-Holland, 1981).
Brachman, Ronald J., and Smith, Brian C., eds., Special Issue on Knowledge
Representation, SIGART Newsletter, vol. 70 (Feb. 1980), pp. 1-138.
Branscomb, Lewis M., “Information: The Ultimate Frontier,” in Science, vol. 203
(12 Jan. 1979), p. 143.
Bresnan, Joan, “A Realistic Transformational Grammar,” in Halle, Morris; Bres-
nan, Joan; and Miller, George A., eds., Linguistic Theory and Psychological
Reality (Cambridge, Mass.: MIT Press, 1978), pp. 1-59.
Briet, Suzanne, Qu’est-ce que la documentation? (Paris: Editions Documentaires
Industrielles et Techniques, 1951).
Brillouin, Leon, “Negentropy Principle of Information,” in Journal of Applied
Physics, vol. 24 (Sept. 1953), pp. 1152-1163.
Brillouin, Leon, Science and Information Theory (New York: Academic Press,
1956; 2nd ed. 1962).
Brockett, Roger W., Finite Dimensional Linear Systems (New York: Wiley, 1971).
Brockett, Roger W., “The Geometry of Partial Realization Problems,” in Proceed¬
ings of a Conference on Decision and Control organized by the Institute of
Electrical and Electronics Engineers (New York: IEEE Press, 1978).
Broglie, Louis de, “L’Avenir de la physique,” in Broglie, Louis de, et al., L’Avenir
de la science (Paris: Plon, 1941), pp. 1-35.
Broglie, Louis de, La cybernetique: theorie du signal et de Tinformation (Paris:
Editions de la Revue d’Optique Theorique et Instrumentale, 1951), p. v.
Brookes, Bertram C., “Robert Fairthorne and the Scope of Information Science,”
in Journal of Documentation, vol. 30 (June 1974), pp. 139-152.
Brookes, Bertram C., “A New Paradigm for Information Science,” in The Informa¬
tion Scientist, vol. 10 (Sept. 1976), pp. 103-111.
Brown, Robert R., Explanation in Social Science (Chicago: Aldine-Atherton, 1963).
Brunovsky, P., “A Classification of Linear Controllable Systems,” in Kybernetika,
vol. 6 (1970), pp. 176-188.
Bryson, Arthur, and Ho, Y. C., Applied Optimal Control (New York: Halsted
Press, 1968).
Buchanan, James, “The Justice of Natural Liberty,” in O’Driscoll, Gerald P., Jr.,
ed., Adam Smith and Modern Political Economy: Bicentennial Essays on The
Wealth of Nations (Ames: Iowa State University Press, 1979) pp. 117-131.
CUMULATIVE LIST OF REFERENCES 681

Buckland, Michael K., Library Services in Theory and Context (New York: Perga-
mon Press, 1983).
Buckley, Walter, Sociology and Modern Systems Theory (Englewood Cliffs, N.J.:
Prentice-Hall, 1967).
Bunge, Mario, Treatise on Basic Philosophy, Vol. IV, Ontology II: A World of
Systems (Dordrecht, Holland, and Boston: D. Reidel, 1979a).
Bunge, Mario, “A Systems Concept of Society: Beyond Individualism and Hol¬
ism,” in Theory and Decision, vol. 10 (Jan. 1979b), PP- 13-30.
Burks, Arthur W., ed., Essays on Cellular Automata (Urbana: University of Illnois
Press, 1970).
Burmeister, Edwin, and Dobell,A. Rodney, Mathematical Theories of Economic
Growth (New York: Macmillan, 1970).
Bush, Vannevar, “As We May Think” (1945). Reprinted in Pylyshyn, Zenon W.,
ed., Perspectives on the Computer Revolution (Englewood Cliffs, N.J.: Pren¬
tice-Hall, 1970), pp. 47-59.
Butler, Pierce, An Introduction to Library Science (Chicago: University of Chicago
Press, 1933).
Butterfield, Herbert,The Origins of Modern Science 1300-1800 (London: Bell,
1949; New York: Macmillan, 1951; Toronto: Clark, Irwin and Co., 1957).
Byrnes,C., and Falb, Peter L. “Applications of Algebraic Geometry in System
Theory,” in American Journal of Mathematics, vol. 101 (1979), pp. 337-363.
Cadwallader, M., “The Cybernetic Analysis of Change in Complex Social Organi¬
zations,” in American Journal of Sociology, vol. 65 (1959), pp. 154-157.
Cannon, Walter B., “Organization for Physiological Homeostasis,” in Physiologi¬
cal Reviews, vol. 9 (July 1929), pp. 399-431.
Cannon, Walter B., The Wisdom of the Body (New York: W. W. Norton, 1939).
Capek, Karel, R.U.R. (Rossum’s Universal Robots) (Garden City, N.Y.: Double¬
day, Page and Co., 1923).
Carnap, Rudolf, “Foundations of Logic and Mathematics,” in Neurath, Otto; Car¬
nap, Rudolf; and Morris, Charles, eds., Foundations of the Unity of Science:
Toward an International Encyclopedia of Unified Science, vol. 1 (Chicago: Uni¬
versity of Chicago Press, 1939a; bound ed., 1969), pp. 139-214.
Carnap, Rudolf, “Logical Foundations of the Unity of Science,” in Neurath, Otto;
Carnap, Rudolf; and Morris, Charles, eds., Foundations of the Unity of
Science: Toward an International Encyclopedia of Unified Science, vol. 1
(Chicago: University of Chicago Press, 1939b; bound ed., 1969).
Carnap, Rudolf, Logical Foundations of Probability (Chicago: University of
Chicago Press, 1950).
Carnap, Rudolf, and Bar-Hillel, Yehoshua, An Outline of a Theory of Semantic
Information (Cambridge, Mass.: MIT Research Laboratory of Electronics Tech¬
nical Report, 1952). Reprinted as chap. 15 in Bar-Hillel, Language and Informa¬
tion: Selected Essays on Their Theory and Application (Reading, Mass.: Addi-
son-Wesley, 1964).
Carroll,C. Edward, The Professionalization of Education for Librarianship
(Metuchen, N.J.: Scarecrow Press, 1970).
682 CUMULATIVE LIST OF REFERENCES

Cash, John L., Connectivity, Complexity, and Catastrophe in Large-Scale Systems

(New York: Wiley, 1979).
Cataloguing Rules: Author and Title Entries. Compiled by Committees of The Li¬
brary Association and of the American Library Association (London: The Li¬
brary Association, 1908).
Cerny, J., and Brunovsky, P., “A Note on Information Without Probability,” in
Information and Control, vol. 25 (June 1974), pp. 134-144.
Chaitin, Gregory J., ‘‘Algorithmic Information Theory,” in IBM Journal of Re¬
search and Development, vol. 21 (July 1977), pp. 350-359.
Chandler, Robert E.; Herman, Robert; and Montroll, Elliott W., ‘‘Traffic Dy¬
namics: Studies in Car Following,” in Operations Research, vol. 6 (Mar./Apr.,
1958), pp. 165-184.
Chase, William G., and Simon, Herbert A., ‘‘The Mind’s Eye in Chess,” in Chase,
William G., ed., Visual Information Processing, proceedings of the Eighth Sym¬
posium on Cognition, held [at Carnegie-Mellon University, Pittsburgh] in 1972
(New York: Academic Press, 1973), pp. 215-281.
Checkland, Peter, Systems Thinking, Systems Practice (Chichester, England:
Wiley, 1981).
CherrybColin, “A History of the Theory of Information,” in The Proceedings of
The Institution of Electrical Engineers, vol. 98 (Sept. 1951), pp. 383-393. Re¬
printed with minor changes as “The Communication of Information (An Histor¬
ical Review),” in American Scientist, vol. 40 (Oct. 1952), pp. 640-664.
Cherry, Colin, ed., Information Theory, proceedings of the Third Symposium on
Information Theory, held [in London] 1955 (London: Butterworths, 1956).
Cherry, Colin, On Human Communication: A Review, A Survey, and A Criticism
(Cambridge, Mass.: MIT Press, 1957; 2nd ed., 1966).
Cherry, Colin, ed., Information Theory, proceedings of the Fourth Symposium on
Information Theory, held [at the Royal Institution, London] Aug. 29-Sept. 2,
1960 (London: Butterworths, 1961).
Chomsky, Noam, Syntactic Structures (The Hague, Netherlands: Mouton, 1957).
Chomsky, Noam, “A Review of Verbal Behavior by B. F. Skinner,” in Language,
vol. 35 (1959), pp. 26-58.
Chomsky, Noam, “Formal Properties of Grammar,” in Luce, Robert Duncan; Bush,
Robert R.; and Galanter, Eugene, eds., Handbook of Mathematical Psychology,
vol. 2 (New York: Wiley, 1963), chap. 12, pp. 323-418.
Chomsky, Noam, Aspects of the Theory of Syntax (Cambridge, Mass.: MIT Press,
1965).
Chomsky, Noam, Cartesian Linguistics: A Chapter in the History of Rationalist
Thought (New York: Harper and Row, 1st ed., 1966).
Chomsky, Noam, Language and Mind (New York: Harcourt, Brace, and World,
1968; enlarged ed., Harcourt, Brace, Jovanovich, 1972).
Chomsky, Noam, “Conditions on Transformations,” in Anderson, Stephen R., and
Kiparsky, Paul, eds., A Festschrift to Morris Halle (Englewood Cliffs, N.J.:
Prentice-Hall, 1973).
Chomsky, Noam, The Logical Structure of Linguistic Theory (New York: Plenum
Press, 1975a).
CUMULATIVE LIST OF REFERENCES 683

Chomsky, Noam, Reflections on Language (New York: Pantheon, 19756).

Chomsky, Noam, “On Cognitive Structures and Their Development: A Reply to
Piaget,” in Piatelli-Palmarini, Massimo, ed., Language and Learning: The De¬
bate between Jean Piaget and Noam Chomsky (Cambridge, Mass.: Harvard
University Press, 1980a), pp. 35-54.
Chomsky, Noam, Rules and Representations (New York: Columbia University
Press, 19806).
Chomsky, Noam, Lectures on Government and Binding (Dordrecht, Holland: Foris,
1981) .
Chomsky, Noam, Some Concepts and Consequences of the Theory of Government
and Binding, Linguistic Inquiry Monograph 6 (Cambridge, Mass.: MIT Press,
1982) .
Chomsky, Noam, and Halle, Morris, The Sound Pattern of English (New York:
Harper and Row, 1968).
Chomsky, Noam, and Miller, George A., “Introduction to the Formal Analysis of
Natural Languages,” in Luce, Robert Duncan; Bush, Robert R.; and Galanter,
Eugene, eds., Handbook of Mathematical Psychology, vol. 2 (New York:
Wiley, 1963), chap. 11, pp. 269-321.
Church, Alonzo, The Calculi of Lambda-Conversion (Princeton, N.J.: Princeton
University Press, 1941).
Churchland, Paul M., Scientific Realism and the Plasticity of Mind (Cambridge :
Cambridge University Press, 1979).
Churchman, C. West, Predictions and Optimal Decision—Philosophical Issues of
Science of Values (Englewood Cliffs, N.J.: Prentice-Hall, 1961).
Churchman, C. West, “An Approach to General Systems Theory,” in Mesarovic,
Mihajlo D., ed., Views on General Systems Theory (New York: Wiley, 1964),
pp. 173-176.
Churchman, C. West, A Challenge to Reason (New York: McGraw-Hill, 1968a).
Churchman, C. West, The Systems Approach (New York: Dell, 19686).
Churchman, C. West, The Design of Inquiring Systems (New York: Basic Books,
1971).
Churchman, C. West, The Systems Approach and Its Enemies (New York: Basic
Books, 1979).
Churchwell, Charles D., The Shaping of American Library Education (Chicago:
American Library Association, 1975), ACRL Monograph 36.
Clark, Margaret, and Fiske, Susan, eds., Affect and Cognition (Hillsdale, N. J.:
Lawrence Erlbaum, 1982).
Coase, Ronald, “The Problem of Social Cost,” in Journal of Law and Economics,
vol. 3 (1960), pp. 1-14.
Coates, J. F., “Technology Assessment—Tool Kit,” in Chemical Technology, vol.
6, no. 6 (1976), pp. 272-383.
Cohen, Morris R., Reason and Nature: An Essay on the Meaning of Scientific
Method (New York: Harcourt, 1931).
Cohen, Paul R., “Models of Cognition: Overview,” in Cohen, Paul R., and Feigen-
baum, Edward A., eds.. The Handbook of Artificial Intelligence, vol. 3 (Los
Altos, Calif.: William Kaufmann, 1982), pp. 1-10.
684 CUMULATIVE LIST OF REFERENCES

Collins, Allan, “Why Cognitive Science?” in Cognitive Science, vol. 1 (Jan. 1977),
pp. 1-2.
Converse, Alvin O., “The Use of Uncertainty in a Simultaneous Search,” in Opera¬
tions Research, vol. 15 (Nov./Dec. 1967), pp. 1088-1095.
Copinger, Walter Arthur, “On the Necessity for the Formation of a Biblio¬
graphical Society of the United Kingdom and Suggestions as to its Operations,”
in The Library, vol. 4 (Jan./Dec. 1892), pp. 1-7.
Costa, Serpil de, “Foundation and Development of IFLA, 1926-1939,” in The
Library Quarterly, vol. 52 (Jan. 1982), pp. 41-58.
Costanza, Robert, “Embodied Energy and Economic Valuation,” in Science, vol.
210 (12 Dec., 1980), p. 1224.
Counelis, James S., “Cybernetics and General Systems—A Unitary Science (Infor¬
mation and the Unity of General Systems Theory and Cybernetics),” in Kyber-
netes, vol. 8 (1979), pp. 25-32.
Cowan, W. Maxwell, ed., Annual Review of Neuroscience (Palo Alto, Calif.: An¬
nual Reviews Inc., 1981).
Craik, Kenneth J. W., The Nature of Explanation (Cambridge: Cambridge Univer¬
sity Press, 1943).
Croce, Benedetto, “The Task of Logic,” in Encyclopaedia of the Philosophical
Sciences, Vol. 1: Logic (London: Macmillan, 1913).
Cuatrecasas, Pedro, and Greaves, Melvyn F., eds., Receptors and Recognition,
series A, vols. 1, 2 (London: Chapman and Hall, 1976).
Cyert, Richard M., and March, James G., A Behavioral Theory of the Firm (En¬
glewood Cliffs, N. J.: Prentice-Hall, 1963).
Cyranski, John F., “Measurement, Theory, and Information,” in Information and
Control, vol. 41 (June 1979), pp. 275-304.
Davis, Martin D., ed., The Undecideable: Basic Papers on Undecideable Proposi¬
tions, Unsolvable Problems, and Uncomputable Functions (Hewlett, N. Y.:
Raven Press, 1965).
Davis, Randall, “Applications of Meta-Level Knowledge to the Construction,
Maintenance, and Use of Large Knowledge Bases,” in Davis, Randall, and
Lenat, Douglas, Knowledge-Based Systems in Artificial Intelligence (New
York: McGraw-Hill, 1982), pp. 229-390.
Davis, Randall, and Lenat, Douglas, Knowledge-Based Systems in Artificial Intel¬
ligence (New York: McGraw-Hill, 1982).
Davis, Watson, “Regarding the Name: Documentation Institute of Science Service”
(Washington, D.C.: Documentation Institute of Science Service, 11 July 1935),
mimeographed.
Dawkins, Richard, The Selfish Gene (New York and London: Oxford University
Press, 1976).
Day, Richard H., “A Microeconomic Model of Business Growth, Decay and Cy¬
cles,” in Unternehmensfusching Band 11, Heft 1 (1967), pp. 1-20.
Day, Richard H., “Adaptive Processes and Economic Theory,” in Day, Richard H.,
and Groves, Theodore, eds., Adaptive Economic Models, proceedings of a Sym¬
posium on Adaptive Economics, held [at the University of Wisconsin, Madison]
Oct. 21-23, 1974 (New York: Academic Press, 1975).
Day, Richard H., “Dynamic Systems and Epochal Change,” in Sabloff, Jeremy A.,
CUMULATIVE LIST OF REFERENCES 685

ed., Simulations in Archeology (Albuquerque: University of New Mexico Press,

1981a).
Day, Richard H., “Unstable Economic Systems,” in Economic Notes, vol. 10,
no. 3 (Siena, Italy: Monte dei Pasche di Siena, 1981), pp. 3-15.
Day, Richard H., “Chaos in Recursive Economics,” in Journal of Economic Behav¬
ior and Organization (forthcoming 1983).
Day, RichardH., and Cigno, A., Modelling Economic Changes: The Recursive
Programming Approach (Amsterdam: North-Holland, 1978).
Debons, Anthony; King, Donald W.; Mansfield, Una; and Shirey, Donald L.,
The Information Professional: Survey of an Emerging Field (New York: Marcel
Dekker, 1981).
de Broglie, Louis [see “Broglie, Louis de”].
de Costa, Serpil [see “Costa, Serpil de”].
de Groot, Adriaan D. [see “Groot, Adriaan D. de”].
de Moivre, Abraham [see “Moivre, Abraham de”].
Demski, Joel S., “Information Improvement Bounds,” in Journal of Accounting
Research (spring 1972).
Demski, Joel S., Information Analysis (Reading, Mass.: Addison-Wesley, 1972; 2nd
ed., 1980).
Demski, Joel S., and Feltham, Gerald A., “Forecast Evaluation,” in The Account¬
ing Review, vol. 47 (July 1972), pp. 533-548.
Demski, JoelS., and Feltham, Gerald A., Cost Determination—A Conceptual
Approach (Ames: Iowa State University Press, 1976).
Demski, JoelS., and Feltham, Gerald A., “References in Information Econom¬
ics,” Working Paper, Faculty of Commerce, University of British Columbia,
Vancouver, B.C., 1981.
Demski, Joel S., and Feltham, Gerald A., Economic Returns to Accounting Infor¬
mation in a Multi-Person Setting (preliminary title; forthcoming in the series
Studies in Accounting Research of the American Accounting Association,
Sarasota, Florida: AAA, 1983).

Denbigh, Kenneth G., The Principles of Chemical Equilibrium, with Applications in

Chemistry and Chemical Engineering (Cambridge: Cambridge University Press,
1955).

Dennett, Daniel C., Brainstorms: Philosophic Essays on Mind and Psychology

(Montgomery, Vt.: Bradford Books/MIT Press, 1978a).
Dennett, Daniel C., “Intentional Systems,” in Dennett, Daniel C., Brainstorms:
Philosophic Essays on Mind and Psychology (Montgomery, Vt.: Bradford
Books/MIT Press, 1978b), PP- 3-22.
Denning, Peter J., “Working Sets Past and Present,” in IEEE Transactions on
Software Engineering, vol. 6 (Jan. 1980), pp. 64-84.
Dennis, Jack B., and Misunas, David P., “A Preliminary Architecture for a Basic
Data-Flow Processor,” in Proceedings of the Second Annual Symposium on
Computer Architecture, held [in Houston] by the IEEE Computer Society and
the Association for Computing Machinery, in cooperation with the University of
Houston, Jan. 20-22, 1975 (New York: Institute of Electrical and Electronics
Engineers, 1975), pp. 126-132.
686 CUMULATIVE LIST OF REFERENCES

Destutt de Tracy, Antoine Louis Claude, Projet d'elements d’ideologic a 1’usage

des ecoles centrales de la Republique francaise, 4 vols., Paris, an IX [1801]—
1818 (Paris: 3rd ed., Courcier, 1817).
Deutsch, Karl, “The Nerves of Government,” in General Systems Yearbook, vol.
21 (1963), pp. 125-176.
Dev, Parvati, “Perception of Depth Surfaces in Random-Dot Stereograms: A
Neural Model,” in International Journal of Man-Machine Studies, vol. 7 (July
1975), pp. 511-528.
Dewey, John, The Quest for Certainty: A Study of the Relation of Knowledge and
Action [the Gifford Lectures, University of Edinburgh] (New York: Minton,
Balch, and Co., 1929).
Dewey, Melvil, Classification and Subject Index for a Library (Amherst, Mass.:
1876; 18th ed., Lake Placid Club, New York: Forest Press, 1971).
Dickson, Gary R.; Senn, James A,; and Chervany, Norman L., “Research in Man¬
agement Information Systems: The Minnesota Experiments,” in Management
Science, vol. 23 (May 1977), pp. 913-923.
Dijkstra, Edsger W., “Notes on Structured Programming,” in Dahl, Ole-Johan;
Dijkstra, Edsger W.; and Hoare, C. A. R., Structured Programming (London
and New York: Academic Press, 1972).
Dijkstra, Edsger W., A Discipline of Programming (Englewood Cliffs, N. J.:
Prentice-Hall, 1976).
Dolan, F. T., “The Role of the Information Scientist,” in International Journal of
Man-Machine Studies, vol. 1 (1969), pp. 39-50.
Dowds, John P., “Application of Information Theory in Establishing Oil Field
Trends,” in Parks, George A., ed., Computers in the Mineral Industries (Stan¬
ford, Calif.: School of Earth Sciences, Stanford University, 1964).
Dray, William H., “Holism and Individualism in History and Social Science,” in
Encyclopedia of Philosophy, vol. 4 (New York: Macmillan and Free Press,
1967), pp. 53-58.
Dresher, B. Elan, and Hornstein, Norbert, “On Some Supposed Contributions of
Artificial Intelligence to the Scientific Study of Language,” in Cognition, vol. 4
(Dec. 1976), pp. 321-398.
Dresher, B. Elan, and Hornstein, Norbert, “Reply to Schank and Wilensky,” in
Cognition, vol. 5 (June 1977u), pp. 147-149.
Dresher, B. Elan, and Hornstein, Norbert, “Reply to Winograd,” in Cognition,
vol. 5 (Dec. 1977b), pp. 379-392.
Dreyfus, Hubert L., What Computers Can’t Do: A Critique of Artificial Reason
(New York: Harper and Row, 1972; rev. paperback ed., 1979).
Driesch, Hans, The History and Theory of Vitalism (London: Macmillan, 1914).
Translated from the German.
Dunlap, Knight, The Elements of Scientific Psychology (St. Louis: C. V. Mosby
Company, 1922).
Dworkin, Ronald, Taking Rights Seriously (Cambridge, Mass.: Harvard University
Press, 1978).
Eckhorn, R., and Popel, B., “Rigorous and Extended Application of Information
CUMULATIVE LIST OF REFERENCES 687

Theory to Afferent Visual Systems of a Cat,” in Kybernetik, vol. 16, no. 4

(1974), pp. 191-200.
Edelman, Gerald M., “Group Selection as the Basis for Higher Brain Function,” in
Schmitt, Francis O.; Worden, Frederic G.; Adelman, George; and Dennis,
Stephen G., eds., The Organization of the Cerebral Cortex, proceedings of a
colloquium held [at Woods Hole, Mass.] (Cambridge, Mass.: MIT Press, 1981).
Edelman, Gerald M., and Mountcastle, Vernon B., The Mindful Brain: Cortical
Organization and the Group-Selective Theory of Higher Brain Functions (Cam¬
bridge, Mass.: MIT Press, 1978).
Egan, Margaret E., and Shera, Jesse H., “Foundations of a Theory of Bibliogra¬
phy,” in The Library> Quarterly, vol. 22 (Apr. 1951), pp. 125-137.
Ehrlich, Paul, The Collected Papers of Paul Ehrlich, 2 vols., comp, and ed.
F. Himmelweit (London and New York: Pergamon Press, 1956).
Eigen, Manfred, “Self-Organization of Matter and the Evolution of Biological Mac¬
romolecules,” in Naturwissenschaften, vol. 58 (Oct. 1971), pp. 465-528.
Eigen, Manfred and Schuster, Peter, “The Hypercycle: A Principle of Natural
Self-Organization. Part A: Emergence of the Hypercycle,” in Naturwis¬
senschaften, vol. 64 (Nov. 1977), pp. 541-565.
Eigen, Manfred, and Schuster, Peter, “The Hypercycle: A Principle of Natural
Self-Organization. Part B: The Abstract Hypercycle,” in Naturwissenschaften,
vol. 65 (Jan. 1978a), pp. 7-41.
Eigen, Manfred, and Schuster, Peter, “The Hypercycle: A Principle of Natural
Self-Organization. Part C: The Realistic Hypercycle,” in Naturwissenschaften,
vol. 65 (July 1978/?), pp. 341-369.
Elias, Peter, “Coding and Information Theory,” a paper written in 1959, reprinted
in Evans, C. R., and Robertson, A. D. J., eds., Cybernetics (Baltimore: Univer¬
sity Park Press, 1968), pp. 253-266.
Elias, Peter, and Flower, Richard A., “The Complexity of Some Simple Retrieval
Problems,” in Journal of the Association for Computing Machinery, vol. 22
(July 1975), pp. 367-379.
El Sayed, Y., On the Use ofExergy and Thermoeconomics in the Design of Desali¬
nation Plants (Hanover, N.H.: Thayer School of Engineering, Dartmouth Col¬
lege, Jan. 1968).
El Sayed, Y., and Aplenc, A., Application of the Thermoeconomic Approach to the
Analysis and Optimization of a Vapor Compression Desalting System
(Hanover, N.H.: Thayer School of Engineering, Dartmouth College, Sept.
1968) .
El Sayed, Y., and Evans, Robert B., Thermoeconomics and the Design of Heat
Systems (Hanover, N.H.: Thayer School of Engineering, Dartmouth College,
Oct. 1968).
Emery, Fred E., Systems Thinking (Harmondsworth, England: Penguin Books,
1969) .
Emmet, Boris, and Jeuck, John E.,Catalogues and Counters: A History of Sears,
Roebuck, and Company (Chicago: University of Chicago Press, 1950).
Emonds, Joseph E., “Root and Structure Preserving Transformations” (Ph.D. diss.,
Massachusetts Institute of Technology, 1970).
688 CUMULATIVE LIST OF REFERENCES

Ermentrout, G. B., and Cowan, Jack D., “Large-Scale Spatially Organized Activ¬
ity in Neural Nets,” in SIAM Journal on Applied Mathematics, vol. 38 (Feb.
1980), pp. 1-21.
Ernst, George W., and Newell, Allen, GPS: A Case Study in Generality and
Problem Solving (New York: Academic Press, 1969).
Evans, Robert B., “Basic Relationships Among Entropy, Exergy, Energy, and
Availability,” in Spiegler, K. S., ed., Principles of Desalination (New York:
Academic Press, 1966), pp. 44-66.
Evans, Thomas G., “A Program for the Solution of Geometric-Analogy Intelligence
Test Questions,” in Minsky, Marvin L., ed., Semantic Information Processing
(Cambridge, Mass.: MIT Press, 1968), pp. 271-354.
Ewert, Jorg-Peter, Neuroethology: An Introduction to the Neurophysiological
Fundamentals of Behavior (Berlin and New York: Springer-Verlag, 1980).
Falb, Peter L., “Linear Systems and Invariants” (Lecture Notes, Control Group,
Lund University, Sweden, 1974).
Fama, Eugene F., “Agency Problems and the Theory of the Firm,” in Journal of
Political Economy, vol. 88 (Apr. 1980), pp. 288-307.
Fano, Robert M., Transmission of Information (Cambridge, Mass., and New York:
MIT Press and Wiley, 1961).
Farradane, Jason E. L., “The Institute: The First Twelve Years,” in The Informa¬
tion Scientist, vol. 4 (Dec. 1970), pp. 143-151.
Faurre, P., “Realisations markoviennes de processus stationnaires,” Research Re¬
port No. 13, INRIA (Rocquencourt, France: Institut National de Recherche en
Informatique et en Automatique, 1973).
Faurre, P.; Clerget, M.; and Germain, F., Operateurs rationnels positifs: Applica¬
tion a Vhyperstabilite et aux processus aleatoires (Paris: Dunod, 1979).
Feigenbaum, Edward A., “The Simulation of Verbal Learning Behavior,” in Ex¬
tending Man’s Intellect, proceedings of the Western Joint Computer Confer¬
ence, held [in Los Angeles] by the Institute of Radio Engineers, the American
Institute of Electrical Engineers, and the Association for Computing Machinery,
May 9-11, 1961, vol. 19, pp. 121-132.
Feigenbaum, Edward A., “Artificial Intelligence: Themes in the Second Decade,” in
Morrell, A. J. H., ed., Information Processing 68, proceedings of the IFIP
Congress 1968, organized by the International Federation for Information Pro¬
cessing [Edinburgh, Scotland] Aug. 5-10, 1968 (Amsterdam: North-Holland,
1969), pp. 1008-1024.
Feigenbaum, Edward A., “The Art of Artificial Intelligence, I: Themes and Case
Studies of Knowledge Engineering,” in IJCAI-77: Proceedings of the Fifth In¬
ternational Joint Conference on Artificial Intelligence, held [at the Massachu¬
setts Institute of Technology, Cambridge, Mass.] by the International Joint Con¬
ference on Artificial Intelligence, Aug. 22-25, 1977 (Pittsburgh: Carnegie-Mellon
University), vol. 2, pp. 1014-1029.
Feigenbaum, Edward A., and Feldman, Julian, eds., Computers and Thought
(New York: McGraw-Hill, 1963).
Feigenbaum, Mitchell J., “Universal Behavior in Nonlinear Systems,” in Los
Alamos Science, vol. 1 (summer 1980), pp. 4-27.
CUMULATIVE LIST OF REFERENCES 689

Feinstein, Amiel, Foundations of Information Theory (New York: McGraw-Hill,

1958).
Feldman, Jerome A., and Sutherland, William R., eds., “Rejuvenating Experi¬
mental Computer Science—A Report to the National Science Foundation and
Others,” in Communications of the ACM, vol. 22 (Sept. 1979), pp. 497-502.
Feldman, Julian, “Simulation of Behavior in Binary Choice Experiment,” in Ex¬
tending Man’s Intellect, proceedings of the Western Joint Computer Confer¬
ence, held [in Los Angeles] by the Institute of Radio Engineers, the American
Institute of Electrical Engineers, and the Association for Computing Machinery,
May 9-11, 1961, vol. 19, pp. 133-144.
Feltham, Gerald A., “The Value of Information,” in The Accounting Review, vol.
43 (Oct. 1968), pp. 684-696.
Feltham, Gerald A., Information Evaluation (Sarasota, Florida: American Ac¬
counting Association, 1972).
Feltham, Gerald A., “Financial Accounting Research: Contributions of Informa¬
tion Economics and Agency Theory,” in Mattessich, Richard, et al., Modern
Accounting Research: History, Survey, and Guide (Vancouver, B.C.: Canadian
Certified General Accountants Research Foundation, forthcoming 1984).
Fiengo, Robert, “Semantic Conditions on Surface Structure” (Ph.D. diss., Massa¬
chusetts Institute of Technology, 1974).
Fisher, Ronald A., “Probability Likelihood and Quantity of Information in the
Logic of Uncertain Inference,” in Proceedings of the Royal Society, series A,
vol. 146 (Aug. 1, 1934), pp. 1-8.
Fisher, Ronald A., The Design of Experiments (London: Oliver and Boyd, 1935).
Flexner, Abraham, “Is Social Work a Profession?” in School and Society, vol. 1
(June 1915), pp. 902-906.
Flood, Merrill M., “The Systems Approach to Library Planning,” in Intellectual
Foundations of Library Education, proceedings of the Twenty-Ninth Annual
Conference of the Graduate Library School (Chicago: University of Chicago
Press, 1965), pp. 38-50.
Fodor, Jerry A., “Methodological Solipsism Considered as a Research Strategy in
Cognitive Psychology,” in The Behavioral and Brain Sciences, vol. 3 (Mar.
1980), pp. 63-109.
Fodor, Jerry A., Representations: Philosophical Essays on the Foundations of Cog¬
nitive Science (Cambridge, Mass.: Harvard University Press, 1981).
Fodor, JerryA.; Bever, Thomas G.; and Garrett, Merrill F., The Psychology of
Language: An Introduction to Psycholinguistics and Generative Grammar
(New York: McGraw-Hill, 1974).
Fodor, Jerry A., and Garrett, Merrill F., “Some Syntactic Determinants of Sen-
tencial Complexity,” in Perception and Psychophysics, vol. 2 (1967), pp. 289-
296.
Forrester, Jay W., Industrial Dynamics (Cambridge, Mass.: MIT Press, 1962).
Forrester, Jay W., World Dynamics (Cambridge, Mass.: Wright-Alien, 1971).
Forte, B., and Kampe de Feriet, J., “Information et probability,” in C. R. Acad.
Sci. [Comptes Rendus de l’Academie de Sciences (Paris)], series A, vol. 265
(1967), pp. 110-114, 142-146, 350-353.
690 CUMULATIVE LIST OF REFERENCES

Frank, Philipp, Philosophy of Science: The Link between Science and Philosophy
(Englewood Cliffs, N.J.: Prentice-Hall, 1957).
Frederiks, J. A. M., “Disorders of the Body Schema,” in Vinken, P. J., and Bruyn,
G. W., eds., Handbook of Clinical Neurology, Disorders of Speech, Perception,
and Symbolic Behavior (Amsterdam: North-Holland, 1969-), vol. 4, pp. 207-
240.
Frege, Friedrich Ludwig Gottlob, Begriffsschrift, eine der arithmetischen nach-
gebildete Formelsprache des reinen Denkens (Halle A/S, Germany: L. Nebert,
1879).
Freidin, Robert A., “Cyclicity and the Theory of Grammar,” in Linguistic Inquiry,
vol. 9 (1978), pp. 519-550.
Fried, Charles, Right and Wrong (Cambridge, Mass.: Harvard University Press,
1978) .
Fromkin, Victoria A., et al., Speech Errors as Linguistic Evidence (The Hague,
Netherlands: Mouton, 1973).
Fuhrman, P., “Functional Models, Factorizations, and Linear Systems”—talks
given at the NATO-AMS Advanced Study Institute on Algebraic and Geometric
Methods in Linear System Theory (Cambridge, Mass.: Harvard University,
1979) .
Furth, Hans G., “The Operative and Figurative Aspects of Knowledge and Infor¬
mation,” in Debons, Anthony, ed., Information Science: Search for Identity,
proceedings of the 1972 NATO Advanced Study Institute in Information Sci¬
ence, held [in Seven Springs, Champion, Penn.] Aug. 12-20, 1972 (New York:
Marcel Dekker, 1974), pp. 21-27.
Gabor, Dennis, “Theory of Communication,” in Journal of the Institution of Electri¬
cal Engineers, vol. 93, no. 3 (1946), pp. 429-457.
Gallagher, Robert G., Information Theory and Reliable Communication (New
York: Wiley, 1968).
Galvin, Thomas J., “The Challenge of Technology in the Eighties,” lecture pre¬
sented at Ann Arbor, University of Michigan, “Issues and Challenges of the
Eighties: A Michigan Libraries Forum,” 1981.
Gardner, Beatrice T., and Gardner, R. Allen, “Two-Way Communication with
an Infant Chimpanzee,” in Schrier, Allan M., and Stollnitz, Fred, edsBehavior
of Non-Human Primates: Modern Research Trends, vol. 4 (New York: Aca¬
demic Press, 1971), pp. 117-184.
Garfield, Eugene; Malin, Morton V.; and Small, Henry, “Citation Data as Sci¬
ence Indicators,” in Elkana, Yehuda, et al., eds., Toward a Metric of Science:
The Advent of Science Indicators (New York: Wiley, 1978).
Garner, Wendell R., Uncertainty and Structure as Psychological Concepts (New
York: Wiley, 1962).
Gatlin, Lila L., Information Theory and the Living System (New York: Columbia
University Press, 1972).
Gauld, Alan, and Shotter, John, Human Action and Its Psychological Investiga¬
tion (London and Boston: Routledge and Kegan Paul, 1977).
Gazis, Denos C.; Herman, Robert; and Potts, Renfrey B., “Car Following Theory
of Steady-State Traffic Flow,” in Operations Research, vol. 7 (July/Aug. 1959),
pp. 499-505.
CUMULATIVE LIST OF REFERENCES 691

Gazzaniga, Michael S., The Bisected Brain (New York: Appleton-Century-Crofts

1970).
Gazzaniga, Michael S., “1981 Nobel Prize for Physiology or Medicine,” in Sci¬
ence, vol. 214 (30 Oct. 1981), pp. 517-518.
Gel’fand, Izrail’ Moiseevich; Gurfinkel’, V. A.; Tsetlin, M. L.; and Shik, M. L.,
eds., Models of the Structural-Functional Organization of Certain Biological
Systems, trans. Carol R. Beard (Cambridge, Mass.: MIT Press, 1971).
Gergely, Tamas, “On the Unity Between Cybernetics and General System The¬
ory,” in Kybernetes, vol. 8 (1979), pp. 45-49.
Gibbs, Josiah Willard, Elementary Principles in Statistical Mechanics: Developed
with Especial Reference to the Rational Foundation of Thermodynamics (New
York: Scribners, 1902; 2nd ed. New York: Dover Publications, 1960).
Gibbs, Josiah Willard, The Collected Works of J. Willard Gibbs, vol. 2, chap. 2,
theorem 5 (New Haven, Conn.: Yale University Press, 1948). Original publica¬
tion 1906, under the title The Scientific Papers of J. Willard Gibbs, pp. 20-25.
Gigley, Helen, A Computational Neurolinguistic Approach to Processing Models
of Sentence Comprehension: COINS Technical Report 82-9 (Amherst: Univer¬
sity of Massachusetts at Amherst, Dept, of Computer and Information Science,
1982).
Givol, David, “A Structural Basis for Molecular Recognition: The Antibody Case,”
in Cuatrecasas, Pedro, and Greaves, Melvyn F., eds., Receptors and Recogni¬
tion, vol. 2 (London: Chapman and Hall, 1976), series A, pp. 1-42.
Glass, Arnold L.; Holyoak, Keith J.; and Santa, John L., Cognition (Reading,
Mass.: Addison-Wesley, 1979).
Glushkov, Viktor Mikhailovich, Introduction to Cybernetics, trans. by Scripta
Technica Inc., (New York and London: Academic Press, 1966).
Goguen, Jack, “L-Fuzzy Sets,” Journal of Mathematical Analysis and Applica¬
tions, vol. 18 (Apr. 1967), pp. 145-174.
Goldstein, Ira, and Papert, Seymour, “Artificial Intelligence, Language, and the
Study of Knowledge,” in Cognitive Science, vol. 1 (Jan. 1977), pp. 84-124.
Goldstein, Kurt, Language and Language Disturbances: Aphasic Symptom Com¬
plexes and Their Significance for Medicine and Theory of Language (New York:
Grune and Stratton, 1948).
Good, Irving J., “Weighing Evidence,” in Good, Irving J., Probability and the
Weighing of Evidence (London: Charles Griffin and Co.; New York: Hafner,
1950), chap. 6.
Goodwin, Brian C., Temporal Organization in Cells; A Dynamic Theory of Cellular
Control Processes (London and New York: Academic Press, 1963).
Gorn, Saul, “The Computer and Information Sciences: A New Basic Discipline,”
in SIAM Review, vol. 5 (Apr. 1963), pp. 150-155.
Gorn, Saul, “The Individual and Political Life of Information Systems,” in Heil-
prin, Laurence B.; Markuson, Barbara E.; and Goodman, Frederick L., eds.,
Education for Information Science, proceedings of a symposium held at
Airlie House, Virginia, 1965 (Washington, D.C.: Spartan Books, 1965), pp.
33-40.
Gorn, Saul, “The Computer and Information Sciences in the Community of Disci¬
plines,” in Behavioral Science, vol. 12 (Nov. 1967), pp. 433-452.
692 CUMULATIVE LIST OF REFERENCES

Gorn, Saul, “The Identification of the Computer and Information Sciences: Their
Fundamental Semiotic Concepts and Relationships,” in Foundations of Lan¬
guage: International Journal of Language and Philosophy, vol. 4 (Nov. 1968),
pp. 339-372.
Greaves, Melvyn F., “Cell Surface Receptors: A Biological Perspective,” in Cuat-
recasas, Pedro, and Greaves, Melvyn F., eds., Receptors and Recognition, vol.
1 (London: Chapman and Hall, 1976), series A, pp. 1-32.
Gregory, Richard L., “On How So Little Information Controls So Much Behav¬
ior,” in Waddington, Conrad H., ed., Towards a Theoretical Biology, Vol. 2:
Sketches, proceedings of an IUBS Symposium (Chicago: Aldine-Atherton,
1969), pp. 236-247.
Greibach, Sheila A., “Formal Languages: Origins and Directions,” in Symposium
on the Foundations of Computer Science, proceedings of the Twentieth Annual
Symposium on Foundations of Computer Science, held [in San Juan, Puerto
Rico] by the IEEE Computer Society’s Technical Committee on Mathematical
Foundations of Computing (New York: Institute of Electrical and Electronics
Engineers, 1979), pp. 66-90.
Grenander, Ulf, Lectures in Pattern Theory, Vol. 1: Pattern Synthesis (New York:
Springer-Verlag, 1976).
Grenander, Ulf, Lectures in Pattern Theory, Vol. 2: Pattern Analysis (New York:
Springer-Verlag, 1978).
Groot, Adriaan D. de, Thought and Choice in Chess (The Hague, Netherlands:
Mouton, 1965).
Grossberg, Stephen, “Adaptive Pattern Classification and Universal Recoding: I.
Parallel Development and Coding of Neural Feature Detectors,” in Biological
Cybernetics, vol. 23, no. 2 (1976), pp. 121-134.
Gunderson, Keith, Mentality and Machines (Garden City, N.Y.: Doubleday, 1971).
Habermas, Jurgen, Toward a Rational Society: Student Protest, Science, and Poli¬
tics, trans. Jeremy J. Shapiro (Boston: Beacon Press, 1970).
Haberstroh, Chadwick J., “Control as an Organizational Process,” in Management
Science, vol. 6 (Jan. 1960), pp. 165-171.
Haken, Hermann, Synergetics: An Introduction: Nonequilibrium Phase Transitions
and Self-Organization in Physics, Chemistry, and Biology (Berlin and New
York: Springer-Verlag, 1977; 2nd ed., 1978).
Haken, Hermann, “Synergetics: Are Cooperative Phenomena Governed by Uni¬
versal Principles?” in Naturwissenschaften, vol. 67 (Mar. 1980), pp. 121-128.
Hall, Thomas Steele, Ideas of Life and Matter: Studies in the History of General
Physiology 600BC-1900AD, Vol. I: From Pre-Socratic Times to the Enlighten¬
ment (Chicago: University of Chicago Press, 1969).
Hanson, Allen R., and Riseman, Edward M., eds., Computer Vision Systems,
proceedings of a Workshop on Computer Vision Systems, held [at the Univer¬
sity of Massachusetts, Amherst] 1977 (New York: Academic Press, 1978).
Hardin, Garrett, Nature and Man’s Fate (New York: Mentor, 1961).
Hardin, Garrett, “The Tragedy of the Commons,” in Science, vol. 162 (1968), p.
1243.
Hardin, Garrett, Exploring New Ethics for Survival (New York: Viking Press,
1977).
CUMULATIVE LIST OF REFERENCES 693

Harris, Zellig S., Methods in Structural Linguistics (Chicago: University of

Chicago Press, 1951).
Harris, ZelligS. “Co-occurrence of Transformation in Linguistic Structure,” in
Language, vol. 33 (1957), pp. 283-340.
Harter, Stephen P., “‘Information’ in Every Name,” in Bulletin of the American
Society for Information Science, vol. 9 (Oct. 1982), pp. 40-41.
Hartley, Ralph V. L., “Transmission of Information,” in Bell System Technical
Journal, vol. 7 (July 1928), pp. 535-563.
Hartmanis, Juris, “Observations about the Development of Theoretical Computer
Science,” in Symposium on the Foundations of Computer Science, proceedings
of the Twentieth Annual Symposium on Foundations of Computer Science, held
[in San Juan, Puerto Rico] by the IEEE Computer Society’s Technical Commit¬
tee on Mathematical Foundations of Computing (New York: Institute of Electri¬
cal and Electronics Engineers, 1979), pp. 224-233.
Hartmanis, Juris, “Quo Vadimus: Computer Science in a Decade,” in Communica¬
tions of the ACM, vol. 24 (June 1981), pp. 351-369.
Haugeland, John, “The Nature and Plausibility of Cognitivism,” in The Behavioral
and Brain Sciences, vol. 1 (June 1978), pp. 215-260.
Hayek, Friedrich A., “The Counter-Revolution of Science,” in Economica, N.S.
vol. 8 (1941). Reprinted in The Counter-Revolution of Science (Glencoe, Ill.: The
Free Press, 1952; also Indianapolis: Liberty Press, 1979).
Hayek, Friedrich A., “Notes on the Evolution of Systems of Rules of Conduct,” in
Studies in Philosophy, Politics, and Economics (Chicago: University of Chicago
Press, 1967).
Hayek, Friedrich A., Law, Legislation, and Liberty, Vol. I: Rules and Order
(Chicago: University of Chicago Press, 1973).
Hayes, Patrick J., “Some Comments on Sir James Lighthill’s Report on Artificial
Intelligence,” in AISB Study Group European Newsletter, Issue 14 (July 1973),
pp. 36-53.
Hayes, Patrick J., “The Naive Physics Manifesto,” in Michie, Donald, ed.. Expert
Systems in a Microelectronic Age (Edinburgh: Edinburgh University Press,
1979), pp. 463-502.
Hayes, Robert M., “Information Science in Librarianship,” in Libri, vol. 19, no. 3
(1969), pp. 216-236.
Hazewinkel, M., “Moduli and Canonical Forms for Linear Dynamical Systems II:
The Topological Case,” in Mathematical Systems Theory, vol. 10 (1977), pp.
363-385.
Hazewinkel, M., and Kalman, Rudolf E., “On Invariants, Canonical Forms and
Moduli for Linear, Constant, Finite Dimensional, Dynamical Systems,” in Pro¬
ceedings of the CNR-CISM Symposium on Algebraic System Theory, held [at
Udine] 1975 (Heidelberg, Germany: Springer-Verlag, 1976), pp. 48-60.
Heilprin, Laurence B., “Operational Definitions,” in Debons, Anthony, ed., Infor¬
mation Science: Search for Identity, proceedings of the 1972 NATO Advanced
Study Institute in Information Science, held [in Seven Springs, Champion,
Penn.] Aug. 12-20, 1972 (New York: Marcel Dekker, 1974), pp. 115-138.
Held, Richard; Leibowitz, Herschel W.; and Teuber, Hans-Lukas, eds., Hand-
694 CUMULATIVE LIST OF REFERENCES

book of Sensory Physiology, Vol. VIII: Perception (Berlin, Heidelberg, and New
York: Springer-Verlag, 1978).
Hempel, Carl G., “Fundamentals of Concept Formation in Empirical Science,” in
Neurath, Otto; Carnap, Rudolf; and Morris, Charles, eds., Foundations of the
Unity of Science: Toward an International Encyclopedia of Unified Science,
vol. 2 (Chicago: University of Chicago Press, 1939; bound ed., 1969), pp. 651 —
746.
Hempel, Carl G., Philosophy of Natural Science (Chicago: Aldine-Atherton, 1963).
Herman, E. Edward, and Kuhn, Alfred, Collective Bargaining and Labor Rela¬
tions (Englewood Cliffs, N.J.: Prentice-Hall, 1981).
Herman, Robert, and Montroll, Elliott W., “Some Statistical Observations from
a 75 Year Run of Sears Roebuck Catalogues,” in Landman, Uzi, ed., Statistical
Mechanics and Statistical Methods in Theory and Application: A Tribute to
Elliott W. Montroll, proceedings of a Symposium on Statistical Mechanics and
Statistical Methods in Theory and Research, held [at the University of Roches¬
ter, N.Y.] Nov. 4, 1976 (New York: Plenum Press, 1977), pp. 785-803.
Herman, Robert;Montroll, Elliott W.; Potts, Renfrey B.; and Rothery,
Richard W., “Traffic Dynamics: Analysis of Stability in Car Following,” in
Operations Research, vol. 7 (Jan./Feb. 1959), pp. 86-106.
Hermann, Robert T., and Martin, Clyde F., “Applications of Algebraic Geometry
to System Theory, Part I,” in IEEE Transactions on Automatic Control, vol.
AC-22 (Feb. 1977), pp. 19-25.
Hesse, Hermann, Magister Ludi (Das Glasperlenspeit) [the Nobel Prize novel],
trans. Mervyn Savill (New York: Holt, 1949).
Hick, W. Edmund, “On the Rate of Gain of Information,” in Quarterly Journal of
Experimental Psychology, vol. 4 (Feb. 1952), pp. 11-26.
Hilgard, Ernest R., “Pain Perception in Man,” in Held, Richard; Leibowitz,
Herschel W.; and Teuber, Hans-Lukas, eds., Handbook of Sensory Physiology,
Vol. VIII: Perception (Berlin, Heidelberg, and New York: Springer-Verlag,
1978), pp. 849-875.
Hilgard, Ernest R., and Bower, Gordon H., Theories of Learning (New York:
Appleton-Century-Crofts, 1948; 4th ed., 1975).
Hillman, Donald J., “Knowledge Transfer Systems,” in Weiss, Edward C., ed.,
The Many Faces of Information Science (Boulder, Colo.: Westview Press,
1977), pp. 75-103.
Hillman, James, The Myth of Analysis: Three Essays in Archetypal Psychology
(Evanston, Ill.: Northwestern University Press, 1972).
Hilton, Ronald W., “The Determinants of Information Value: Synthesizing Some
General Results,” in Management Science, vol. 27 (Jan. 1981), pp. 57-64.
Hinton, G. E., “Shape Representation in Parallel Systems,” in Proceedings of the
Seventh International Joint Conference on Artificial Intelligence [Vancouver,
B.C., 1981], pp. 1088-1096.
Hirshleifer, Jack, and Riley, JohnG., “The Analytics of Uncertainty and Infor¬
mation—An Expository Survey,” in Journal of Economic Literature, vol. 17
(Dec. 1979), pp. 1375-1421.
Hobbes, Thomas, Leviathan, or, The Matter, Form and Power of a Commonwealth
Ecclesiastical and Civil (London: Andrew Cooke, 1651).
CUMULATIVE LIST OF REFERENCES 695

Hofstadter, Douglas R., Godel, Escher, Bach: an Eternal Golden Braid (New
York: Basic Books, 1979).
Hofstadter, Douglas R., “Metamagical Themas: How Might Analogy, the Core of
Human Thinking, be Understood by Computers?,” in Scientific American,
Sept. 1981, p. 18.
Hofstadter, Douglas R., Who Shoves Whom Around Inside the Careenium?
(Bloomington: Indiana University Computer Science Department Technical Re¬
port No. 130, July 1982a; forthcoming in Synthese).
Hofstadter, Douglas R., ‘‘Metamagical Themas: Can Inspiration Be Mechanized?”
in Scientific American, Sept. 1982ft, p. 16.
Hofstadter, Douglas R., “Metamagical Themas: On Variations on a Theme as the
Essence of Imagination,” in Scientific American, Oct. 1982c, p. 20.
Hofstadter, Douglas R., The Tumult of Inner Voices (forthcoming, Southern Utah
State College, 1982d).
Hofstadter, Douglas R.; Clossman, Gray A.; and Meredith, Marsha J., Shake¬
speare’s Plays Weren’t Written by Him, But by Someone Else of the Same
Name: An Essay on Intensionality and Frame-Based Knowledge Representa¬
tion Systems (Bloomington: Indiana University Computer Science Department
Technical Report No. 96, July 1980).
Hofstadter, Douglas R., and Dennett, Daniel C., eds., The Mind’s I: Fantasies
and Reflections of Self and Soul (New York: Basic Books, 1981).
Holstrom, Bengt R., “Moral Hazard and Observability,” in Bell Journal of Eco¬
nomics, vol. 10 (spring 1979), pp. 74-91.
Holzner, Burkart, and Marx, John H., Knowledge Application: The Knowledge
System in Society (Boston: Allyn and Bacon, 1979).
Homans, George C., The Nature of Social Change (New York: Harcourt, Brace, and
World, 1967).
Hoos, Ida R., Systems Analysis in Public Policy: A Critique (Berkeley: University of
California Press, 1972).
Hopcroft, John E., and Ullman, Jeffrey D., Introduction to Automata Theory,
Languages, and Computation (Reading, Mass.: Addison-Wesley, 1979).
Hopf, Eberhard, “Abzweigung einer periodischen Losung von einer Stationaren
Losung eines Differentzialsystems,” in Berichte iiber die Verhandlungen Sach-
sischen Akademie der Wissenschaften zu Leipzig, Mathematisch-Naturwissen-
schaftliche Klasse, vol. 95 (1943), pp. 3-22.
Howard, Ronald A., “Information Value Theory,” in IEEE Transactions on Sys¬
tems Science and Cybernetics, vol. SSC-2 (Aug. 1966), pp. 22-26.
Hume, David, Dialogues Concerning Natural Religion (1779), in Green, T. H., and
Grose, T. H., eds., A Treatise on Human Nature (London: Longmans, Green,
and Coy., new ed., 1878, vol. II), pp. 380-468; reprinted in The Empiricists
(Garden City, N.Y.: Doubleday, 1961).
I Ching, or Book of Changes, trans. Richard Wilhelm and Cary F. Baynes (Prince¬
ton, N.J.: Princeton University Press, Bollingen Series 19, 1951).
IEEE Transactions on Information Theory (New York: Institute of Electrical and
Electronics Engineers, 1954, 1956).
Ingarden,R., and Urbanik, K., “Information Without Probability,” in Colloquium
Mathematicum, vol. 9 (1962), pp. 131-149.
696 CUMULATIVE LIST OF REFERENCES

Institute of Radio Engineers, IRE Transactions on Information Theory, vol. IT-1

(New York: IRE Professional Group on Information Theory, Sept., 1954).
Institute of Radio Engineers, IRE Transactions on Information Theory, vol. IT-2
(New York: IRE Professional Group on Information Theory, Sept. 1956).
Itami, Hiroyuki, Adaptive Behavior: Management Control and Information Analy¬
sis (Sarasota, Florida: American Accounting Association, 1977).
Jackendoff, Ray S., Semantic Interpretation in Generative Grammar (Cambridge,
Mass.: MIT Press, 1972).
Jackson of Burnley, Willis Jackson, Baron, ed., Proceedings of Symposium on
Information Theory, held [at the Royal Society, London] 1950 (London: Minis¬
try of Supply, 1950).
Jackson of Burnley, Willis Jackson, Baron, ed., Communication Theory, papers
read at a symposium on “Applications of Communication Theory,” held [at the
Institution of Electrical Engineers, London] Sept. 22-26, 1952 (London: Butter-
worths, 1953).
Jacob, Francois, La logique du vivant: Une histoire de Theredite (Paris: Gallimard,
1971), trans. by Betty E. Spillmann as The Logic of Living Systems (London:
Allen Lane, 1974).
Jakobson, Roman, Selected Writings, vol. 1: Phonological Studies (The Hague,
Netherlands: Mouton, 1962).
Jakobson, Roman, Selected Writings, vol. 2: Word and Language (The Hague,
Netherlands: Mouton, 1971).
Jamison, Dean, “Bayesian Information Usage,” in Hintikka, Jaako, and Suppes,
Patrick, eds., Information and Inference (Dordrecht, Holland: Reidel, 1970), pp.
28-57.
Jantsch, Erich, The Self-Organizing Universe; Scientific and Human Implications
of the Emerging Paradigm of Evolution (Oxford, England: Pergamon Press,
1980).
Jaspers, Karl, The Future of Mankind, trans. E. B. Ashton (Chicago: Chicago
University Press, 1961).
Jaynes, Edwin T., “Information Theory and Statistical Mechanics,” in The Physical
Review, vol. 106, 2nd series (May 1957a), pp. 620-630.
Jaynes, Edwin T., “Information Theory and Statistical Mechanics,” in The Physical
Review, vol. 108, 2nd series (Oct. 19576), pp. 171-190.
Jaynes, Edwin T., “Note on Unique Decipherability,” in IRE Transactions on Infor¬
mation Theory, vol. IT-5 (Sept. 1959), pp. 98-102.
Jaynes, Edwin T., “New Engineering Applications of Information Theory,” in Bog-
danoff, John L., and Kozin, Frank, eds., Proceedings of the First Symposium on
Engineering Applications of Random Function Theory and Probability (New
York: Wiley, 1963), pp. 163-203.
Jaynes, Edwin T., and Scalapino, D., “Non-Local Transport Theory,” a paper
presented at the first meeting of the Society for Natural Philosophy [Baltimore],
Mar. 25, 1963.
Jensen, Michael C., and Meckling, William H., “Theory of the Firm: Managerial
Behavior, Agency Cost and Ownership Structure,” in Journal of Financial Eco¬
nomics, vol. 3 (Oct. 1976), pp. 305-360.
CUMULATIVE LIST OF REFERENCES 697

Jewett, Charles C., “A Plan for Stereotyping Catalogues by Separate Titles; and for
Forming a General Stereotyped Catalogue of Public Libraries in the United
States,” in Proceedings of the American Association for the Advancement of
Science, vol. 4 (1851), pp. 165-176.
Johns, Ada Winifred, Special Libraries: Development of the Concept, their Organi¬
zations, and their Services (Metuchen, N.J.: Scarecrow Press, 1968).
Josiah Macy Jr. Foundation, Cybernetics, proceedings of the Sixth Conference
(1949) to the Tenth Conference (1953) (New York: Josiah Macy Jr. Foundation,
last pub. 1955).
Julesz, Bela, Foundation of Cyclopean Perception (Chicago: University of Chicago
Press, 1971).
Kalman, Rudolf E., “A New Approach to Linear Filtering and Prediction Prob¬
lems,” in Transactions of the American Society of Mechanical Engineers, Jour¬
nal of Basic Engineering, series D, vol. 82 D (Mar. 1960a), pp. 35-45.
Kalman, Rudolf E., “On the General Theory of Control Systems,” in Proceedings
of the First IFAC Congress, held [in Moscow] 1960 (London: Butterworths,
19606), vol. 1, pp. 481-491.
Kalman, Rudolf E., “Canonical Structure of Linear Dynamical Systems,” in Pro¬
ceedings of the National Academy of Sciences (USA), vol. 48 (1962), pp. 596-
600.
Kalman, Rudolf E., “Linear Stochastic Filtering Theory—Reappraisal and Out¬
look,” in Fox, J., ed., Proceedings of a Symposium on System Theory (Brook¬
lyn, N.Y.: Polytechnic Institute of Brooklyn, 1965), pp. 197-205.
Kalman, Rudolf E., “On Minimal Partial Realizations of a Linear Input/Output
Map,” in Kalman, Rudolf E., and De Claris, N., eds., Aspects of Network and
System Theory (New York: Holt, Rinehart, and Winston, 1971), pp. 385-408.
Kalman, Rudolf E., “Kronecker Invariants and Feedback,” in Weiss, L., ed.,
Proceedings of the 1971 NRL-MRC Conference on Ordinary Differential Equa¬
tions (New York: Academic Press, 1972), pp. 459-471.
Kalman, Rudolf E., “Comments on the Scientific Aspects of Modeling,” in Marois,
M., ed., Towards a Plan of Action for Mankind (Amsterdam: North-Holland,
1974), pp. 492-505.
Kalman, Rudolf E., “Realization Theory of Linear Dynamical Systems,” in Con¬
trol Theory and Functional Analysis (Vienna: International Atomic Energy
Agency, 1976).
Kalman, Rudolf E., “On Partial Realizations, Transfer Functions, and Canonical
Forms,” in Acta Polytechnica Scandinavica, Mathematics and Computer Sci¬
ence Series no. 31 (1979), pp. 9-32.
Kalman, Rudolf E., “A System-Theoretic Critique of Dynamic Economic Mod¬
els,” in International Journal of Policy Analysis and Information Systems, vol.
4 (1980), pp. 3-22.
Kalman, Rudolf E., “Identifiability and Problems of Model Selection in Econo¬
metrics”—invited paper presented at the Fourth World Congress of the Econ¬
ometric Society, held [in Aix-en-Provence, France] 1980 (rev. version, Jan.
1981).
Kalman, Rudolf E., “Finitary Models for Time Series and the Identification of
698 CUMULATIVE LIST OF REFERENCES

Dynamics,” in Krishnaiah, Paruchuri R., ed., Developments in Statistics (New

York: Academic Press, 1982).
Kalman, Rudolf E., and Bucy, Richard S., “New Results in Linear Prediction and
Filtering Theory,” in Transactions of the American Society of Mechanical En¬
gineers, Journal of Basic Engineering, series D, vol. 83D (Mar. 1961), pp. 95-
108.
Kalman, Rudolf E.; Falb, Peter L.; and Arbib, Michael A., Topics in Mathemat¬
ical System Theory (New York: McGraw-Hill, 1969).
Kamien, Ralph, and Schwartz, Nancy, Dynamic Optimization: The Calculus of
Variations and Optimal Control in Economics and Management (New York:
Elsevier, 1981).
Kaneko, Tatsuo; Atlas, Ronald M.; and Krichevsky, Micah, “Diversity of Bacte¬
rial Populations in Beaufort Sea,” in Nature, vol. 270 (Dec. 1977), pp. 596-599.
Kant, Immanuel, Foundations of the Metaphysic of Morals (London: 1st ed. in
English, 1799; another ed., Indianapolis: Bobbs-Merrill, 1959).
Kant, Immanuel, Critick of Pure Reason (London: William Pickering, 1848).
Kaplan, Abraham, “The Age of the Symbol—A Philosophy of Library Education,”
in The Library Quarterly, vol. 34 (Oct. 1964), pp. 295-304.
Karetzky, Stephen, Reading Research in Librarianship: A History and Analysis
(Westport, Conn.: Greenwood Press, 1982).
Karmiloff-Smith, Annette, “Micro- and Macrodevelopmental Changes in Lan¬
guage Acquisition and Other Representational Systems,” in Cognitive Science,
vol. 3 (Apr./June 1979), pp. 91-117.
Karmiloff-Smith, Annette, and Inhelder, Barbel, “If You Want to Get Ahead,
Get a Theory,” in Cognition, vol. 3, no. 3 (1975), pp. 195-212.
Katz, Jerrold J., Underlying Reality of Language and Its Philosophical Import
(New York: Harper and Row, 1971).
Katz, Jerrold J., and Postal, Paul M., An Integrated Theory of Linguistic De¬
scriptions (Cambridge, Mass.: MIT Press, 1964).
Kendall, D. G., “Entropy, Probability, and Information,” in International Statisti¬
cal Review, vol. 41 (1973), pp. 59-68.
Kent, Allen, “Field Definition—Information Science,” in Section 7: “Computer
and Information Sciences,” Peterson’s Guides to Graduate Study, Book 5,
1982.
Keyser, Samuel J., and O’Neil, Wayne, “The Evolution of the English Plural,” in
Linguistic Research, vol. 1, no. 2 (1980).
Khinchin, Aleksandr I., Mathematical Foundations of Information Theory, trans.
R. A. Silverman and M. D. Friedman of two pieces: “The Entropy Concept in
Probability Theory” (1953) and “On the Fundamental Theorems of Information
Theory” (1956) (New York: Dover, 1957).
Kilmer, William L.; McCulloch, Warren S.; and Blum, Jay, “A Model of the
Vertebrate Central Command System,” in International Journal of Man-
Machine Studies, vol. 1 (July 1969), pp. 279-309.
King, Donald W., “Thoughts for ASIS Members,” in Bulletin of the American
Society for Information Science, vol. 6 (Aug. 1980), pp. 3-4.
CUMULATIVE LIST OF REFERENCES 699

King, Donald W.; Krauser, Cheri; and Sague, Virginia M., “Profile of ASIS
Membership,” in Bulletin of the American Society for Information Science, vol.
6 (Aug. 1980), pp. 9-17.
King, Gilbert W., et al., Automation and the Library of Congress (Washington,
D.C.: Library of Congress, 1963).
Kirzner, Israel M., Competition and Entrepreneurship (Chicago: University of
Chicago Press, 1973).
Kleene, Stephen C., “Origins of Recursive Function Theory,” in Symposium on the
Foundations of Computer Science, proceedings of the Twentieth Annual Sym¬
posium on Foundations of Computer Science, held [in San Juan, Puerto Rico] by
the IEEE Computer Society’s Technical Committee on Mathematical Founda¬
tions of Computing (New York: Institute of Electrical and Electronics En¬
gineers, 1979), pp. 371-381.
Klima, Edward S., and Bellugi, Ursula, The Signs of Language (Cambridge,
Mass.: Harvard University Press, 1979).
Kling, Rob, “Social Analyses of Computing: Theoretical Perspectives in Recent
Empirical Research,” in Computing Surveys, vol. 12 (Mar. 1980), pp. 61-110.
Klir, George Jiri, An Approach to General System Theory (New York: Van Nos¬
trand Reinhold, 1969).
Klir, George Jiri, “0n the Relation Between Cybernetics and General Systems
Theory,” in Rose, John, ed., Progress of Cybernetics, proceedings of the First
International Congress of Cybernetics, held [in London] 1969 (London and New
York: Gordon and Breach Science Publishers, 1970), vol. 1, pp. 155-165.
Knuth, Donald E., The Art of Computer Programming, Vol. 1: Fundamental Al¬
gorithms (Reading, Mass.: Addison-Wesley, 1968).
Knuth, Donald E., The Art of Computer Programming (Reading, Mass.: Addison-
Wesley, 1973-1981), 3 vols.
Knuth, Donald E., “Computer Programming as an Art” [the 1974 ACM Turing
Award Lecture], in Communications of the ACM, vol. 17 (Dec. 1974a), pp. 667-
673.
Knuth, Donald E., “Computer Science and Its Relation to Mathematics,” in Amer¬
ican Mathematical Monthly, vol. 81, no. 4 (1974b), pp. 323-343.
Kochen, Manfred, “Extension of Moore-Shannon Model for Relay Circuits,” in
IBM Journal of Research and Development, vol. 3 (Apr. 1959), pp. 169-186.
Kochen, Manfred, Some Problems in Information Science (Metuchen, N.J.: Scare¬
crow Press, 1965).
Kochen, Manfred, ed., The Growth of Knowledge: Readings on Organization and
Retrieval of Information (New York: Wiley, 1967).
Kochen, Manfred, “Stability in the Growth of Knowledge, in Saracevic, Tefko,
ed., Introduction to Information Science (New York and London: Bowker,
1970), pp. 48-55.
Kochen, Manfred, “WISE: A World Information Synthesis and Encyclopaedia,” in
Journal of Documentation, vol. 28 (Dec. 1972), pp. 322—343.
Kochen, Manfred, “Views on the Foundations of Information Science, in Debons,
Anthony, ed., Information Science: Search for Identity, proceedings of the 1972
700 CUMULATIVE LIST OF REFERENCES

NATO Advanced Study Institute in Information Science, held [in Seven

Springs, Champion, Penn.] Aug. 12-20, 1972 (New York: Marcel Dekker, 1974),
pp. 171-187.
Kochen, Manfred, ed., Information for Action, proceedings of a symposium orga¬
nized by the American Association for the Advancement of Science (New York:
Academic Press, 1975).
Kochen, Manfred, “Paper as a Medium of Communication: Trends in Alternate
Technologies,” in International Forum on Information and Documentation
(forthcoming 1982).
Kochen, Manfred, and Deutsch, Karl W., Decentralization: Sketches Towards a
Rational Theory (Cambridge, Mass.: Ogilvie-Hill Press, 1980).
Koffka, Kurt, Principles of Gestalt Psychology (New York: Harcourt, 1935).
Kohler, Wolfgang, Gestalt Psychology: An Introduction to New Concepts in Mod¬
ern Psychology (New York: Liveright, 1929; rev. ed., 1947; paperback ed., New
York: New American Library, 1947, 1961).
Kolmogorov, A. N., “Three Approaches to the Quantitative Definition of Informa¬
tion,” in Problems of Information Transmission, vol. 1 (Jan./Mar., 1965), pp. 1-
7. English trans. of the Russian Problemy Peredachi Informatsii.
Kornfeld, W. A., and Hewitt, Carl E., The Scientific Community Metaphor
(Artificial Intelligence Laboratory, M.I.T., Report No. AIM-641, 1981).
Kosslyn, Stephen M.; Pinker, Steven; Smith, George E.; and Shwartz, Steven
P., “On the Demystification of Mental Imagery” [with peer commentaries], in
The Behavioral and Brain Sciences, vol. 2 (Dec. 1979), pp. 535-581.
Kotelnikov, Vladimir A., The Theory of Optimum Noise Immunity, trans. R. A.
Silverman (New York: McGraw-Hill, 1959).
Kowalski, Robert, Logic for Problem Solving (New York: North-Holland, 1979).
Kristol, Irving, “Rationalism in Economics,” in Bell, Daniel, and Kristol, Irving,
eds., The Crisis in Economic Theory (New York: Basic Books, 1981).
Kuhn, Alfred, The Study of Society: A Unified Approach (Homewood, Ill.: Irwin-
Dorsey Press, 1963).
Kuhn, Alfred, The Logic of Social Systems (San Francisco: Jossey-Bass, 1974).
Kuhn, Alfred, Unified Social Science (Homewood, Ill.: Dorsey Press, 1975).
Kuhn, Alfred, “Let’s Bring the Skeleton Out of the Closet,” in Lasker, E. G., ed..
Applied Systems and Cybernetics (New York: Pergamon, 1981).
Kuhn, Alfred, and Beam, Robert D., The Logic of Organization (San Francisco:
Jossey-Bass, 1982).

Kuhn, Thomas S., The Structure of Scientific Revolutions (Chicago: University of

Chicago Press, 1962o; 2nd enlarged ed., 1970).

Kuhn, Thomas S., “The Historical Structure of Scientific Discovery,” in Science,

vol. 136 (Apr./June 19626), pp. 760-764.

Kuhn, Thomas S., “Second Thoughts on Paradigms,” in Suppe, Frederick, ed.. The
Structure of Scientific Theories (Urbana: University of Illinois Press, 1974), pp
459-482.

Kuhn, Thomas S., The Essential Tension: Selected Studies in Scientific Tradition
and Change (Chicago: University of Chicago Press, 1977).
CUMULATIVE LIST OF REFERENCES 701

Laferriere, Daniel, “Making Room for Semiotics,” in Academe, vol. 65 (Nov.

1979), pp. 434-440.
Lakatos, Imre, “Falsification and the Methodology of Scientific Research Pro¬
grammes,” in Lakatos, Imre, and Musgrave, Alan, eds., Criticism and the
Growth of Knowledge, proceedings of the International Colloquium in the Phil¬
osophy of Science, London, 1965 (Cambridge: Cambridge University Press,
1970), vol. 4, pp. 91-195.
Lakatos, Imre, and Musgrave, Alan, Criticism and the Growth of Knowledge
(Cambridge: Cambridge University Press, 1970).
Lakoff, George, “Some Remarks on AI and Linguistics,” in Cognitive Science, vol.
2 (1978), pp. 267-275.
Lakoff, George, and Ross, John R., “Is Deep Structure Necessary?” in McCawley,
James D., ed., Syntax and Semantics: Vol. 7: Notes from The Linguistic Under¬
ground (New York: Academic Press, 1976), pp. 159-164.
Lamberton, Donald M., ed., Economics of Information and Knowledge (Har-
mondsworth, England: Penguin Books, 1971).
Lange, Oskar, Wholes and Parts: A General Theory of System Behaviour, trans.
Eugeniusz Lepa of Calosc i rozwoj w swietle cybernetyki [Warsaw: Polish
Scientific Publishers, 1962] (Oxford, England: Pergamon Press, 1965).
Langefors, Borje, “Information Systems Theory,” in Information Systems, vol. 2,
no. 3 (1977), pp. 207-219.
Langendoen, D. Terrence, and Postal, Paul M., “The Vastness of Natural Lan¬
guage” (1982).
Langer, Susanne K., Philosophical Sketches (Baltimore: Johns Hopkins Press,
1962).
Langlois, Richard N., “Knowledge, Order, and Technology” (Ph.D. diss., Stan¬
ford University, 1981).
Lashley, Karl S., “The Problem of Serial Order in Behavior,” In Jeffress, Lloyd
A., ed., Cerebral Mechanisms in Behavior, proceedings of the Hixon Sym¬
posium, held [at the California Institute of Technology, Pasadena] 1948 (New
York: Wiley-Interscience, 1951), pp. 112-136.
Lees, Sidney, “Uncertainty and Imprecision,” in Transactions of the American
Society of Mechanical Engineers, Journal of Basic Engineering, series D, vol. 88
(June 1966), pp. 369-378.
Lees, Sidney, “On Measurement,” Applied Mechanics Reviews (forthcoming 1983).
Lenat, Douglas, The Heuristics of Nature (Stanford University, Heuristic Program¬
ming Project, Working Paper No. HPP-81-22, 1981).
Lenneberg, Eric H., Biological Foundations of Language (New York: Wiley, 1967).
Leontief, Wassily, “Academic Economics,” letter to the editor, Science, vol. 217
(9 July 1982).
Lesser, Victor,“AI and Brain Theory (BT) Research at the University of Massachu¬
setts at Amherst,” in AI Magazine (winter 1981-1982), pp. 16-20.
Lesser, Victor, and Erman, Lee, “A Retrospective View of the Hearsay-II Ar¬
chitecture,” in Proceedings of the Fifth International Joint Conference on
Artificial Intelligence (Cambridge, Mass.: MIT Artificial Intelligence Labora¬
tory, 1977), pp. 790-800.
702 CUMULATIVE LIST OF REFERENCES

Levy, J., “Lateral Specialization of the Human Brain: Behavioral Manifestations

and Possible Evolutionary Basis,” in Kiger, John A., Jr., ed.. The Biology of
Behavior (Corvallis: Oregon State University Press, 1972), pp. 159-180.
Lewin, Kurt, Principles of Topological Psychology, trans. Fritz Heider (New York
and London: McGraw-Hill, 1936).
Lewis, Gilbert N., “The Symmetry of Time in Physics,” in Science, vol. 71 (6 June
1930), pp. 569-577.
Licklider, J. C. R., “Man-Computer Symbiosis,” in IRE Transactions on Human
Factors in Electronics, vol. HFE-1 (1960), pp. 4-11.
Lieberman, Philip, “Primate Vocalizations and Human Linguistic Ability,” in Jour¬
nal of the Acoustical Society of America, vol. 44 (1968), pp. 1574-1584.
Lindsay, Peter H., and Norman, Donald A., Human Information Processing: An
Introduction to Psychology (New York: Academic Press, 1972; 2nd ed., 1977).
Lindsay, Robert; Buchanan, Bruce; Feigenbaum, Edward A.; and Lederberg,
Applications of Artificial Intelligence for Organic Chemistry: The
Joshua,
DENDRAL Project (New York: McGraw-Hill, 1980).
Lions, Jacques Louis, Controle optimal de systemes gouvernes par des equations
aux derivees partielles (Paris: Dunod, 1968).
Lions, Jacques Louis, Function Spaces and Optimal Control of Distributed Systems
(Paris: College de France, 1977).
Locke, John, An Essay Concerning Humane Understanding (London: Thomas Bas¬
set, 1690; 32nd ed., W. Tegg, 1860).
Locke, William Nash, and Booth, A. Donald, eds., Machine Translation of Lan¬
guages: Fourteen Essays (Cambridge, Mass., and New York: Technology Press
of MIT and Wiley, 1957).
Loebner, Egon E., “Subhistories of the Light-Emitting Diode (LED),” in IEEE
Transactions on Electron Devices, vol. 23 (1976), pp. 675-699.
Loebner, Egon E., and Borden, H. “Ecological Niches for Optoelectronic De¬
vices,” in WESCON, vol. 13, session no. 20 (1969), pp. 1-8.
Lorenz, Edward, “Deterministic, Non-periodic Flow,” in Journal of Atmospheric
Sciences, vol. 82 (Mar. 1963), pp. 985-992.
Lorenz, Konrad, Evolution and Modification of Behavior (Chicago: University of
Chicago Press, 1965).
Lotke, Alfred J., Elements of Mathematical Biology (New York: Dover, 1924;
another ed. 1956).
Lowen, R., “Compact Hausdorff Fuzzy Topological Spaces are Topological,” in
Topology and Its Applications, vol. 12 (Jan. 1981), pp. 65-74.
Lucretius, Titus Lucretius Carus, On the Nature of Things (in six books), trans.
J. S. Watson (London: H. G. Bohn, 1851; 1st ed. in Latin, Verona, 1486).
Luenberger, David G., “Observing the State of a Linear System,” in IEEE Transac¬
tions on Military Electronics, vol. MIL-8 (Jan. 1964), pp. 74-80.
Luenberger, David G., “Observers for Multivariable Systems,” in IEEE Transac¬
tions on Automatic Control, vol. AC-11 (Apr. 1966), pp. 190-197.
Luenberger, David G., “An Introduction to Observers,” in IEEE Transactions on
Automatic Control, vol. AC-16 (Dec. 1971), pp. 596-602.
CUMULATIVE LIST OF REFERENCES 703

Luenberger, David G., Introduction to Dynamic Systems (New York: Wiley, 1979).
Lukes, Steven,“Types of Individualism,” in Dictionary of the History of Ideas, vol.
2 (New York: Scribner’s, 1973), pp. 594-604.
MacDougall, William, Body and Mind; A History and a Defense of Animism
(Boston: Beacon Press, 1961; 1st ed., London: Methuen and Co., 1911).
Machlup, Fritz, The Production and Distribution of Knowledge in the United States
(Princeton, N.J.: Princeton University Press, 1962).
Machlup, Fritz, Essays in Economic Semantics (Englewood Cliffs, N.J.: Prentice-
Hall, 1963).
Machlup, Fritz, “Theories of the Firm: Marginalist, Behavioral, Managerial,” in
American Economic Review, vol. 57 (Mar. 1967), pp. 1-33.
Machlup, Fritz, “If Matter Could Talk,” in Morgenbesser, Sidney; Suppes, Pat¬
rick; and White, Morton, eds.. Philosophy, Science, and Method (New York:
St. Martin’s Press, 1969a), pp. 286-305. Reprinted in Machlup, Fritz, Methodol¬
ogy of Economics and Other Social Sciences (New York: Academic Press,
1978), pp. 309-332.
Machlup, Fritz, “Positive and Normative Economics: An Analysis of the Ideas,” in
Heilbroner, Robert L., ed.. Economic Means and Social Ends: Essays in Polit¬
ical Economics (Englewood Cliffs, N.J.: Prentice-Hall, 1969b), pp. 99-129. Re¬
printed in Machlup, Fritz, Methodology of Economics and Other Social Sci¬
ences (New York: Academic Press, 1978), pp. 425-450.
Machlup, Fritz, Methodology of Economics and Other Social Sciences (New York:
Academic Press, 1978).
Machlup, Fritz, “Meeting Review: An Economist’s Reflections on an Institute for
Advanced Study of Information Science,” in Journal of the American Society
for Information Science, vol. 30 (Mar. 1979a), pp. 111—113.
Machlup, Fritz, “Uses, Value, and Benefits of Knowledge,” in Knowledge, vol. 1
(Sept. 1979b), pp. 62-81.
Machlup, Fritz, Knowledge: Its Creation, Distribution, and Economic Signifi¬
cance, Volume I: Knowledge and Knowledge Production (Princeton, N.J.:
Princeton University Press, 1980).
Machlup, Fritz, Knowledge: Its Creation, Distribution, and Economic Signifi¬
cance, Volume II: The Branches of Learning (Princeton, N.J.: Princeton Uni¬
versity Press, 1982).
Machlup, Fritz, Knowledge: Its Creation, Distribution, and Economic Signifi¬
cance, Volume III: The Economics of Information and Human Capital (Prince¬
ton, N.J.: Princeton University Press, 1984).
Machlup, Fritz; Leeson, Kenneth;'and Associates,Information Through the
Printed Word: The Dissemination of Scholarly, Scientific, and Intellectual
Knowledge, Vol. 1: Book Publishing (New York: Praeger, 1978a).
Machlup, Fritz; Leeson, Kenneth; and Associates, Information Through the
Printed Word: The Dissemination of Scholarly, Scientific, and Intellectual
Knowledge, Vol. 2: Journals (New York: Praeger, 1978b).
Machlup, Fritz; Leeson, Kenneth; and Associates, Information Through the
Printed Word: The Dissemination of Scholarly, Scientific, and Intellectual
Knowledge, Vol. 3: Libraries (New York: Praeger, 1978c).
704 CUMULATIVE LIST OF REFERENCES

Machlup, Fritz; Leeson, Kenneth; and Associates, Information Through the

Printed Word: The Dissemination of Scholarly, Scientific, and Intellectual
Knowledge, Vol. 4: Books, Journals, and Bibliographic Services (New York:
Praeger, 1980).
MacKay, Donald M., “Quantal Aspects of Scientific Information,” in Philosophical
Magazine, vol. 41 (1950), pp. 289-311.
MacKay, Donald M., ‘‘The Nomenclature of Information Theory,” in Proceedings
of the First International Symposium on Information Theory, held [in London]
Sept., 1950, sponsored by the American Institute of Radio Engineers. Litho¬
printed, 1953. Reprinted in Foerster, Heinz M. von, ed., Cybernetics, transac¬
tions of the Eighth Conference sponsored by the Josiah Macy Jr. Foundation
(New York: Josiah Macy Jr. Foundation, 1951 <a).
MacKay, Donald M., “In Search of Basic Symbols,” in Foerster, Heinz M. von,
ed., Cybernetics, transactions of the Eighth Conference sponsored by the Josiah
Macy Jr. Foundation (New York: Josiah Macy Jr. Foundation, 195\b), pp. 181—
221.
MacKay, Donald M., “On Comparing the Brain with Machines,” in American
Scientist, vol. 42 (Apr. 1954a), pp. 261-268.
MacKay, Donald M., “Operational Aspects of Some Fundamental Concepts of
Human Communication,” in Synthese, vol. 9 (19546), pp. 182-198.
MacKay, Donald M., “Moving Visual Images Produced by Regular Stationary
Patterns,” in Nature, vol. 180 (Oct. 26, 1957a), pp. 849-850.
MacKay, Donald M., “Some Further Visual Phenomena Associated With Regular
Patterned Stimulation,” in Nature, vol. 180 (Nov. 23, 19576), pp. 1145-1146.
MacKay, Donald M., “Information Theory: Its Uses and Abuses,” in Bulletin of
the British Psychological Society, vol. 38 (1959), pp. 70-71.
MacKay, Donald M., “Interactive Processes in Visual Perception,” in Rosenblith,
Walter A., ed., Sensory Communication (Cambridge, Mass., and New York:
MIT Press and Wiley, 1960), pp. 339-355.
MacKay, Donald M., “Information and Learning,” in Billing, H., ed.. Learning
Automata (Munich: Oldenbourg, 1961a), pp. 40-49.
MacKay, Donald M., “The Place of ‘Meaning’ in the Theory of Information,” in
Cherry, Colin, ed., Information Theory, papers read at a symposium on Infor¬
mation Theory, held [at the Royal Institution, London] Aug. 29-Sept. 2, 1960
(London: Butterworths, 19616), pp. 215-225. Reprinted in MacKay, Donald M.,
Information, Mechanism and Meaning (Cambridge, Mass.: MIT Press, 1969),
pp. 79-93.
MacKay, Donald M., “The Informational Approach,” in Cybernetics of Neural
Processes, La ricerca scientifica, 31 (Rome: Consiglio Nazionale delle Ricerche,
1964a), pp. 123-128.
MacKay, Donald M., “Information Theory in the Study of Man,” in Cohen, John,
ed., Readings in Psychology (London: Allen and Unwin, 19646), pp. 214-235.
MacKay, Donald M., “Information in Brains and Machines,” in Kalenich, Wayne
A., ed., Proceedings of the IFIP Congress 1965, organized by the International
Federation for Information Processing [New York] May 24-29, 1965 (Washing¬
ton, D.C.: Spartan Books; London: Macmillan, 1965), vol. 2, pp. 637-643.
CUMULATIVE LIST OF REFERENCES 705

MacKay, Donald M., Information, Mechanism and Meaning (Cambridge, Mass.:

MIT Press, 1969).
Majumder,D. Dutta, “Cybernetics and General Systems—A Unitary Science?” in
Kybernetes, vol. 8 (1979), pp. 7-15.
N., The Soul of the White Ant, trans. Winifred De Kok (New York:
Marais, Eugene
Dodd, Mead, 1937; Kraus reprint, 1969).
March, James G., and Simon, Herbert A., Organizations (New York: Wiley, 1958).
Marcus, Mitchell P., A Theory of Syntatic Recognition for Natural Language
(Cambridge, Mass.: MIT Press, 1980).
Marr, David C., “Approaches to Biological Information Processing,” book review
of Conrad, Michael; Giittinger, W.; and Dal Cin, M., eds., Physics and Mathe¬
matics of the Nervous System, proceedings of a summer school organized by the
International Centre for Theoretical Physics, Trieste, and the Institute for Infor¬
mation Sciences, University of Tubingen, held [in Trieste, Italy] Aug. 21-31,
1973 (Berlin and New York: Springer-Verlag, 1974), in Science, vol. 190 (Nov.
1975), pp. 875-876.
Marr, David C., “Early Processing of Visual Information,” in Philosophical Trans¬
actions of the Royal Society of London, vol. 275 (1976), pp. 483-534.
Marr, David C., “Artificial Intelligence—A Personal View,” in Artificial Intelli¬
gence, vol. 9 (Aug. 1977), pp. 37-48.
Marr, David C., Vision: A Computational Investigation into the Human Represen¬
tation and Processing of Visual Information (San Francisco: W. H. Freeman,
1982).
Marr, David C., and Poggio, T., “From Understanding Computation to Under¬
standing Neural Circuitry,” in Neurosciences Research Program Bulletin, vol.
15 (1977), pp. 470-488.
Marschak, Jacob, “Towards an Economic Theory of Organization and Informa¬
tion,” in Thrall, R. M.; Coombs, C. H.; and Davis, R. L., eds., Decision Pro¬
cesses (New York: Wiley, 1954), pp. 187-220.
Marschak, Jacob, “Problems in Information Economics,” in Bonini, C. P.;
Jaedicke, R. K.; and Wagner, H. M., eds., Management Controls: New Direc¬
tions in Basic Research (New York: McGraw-Hill, 1964), pp. 38-90.
Marschak, Jacob, Economic Information, Decision and Interpretation—Selected
Essays, vols. 1-3 (Dordrecht, Holland, and Boston: D. Reidel, 1974).
Marschak, Jacob, and Radner, Roy, Economic Theory of Teams (New Haven,
Conn.: Yale University Press, 1972).
Martin, Clyde F., and Hermann, Robert, “Applications of Algebraic Geometry to
Systems Theory: The McMillan Degree and Kronecker Indices of Transfer
Functions in Topological and Homomorphic Invariants, in SIAM Journal of
Control and Optimization, vol. 16 (1978), pp. 743-755.
Martin, Nathaniel F. G., and England, James W., Mathematical Theory of En¬
tropy (Reading, Mass.: Addison-Wesley, 1981).
Martin, Paul E., “What Can Cybernetics Do For Industry?” in Austria Today, no.
2 (1981), pp. 11-14.
Maruyama, Magoroh, “The Second Cybernetics: Deviation-Amplifying Mutual
Causal Processes,” in The American Scientist, vol. 51 (June 1963), pp. 164-179.
706 CUMULATIVE LIST OF REFERENCES

Mattessich, Richard, “The Incorporation and Reduction of Value Judgments in

Systems,” in Management Science, vol. 21 (Sept. 1974a), pp. 1-9.
Mattessich, Richard, “Informations- und Erkenntnisekonomik: Treffpunkt von
Philosophic und Wirtschaftswissenschaft,” in Zeitschrift fur betriebswirtschaft-
liche Forschung, vol. 26 (Dec. 19746), pp. 777-784.
Mattessich, Richard, “Information Economics and the Notion of ‘MIS’,” in
Grochla, E., and Szyperski, N., eds., Information Systems and Organizational
Structure (New York: Walter De Gruyter, 1975), pp. 342-364.
Mattessich, Richard, Instrumental Reasoning and Systems Methodology—An
Epistemology of the Applied and Social Sciences (Dordrecht, Holland, and
Boston: D. Reidel, 1978).
Mattessich, Richard, “Axiomatic Representation of the Systems Framework:
Similarities and Differences between Mario Bunge’s World of Systems and My
Own Systems Methodology,” in Cybernetics and Systems, vol. 13 (1982), pp.
51-75.
Mattessich, Richard, ed., Modern Accounting Research: A Survey and Guide (Van¬
couver: Research Foundation of the Canadian Certified General Accountants’
Association, forthcoming 1983).
Maturana, Humberto R., “Biology of Language: The Epistemology of Reality,” in
Miller, George A., and Lenneberg, Elizabeth, eds., Psychology and Biology of
Language and Thought: Essays in Honor of Eric Lenneberg (New York: Aca¬
demic Press, 1978), pp. 27-63.
Maxwell, James Clerk, “On Governors,” in Proceedings of the Royal Society of
London, vol. 16 (Mar. 1868), pp. 270-283. Reprinted in Niven, William David¬
son, ed., The Scientific Papers of James Clerk Maxwell (Cambridge: Cambridge
University Press, 1890; New York: Dover, 1965), vol. 2, pp. 105-120.
May, R., “The Grammar of Quantification” (Ph.D. diss., Massachusetts Institute of
Technology, 1977).
May, Robert, “Simple Mathematical Models with Very Complicated Dynamics,” in
Nature, vol. 261 (1976), pp. 459-467.
May, Rollo, Stability and Complexity in Model Ecosystems (Princeton, N.J.:
Princeton University Press, 1973).
Mayne, A. J., “Cybernetics and General Systems—A Unitary Science?” in Kyber-
netes, vol. 8 (1979), pp. 39-43.
McCarthy, John, “Programs with Common Sense,” in Mechanization of Thought
Processes (London: HMSO, 1959).
McCorduck, Pamela, Machines Who Think: A Personal Inquiry into the History and
Prospects of Artificial Intelligence (San Francisco: W. H. Freeman, 1979).
McCulloch, Warren S., “A Historical Introduction to the Postulational Founda¬
tions of Experimental Epistemology,” in McCulloch, Warren S., Embodiments
of Mind (Cambridge, Mass.: MIT Press, 1965), pp. 359-372.
McCulloch, Warren S., and Pitts, Walter H., “A Logical Calculus of the Ideas
Immanent in Nervous Activity,” in Bulletin of Mathematical Biophysics, vol. 5
(1943), pp. 115-133.
McEliece, Robert J., The Theory of Information and Coding (Reading, Mass.: Addi-
son-Wesley, 1977).
CUMULATIVE LIST OF REFERENCES 707

McGill, William J., “Multivariate Information Transmission,” in Psychometrika,

vol. 19 (June 1954), pp. 97-116.
McRae, Robert, “The Unity of the Sciences: Bacon, Descartes, and Leibniz,” in
Journal of the History of Ideas, vol. 18 (Jan. 1957), pp. 27-48.
Mead, George Herbert, Mind, Self, and Society (Chicago: University of Chicago
Press, 1934).
Meadows, Donnella H., et al., The Limits to Growth: A Report for the Club of
Rome’s Project on the Predicament of Mankind (New York: Universe Books,
1972).
G., eds., System Identification: Advances
Mehra, Raman K., and Lainiotis, Dimitri
and Case Studies (New York: Academic Press, 1976).
Merton, Robert K., The Sociology of Science: Theoretical and Empirical Investiga¬
tions, ed. Norman W. Storer (Chicago: University of Chicago Press, 1973).
Mesarovic, Mihajlo D., “On Self-Organizational Systems,” in Yovits, Marshall C.;
Jacobi, George T.; and Goldstein, Gordon D., eds., Self-Organizing Systems,
proceedings of a conference on self-organizing systems, held [in Chicago] 1962
(Washington, D.C.: Spartan Books, 1962), pp. 9-36.
Mesarovic, Mihajlo D., ed., Views on General Systems Theory, proceedings of the
Second Systems Symposium, held [at Case Institute of Technology] 1963 (New
York: Wiley, 1964a).
Mesarovic, Mihajlo D., “Foundations for a General Systems Theory,” in
Mesarovic, Mihajlo D., ed., Views on General Systems Theory, proceedings of
the Second Systems Symposium, held [at Case Institute of Technology] 1963
(New York: Wiley, 1964ft), pp. 1-24.
ed., Systems Theory and Biology, proceedings of the Third
Mesarovic, Mihajlo D.,
Systems Symposium at Case Institute of Technology (New York: Springer-
Verlag, 1968a).
Mesarovic, Mihajlo D., “Systems Theory and Biology—View of a Theoretician,”
in Mesarovic, Mihajlo D., ed., Systems Theory and Biology, proceedings of the
Third Systems Symposium at Case Institute of Technology (New York: Sprin-
ger-Verlag, 1968ft), pp. 59-87.
Mesarovic, Mihajlo D.; Macro, Donald; and Takahara, Yasuhiko, Theory of
Hierarchical, Multi-Level Systems (New York: Academic Press, 1970).
Mesarovic, Mihajlo D., and Pestel, Eduard, Mankind at the Turning Point: The
Second Report to the Club of Rome (New York: Dutton, 1974).
Mesarovic, Mihajlo D., and Takahara, Yasuhiko,General Systems Theory: Math¬
ematical Foundations (New York: Academic Press, 1975).
Meyer, Jean, “Essai d’application de certains modeles cybernetiques a la coordina¬
tion chez les insectes sociaux,” in Insectes Sociaux, vol. 13, no. 2 (1966), p. 127.
Meyer-Eppler, Werner, “Vorwort des Herausgebers,” Grundlagen und Anwen-
dungen der Informationstheorie [Kommunikation und Kybernetik in Einzeldar-
stellungen, Band 1, Herausgegeben von W. Meyer-Eppler] (Berlin: Springer-
Verlag, 1959).
Michalski, RyszardS.; Carbonell, Jaime; and Mitchell, Tom, eds., Machine
Learning—An Artificial Intelligence Approach (Palo Alto, Calif.: Tioga Press,
1983).
708 CUMULATIVE LIST OF REFERENCES

Mill, John Stuart, “On the Definition of Political Economy; and on the Method of
Investigation Proper to It,” in London and Westminster Review, October, 1836;
reprinted in Mill, John Stuart, Essays on Some Unsettled Questions of Political
Economy (London: John W. Parker, 1844).
Mill, John Stuart, Essays on Some Unsettled Questions of Political Economy
(London: John W. Parker, 1844; 1st ed. 1836).
Miller, George A., Language and Communication (New York: McGraw-Hill,
1951).
Miller, George A., “What Is Information Measurement?” in American Psycholo¬
gist, vol. 8 (Jan. 1953), pp. 3-11.
Miller, George A., “The Magical Number Seven, Plus or Minus Two: Some Limits
on our Capacity for Processing Information,” in Psychological Review, vol. 63
(Mar. 1956), pp. 81-97.
Miller, George A., “Some Psychological Studies of Grammar,” in American Psy¬
chologist, vol. 17 (1962), pp. 748-762.
Miller, George A., and Chomsky, Noam, “Finitary Models of Language Users,” in
Luce, Robert Duncan; Bush, Robert R.; and Galanter, Eugene, eds.. Handbook
of Mathematical Psychology, vol. 2 (New York: Wiley, 1963), chap. 13, pp. 419—
492.
Miller, George A., and Frick, Frederick C., “Statistical Behavioristics and Se¬
quences of Responses,” in Psychological Review, vol. 56 (Nov. 1949), pp. 311 —
324.
Miller, George A.; Galanter, Eugene; and Pribram, Karl H., Plans and the
Structure of Behavior (New York: Holt, Rinehart, and Winston, 1960).
Miller, George A., and Johnson-Laird, Philip N., Language and Perception (Cam¬
bridge, Mass.: Belknap Press, 1976).
Miller, James G., “The Nature of Living Systems,” in Behavioral Science, vol. 21
(Sept. 1976), pp. 295-319.
Miller, James G., Living Systems (New York: McGraw-Hill, 1978).
Miller, Jonathan, The Body in Question (New York: 1st American ed., Random
House, 1978).
Miller, Laurence, “Has Artificial Intelligence Contributed to an Understanding of
the Human Mind? A Critique of Arguments For and Against,” in Cognitive
Science, vol. 2 (Apr./June 1978), pp. 111-127.
Milsum, John H., Biological Control Systems Analysis (New York: McGraw-Hill,
1966).
Minsky, Marvin L., “Steps Toward Artificial Intelligence,” in Feigenbaum, Edward
A., and Feldman, Julian, eds.. Computers and Thought (New York: McGraw-
Hill, 1963), pp. 406-450.
Minsky, Marvin L., ed., Semantic Information Processing (Cambridge, Mass.: MIT
Press, 1968).
Minsky, Marvin L., “A Framework for Representing Knowledge,” in Winston,
Patrick H., ed., The Psychology of Computer Vision (New York: McGraw-Hill,
1975), pp. 211-277.
Minsky, Marvin L., “Computer Science and the Representation of Knowledge,” in
CUMULATIVE LIST OF REFERENCES 709

Dertouzos, Michael L., and Moses, Joel, eds., The Computer Age: A Twenty-
Year View (Cambridge, Mass.: MIT Press, 1979), pp. 392-421.
Minsky, Marvin L., and Papert, Seymour, Perceptrons: An Introduction to Compu¬
tational Geometry (Cambridge, Mass.: MIT Press, 1969).
Mittelmann, H. D., and Weber, H., eds.. Bifurcation Problems and Their Numeri¬
cal Solution. [Workshop on Bifurcation Problems and Their Numerical Solution,
University of Dortmund, Jan. 15-17, 1980] (Basel, Switzerland, and Boston:
Birkhauser, 1980).
Moivre, Abraham de, The Doctrine of Chances; or, a Method of Calculating the
Probability of Events in Play (London: 1718; 3rd ed. 1756; photo, reproduction
of 3rd ed. together with the biography by Helen M. Walker, New York: Chelsea
Press, 1967).
Monod, Jacques, Chance and Necessity: An Essay on the Natural Philosophy of
Modern Biology, trans. Austryn Wainhouse [of Le hasard et la necessite: Essai
sur la philosophic naturelle de la biologie moderne, Paris: Editions du Seuil,
1970] (New York: 1st American ed., Knopf, 1971).
Montroll, Elliott W., “Theory and Observations of the Dynamics and Statistics of
Traffic on an Open Road,” in Bogdanoff, John L., and Kozin, Frank, eds.,
Proceedings of the First Symposium on Engineering Applications of Random
Function Theory and Probability (New York: Wiley, 1963), pp. 231-269.
Montroll, Elliott W., “On the Entropy Function in Sociotechnical Systems,” in
Proceedings of the National Academy of Sciences, vol. 78 (Dec. 1981), pp.
7839-7843.
Moore, Edward F., and Shannon, Claude E., “Reliable Circuits Using Less Reli¬
able Relays,” in Journal of the Franklin Institute, part 1, vol. 262 (Sept. 1956),
pp. 191-208; part 2, vol. 262 (Oct. 1956), pp. 281-297.
Morris, Charles W., “Foundations of the Theory of Signs,” in Neurath, Otto;
Carnap, Rudolf; and Morris, Charles, eds., Foundations of the Unity of Science:
Toward an International Encyclopedia of Unified Science, vol. 1 (Chicago: Uni¬
versity of Chicago Press, 1939; bound ed., 1969), pp. 77-137.
Moser, Jungen, Stable and Random Motions in Dynamical Systems (Princeton,
N.J.: Princeton University Press, 1973).
Moses, Joel, “Symbolic Integration” (Ph.D. diss., Massachusetts Institute of Tech¬
nology, 1967).
Mountcastle, Vernon B., “An Organizing Principle for Cerebral Function: The
Unit Module and the Distributed System,” in Edelman, Gerald M., and Mount¬
castle, Vernon B., The Mindful Brain (Cambridge, Mass.: MIT Press, 1978), pp.
7-50.
Nalimov, Vasilii V., “The Penetration of the Humanities into Other Fields of
Knowledge,” in Colodny, Robert G., ed., Faces of Science (Philadelphia: ISI
Press, 1981; 1st pub. in Russian in the journal Znanie-sila, no. 5 [1979]).
Naur, Peter; Randell, Brian; and Buxton, John N., eds.,Software Engineering:
Concepts and Techniques, proceedings of the NATO conferences (New York:
Petrocelli/Charter, 1976). Part I reports on the conference held [in Garmisch,
Germany] by the NATO Science Committee, Oct. 7-11, 1968; Part II on the con¬
ference held [in Rome] by the NATO Science Committee, Oct. 27-31, 1969.
710 CUMULATIVE LIST OF REFERENCES

Neisser, Ulric, “Decision-time without Reaction-time: Experiments in Visual Scan¬

ning,” in American Journal of Psychology, vol. 76 (Sept. 1963), pp. 376-385.
Neisser, Ulric, Cognitive Psychology (New York: Appleton-Century-Crofts, 1967).
Neisser, Ulric, Cognition and Reality (San Francisco: W. H. Freeman, 1976).
Neufeld, Richard W. J., “Response-Selection Processes in Paranoid and Non-
Paranoid Schizophrenia,” in Perceptual and Motor Skills, vol. 44 (Apr. 1977),
pp. 499-505.
Neumann, Johann von, Mathematische Grundlagen der Quantenmechanik (Berlin:
Julius Springer, 1932).
Neumann, John von, Mathematical Foundations of Quantum Mechanics, trans.
Robert T. Beyer (Princeton, N.J.: Princeton University Press, 1955; German
ed., 1932).
Neumann, John von, “Probabilistic Fogies and the Synthesis of Reliable Organisms
from Unreliable Components,” in Shannon, Claude E., and McCarthy, John,
eds.. Automata Studies (Princeton, N.J.: Princeton University Press, 1956), pp.
43-98.
Neumann, John von, The Computer and the Brain (New Haven, Conn.: Yale Uni¬
versity Press, 1958).
Nevitt, Barrington, “The Complementarity of Art and Science in General Sys¬
tems,” in Lasker, G. E., ed., Applied Systems and Cybernetics (New York:
Pergamon, 1981), pp. 592-596.
Newell, Allen, “Remarks on the Relationship Between Artificial Intelligence and
Cognitive Psychology,” in Baneiji, Ranan B., and Mesarovic, Mihajlo D., eds.,
Theoretical Approaches to Non-Numerical Problem Solving, proceedings of the
Fourth Systems Symposium, held [at Case Western Reserve University] 1968
(New York: Springer-Verlag, 1970), pp. 363-400.
Newell, Allen, “Production Systems: Models of Control Structures,” in Chase,
William G., ed., Visual Information Processing, proceedings of the Eighth An¬
nual Carnegie Symposium on Cognition, held [at Carnegie-Mellon University,
Pittsburgh] May 19, 1972 (New York: Academic Press, 1973u), pp. 463-526.
Newell, Allen, “You Can’t Play Twenty Questions With Nature and Win: Projec¬
tive Comments on the Papers of this Symposium,” in Chase, William G., ed.,
Visual Information Processing, proceedings of the Eighth Annual Carnegie
Symposium on Cognition, held [at Carnegie-Mellon University, Pittsburgh] May
19, 1972 (New York: Academic Press, 1973b), pp. 283-310.
Newell, Allen, “Artificial Intelligence and the Concept of Mind,” in Schank,
Roger C., and Colby, Kenneth M., eds.. Computer Models of Thought and
Language (San Francisco: W. H. Freeman, 1973c), pp. 1-60.
Problem
Newell, Allen, “One Final Word,” in Tuma, David T., and Reif, F., eds.,
Solving and Education: Issues in Teaching and Research, proceedings of a
conference held [at Carnegie-Mellon University, Pittsburgh] Oct. 9-10, 1978
(Hillsdale, N.J.: Lawrence Erlbaum, 1980u), pp. 175-189.
Newell, Allen, “Physical Symbol Systems,” in Cognitive Science, vol. 4 (Apr./
June 1980b), pp. 135-183.
Newell, Allen, “The Knowledge Level,” in AI Magazine, vol. 2 (1981), pp. 1-20.
Newell, Allen, and Ernst, George, “The Search for Generality,” in Kalenich,
CUMULATIVE LIST OF REFERENCES 711

Wayne A., ed., Proceedings of the IFIP Congress 1965, organized by the Inter¬
national Federation for Information Processing [New York] May 24-29, 1965
(Washington, D.C.: Spartan Books; London: Macmillan, 1965), vol. 1, pp. 17-
24.
Newell, Allen; Perlis, Alan J.; and Simon, Herbert A., “Computer Science,”
letter to the editor, Science, vol. 157 (22 Sept., 1967), pp. 1373-1374.
Newell, Allen, and Shaw, John C., “Programming the Logic Theory Machine,” in
Proceedings of the Western Joint Computer Conference, sponsored by the Insti¬
tute of Radio Engineers, the American Institute of Electrical Engineers, and the
Association for Computing Machinery and held [in Los Angeles] Feb. 26-28,
1957, pp. 230-240.
Newell, Allen; Shaw, John C.; and Simon, Herbert A., “Empirical Explorations
of the Logic Theory Machine: A Case Study in Heuristics,” in Proceedings of
the Western Joint Computer Conference, sponsored by the Institute of Radio
Engineers, the American Institute of Electrical Engineers, and the Association
for Computing Machinery and held [in Los Angeles] Feb. 26-28, 1957, pp. 218—
230.
Newell, Allen, and Simon, Herbert A., Human Problem Solving (Englewood
Cliffs, N.J.: Prentice-Hall, 1972).
Newell, Allen, and Simon, Herbert A., “Computer Science as Empirical Inquiry:
Symbols and Search” [the 1976 ACM Turing Award Lecture], in Communica¬
tions of the ACM, vol. 19 (Mar. 1976a), pp. 113-126.
Newell, Allen, and Simon, Herbert A., “Symbol Manipulation,” in Encyclopedia
of Computer Science (New York: Petrocelli/Charter, 1976b), pp. 1384-1389.
Newmeyer, Frederick J., Linguistic Theory in America: The First Quarter Century
of Transformational Generative Grammar (New York: Academic Press, 1980).
Ng, David S., and Stoeckenius, Jan, “Auditing: Incentives and Truthful Report¬
ing,” in Journal of Accounting Research, vol. 17 (supplement 1979), pp. 1-24.
Nicolis, George, and Prigogine, Ilya, Self-Organization in Nonequilibrium Sys¬
tems: From Dissipative Structures to Order through Fluctuations (New York:
Wiley, 1977).
Nilsson, Nils J., Problem Solving Methods in Artificial Intelligence (New York:
McGraw-Hill, 1971).
Nilsson, Nils J., “Artificial Intelligence,” in Rosenfeld, J. L., ed., Proceedings of
the IFIP Congress, vol. 4 (New York: American Elsevier, 1974), pp. 778-801.
Nilsson, Nils J., Principles of Artificial Intelligence (Palo Alto, Calif.: Tioga Press,
1980).
Nilsson, Nils J., “Artificial Intelligence: Engineering, Science or Slogan?” in Al
Magazine, vol. 3 (Jan. 1982), pp. 2-9.
Nilsson, Nils J., and Raphael, Bertram, “Preliminary Design of an Intelligent
Robot,” in Computer and Information Sciences, vol. 2 (1967), pp. 235-259.
Nirenberg, Marshall W., “The Genetic Code: II,” in Scientific American, vol. 208
(Mar. 1963), pp. 80-94.
Nisbett, Richard, and Ross, Lee, Human Inference: Strategies and Shortcomings
of Social Judgment (Englewood Cliffs, N.J.: Prentice-Hall, 1980).
Norman, Donald A., “Memory, Knowledge, and the Answering of Questions,” in
 712
CUMULATIVE LIST OF REFERENCES

Solso, Robert L., ed., Contemporary Issues in Cognitive Psychology: The Loy¬
ola Symposium on Cognitive Psychology [Chicago] 1972 (Washington, D.C.:
Winston, 1973), pp. 135-165.

Norman, Donald A., “Twelve Issues for Cognitive Science,” in Cognitive Science,
vol. 4 (Jan./Mar. 1980), pp. 1-32.

Norman, Donald A., “Categorization of Action Slips,” in Psychological Review,

vol. 88 (Jan. 1981), pp. 1-15.

Nozick, Robert, Anarchy, State, and Utopia (New York: Basic Books, 1974).
Nyquist, Harry, “Certain Factors Affecting Telegraph Speed,” in Journal of the
American Institute of Electrical Engineers, vol. 43 (Feb. 1924), pp. 124-130.
Ochs, Sidney, Elements of Neurophysiology (New York: Wiley, 1965).
Offray de la Mettrie, Julien Jan, Man a Machine (London: 2nd ed., G. Smith,
1750; ed. with notes by Gertrude Carman Bussey, La Salle, Ill.: Open Court
Publishing Company, 1961). Translated from the French.

Otlet, Paul, “Les sciences bibliographiques et la documentation,” in Institut Inter¬

national de Bibliographie Bulletin 8 (1903), pp. 121-147.
Otlet, Paul, and Goldschmidt, Robert, Sur une forme nouvelle du livre: le livre
microphotographique (Brussels: Institut International de Bibliographie, 1906).
Publication 81.

Otten, Klaus, “Basis for a Science of Information,” in Debons, Anthony, ed..

Information Science: Search for Identity, proceedings of the 1972 NATO Ad¬
vanced Study Institute in Information Science, held [in Seven Springs, Cham¬
pion, Penn.] Aug. 12-20, 1972 (New York: Marcel Dekker, 1974), pp. 91-106.
Otten, Klaus, and Debons, Anthony, “Opinion Paper: Towards a Metascience of
Information: Informatology, ” in Journal of the American Society for Informa¬
tion Science, vol. 21 (Jan./Feb. 1970), pp. 89-94.
Papert, Seymour, Paper given at the NUFFIC [Netherlands Universities Foundation
for International Cooperation] Summer School, 1972 (The Hague, Netherlands-
NUFFIC, 1972).

Papert, Seymour, Mindstorms: Children, Computers, and Powerful Ideas (New

York: Basic Books, 1980).

Parsons, Talcott, “Concrete Systems and ‘Abstracted’ Systems,” in Behavioral

Science, vol. 25 (Jan. 1980), pp. 46-55. Reproduced from Contemporary Sociol¬
ogy, 1979.
Pearson, Charls, “Towards an Empirical Foundation of Meaning” (Ph.D. diss
Georgia Institute of Technology, 1977).

Pearson, Charls, and Slamecka, Vladimir, Semiotic Foundations of Information

Science. Report of the School of Information and Computer Science, GIT-ICS-
77-01 (Atlanta: Georgia Institute of Technology, 1977).

Peirce, Charles Sanders, “On a New List of Categories,” in Hartshorne, Charles

and Weiss, Paul, eds., Collected Papers of Charles Saunders Pierce, Vol. I:
Principles of Philosophy (Cambridge, Mass.: Harvard University Press, 1931),
chap. 6, pp. 287-305. [Originally published in Proceedings of the American
Academy of Arts and Sciences, vol. 7 (May 1867), pp. 287-298.]
Pellionisz, Andras, and Llinas, Rodolfo, “Brain Modelling by Tensor Network
CUMULATIVE LIST OF REFERENCES 713

Theory and Computer Simulation. The Cerebellum: Distributed Processor for

Predictive Coordination,” in Neuroscience, vol. 4 (1979), pp. 323-348.
Peritz, Bluma C., ‘‘The Methods of Library Science Research: Some Results from a
Bibliometric Survey,” in Library Research, vol. 2 (1980-1981), pp. 251-268.
Perlis, Alan J., “Computer Science Is Neither Mathematics nor Electrical En¬
gineering,” in Finerman, Aaron, ed., University Education in Computing Sci¬
ence, proceedings of a conference on Graduate Academic and Related Programs
in Computing Science, held [at the State University of New York, Stony Brook]
June 1967 (New York: Academic Press, 1968), pp. 69-79.
Perlis, Alan J., “Current Research Frontiers in Computer Science,” in Dertouzos,
Michael L., and Moses, Joel, eds., The Computer Age: A Twenty-Year View
(Cambridge, Mass.: MIT Press, 1979), pp. 422-436.
Phillips, Almarin, “The Tableau Economique as a Simple Leontief Model,” in
Quarterly Journal of Economics, vol. 69 (Feb. 1955), pp. 137-144.
Phillips, Dennis C., Holistic Thought in Social Science (Stanford, Calif.: Stanford
University Press, 1976).
Piaget, Jean, Genetic Epistemology, trans. Eleanor Duckworth (New York: Colum¬
bia University Press, 1970).
Piaget, Jean,Biology and Knowledge: An Essay on the Relations between Organic
Regulations and Cognitive Processes, trans. Beatrix Walsh (Edinburgh, Scot¬
land: Edinburgh University Press, 1971).
Piaget, Jean, “The Psychogenesis of Knowledge and Its Epistemological
Significance,” in Piatelli-Palmarini, Massimo, ed., Language and Learning: The
Debate between Jean Piaget and Noam Chomsky (Cambridge, Mass.: Harvard
University Press, 1980), pp. 23-34.
Language and Learning: The Debate between
Piatelli-Palmarini, Massimo, ed.,
Jean Piaget and Noam Chomsky (Cambridge, Mass.: Harvard University Press,
1980).
Picavet, Francois Joseph, Les ideologues: essai sur I’histoire des idees et des
theories scientifiques, philosophiques, religieuses, etc. en France depuis 1789
(Paris: F. Alcan, 1891; reprint. New York: Burt Franklin, 1971).
Picci, G., “Stochastic Realization of Guassian Processes,” in Proceedings of the
IEEE, vol. 64 (1976), pp. 112-122.
Picci, G., “Some Connections Between the Theory of Sufficient Statistics and the
Identifiability Problem,” in SIAM Journal of Applied Mathematics, vol. 33
(1977), pp. 383-398.
Pierce, John R.,Symbols, Signals, and Noise: The Nature and Process of Com¬
munication (New York: Harper, 1961; Torchbook ed., 1965).
Pinsker, M. S., Information and Information Stability of Random Variables and
Processes, trans. Amiel Feinstein (San Francisco: Holden-Day, 1964).
Pitts, Walter H., and McCulloch, Warren S., “How We Know Universals: The
Perception of Auditory and Visual Forms,” in Bulletin of Mathematical
Biophysics, vol. 9 (1947), pp. 127-147.
Plato, The Republic (Glasgow, Scotland: 1st ed. in English, 1763; another ed.,
Oxford: Oxford University Press, 1955).
714 CUMULATIVE LIST OF REFERENCES

Polanyi, Michael, Personal Knowledge (Chicago: University of Chicago Press,

1958).
Pollack, Irwin, “The Information of Elementary Auditory Displays,” in Journal of
the Acoustical Society of America, vol. 25 (July 1953), pp. 765-769.
PONTRYAGIN, L. S.J BOLTYANSKIL, V. G.; GaMRELIDZE, R. V.; AND MISHCHENKO,
E. F., The Mathematical Theory of Optimal Processes (New York: Wiley,
1962).
Poole, William Frederick, An Index to Periodical Literature (Boston: 3rd ed., J. R.
Osgood, 1882; 1st ed. 1848; reprinted with five supplements covering the litera¬
ture from 1802-1906, Gloucester, Mass.: Peter Smith, 1963).
Poore, Jesse H.; Baralt-Torrijos, Jorge; and Chiaraviglio, Lucio, On the Com¬
binatory Definability of Hardware and Software (Atlanta: Georgia Institute of
Technology, 1971). Research Report GITIS-71-04.
Popov, V. M., “Invariant Description of Linear Time-Invariant Controllable Sys¬
tems,” in SIAM Journal of Control, vol. 10 (1972), pp. 252-264.
Popov, V. M., Hyperstability of Control Systems (New York: Springer-Verlag, 1973;
Roumanian ed., Bucharest, 1966).
Popper, Karl R., The Logic of Scientific Discovery (London: Hutchinson, 1959;
Torchbooks, 1965), trans. author with the assistance of Julius Freed and Lan
Freed, from Logik der Forschung: Zur Erkenntnistheorie der modernen Natur-
wissenschaft (Vienna, 1935).
Popper, Karl R., Conjectures and Refutations: The Growth of Scientific Knowledge
(New York: Basic Books, 1962).
Popper, Karl R., The Open Society and its Enemies, vol. 2 (Princeton, N.J.: Prince¬
ton University Press, 5th ed., 1966).
Popper, Karl R., “Epistemology Without a Knowing Subject,” in Popper, Karl R.,
Objective Knowledge: An Evolutionary Approach (Oxford: Clarendon Press,
1972a), pp. 106-152.

Popper, Karl R., Objective Knowledge: An Evolutionary Approach (Oxford: Claren¬

don Press, 1972ft; rev. ed., Oxford University Press, 1979).
Popper, Karl R., and Eccles, John C., The Self and Its Brain (Berlin and New York:
Springer International, 1977).
Premack, David, Intelligence in Ape and Man (Hillsdale, N.J.: Lawrence Erlbaum,
1976).

Prigogine, Ilya, “Unity of Physical Laws and Levels of Description,” in Grene,

Marjorie, ed., Interpretations of Life and Mind (New York: Humanities Press,
1971).

Problemy Peredachi Informatsii, published since 1965 in Moscow in Russian. En¬

glish trans., Problems of Information Transmission, pub. Consultant’s Bureau,
227 West 17th Street, New York, N.Y. 10011.
Procaccia, I.; Shimoni, Y.; and Levine, R. D., “Entropy and Microscopic Dis¬
equilibrium: Isothermal Time Evolution with Application to Vibrational Relax¬
ation,” in Journal of Chemical Physics, vol. 65 (Oct. 1976), pp. 3284-3301.
Proceedings of the First International Symposium on Information Theory, held [in
London] Sept. 1950, sponsored by the American Institute of Radio Engineers.
Lithoprinted, 1953.
CUMULATIVE LIST OF REFERENCES 715

Putnam, Hilary, “Minds and Machines,” in Hook, Sidney, ed., Dimensions of

Mind [a symposium. New York University Institute of Philosophy, 1959] (New
York: New York University Press, 1960).
Pylyshyn, Zenon W., ed.. Perspectives on the Computer Revolution (Englewood
Cliffs, N.J.: Prentice-Hall, 1970).
Pylyshyn, Zenon W., “Computational Models and Empirical Constraints” [with
peer commentaries], in The Behavioral and Brain Sciences, vol. 1 (Mar. 1978),
pp. 93-127.
Pylyshyn, Zenon W., “Complexity and the Study of Artificial and Human Intelli¬
gence,” in Ringle, Martin, ed.. Philosophical Perspectives in Artificial Intelli¬
gence (New York: Humanities Press, 1979), pp. 23-56.
Pylyshyn, Zenon W., “Computation and Cognition: Issues in the Foundations of
Cognitive Science,” in The Behavioral and Brain Sciences, vol. 3 (Mar. 1980),
pp. 111-169.
Pylyshyn, Zenon W., Computation and Cognition: Issues in the Foundations of
Cognitive Science (Cambridge, Mass.: Bradford Books/MIT Press, forthcoming
1984).
Quastler, Henry, ed., Information Theory in Psychology: Problems and Methods,
proceedings of a conference on the Estimation of Information Flow, held [in
Monticello, Ill.] July 5-9, 1954 (Glencoe, Ill.: Free Press, 1955).
Quillian, M. Ross, “Semantic Memory,” in Minsky, Marvin L., ed., Semantic
Information Processing (Cambridge, Mass.: MIT Press, 1968), pp. 216-270.
Quine, Willard van Orman, “The Problem of Meaning in Linguistics,” in Quine,
Willard van Orman, From a Logical Point of View (Cambridge, Mass.: Harvard
University Press, 1953).
Rall, Wilfred, “Theoretical Significance of Dendritic Trees for Neuronal Input-
Output Relations,” in Reiss, Richard F., et al., eds., Neural Theory and Model¬
ing, proceedings of the 1962 Ojai Symposium, held [in Ojai, Calif.] Dec. 4-6,
1962 (Stanford, Calif.: Stanford University Press, 1964), pp. 73-97.
Ralston, Anthony, Introduction to Programming and Computer Science (New
York: McGraw-Hill, 1971).
Ramo, Simon, Cure for Chaos: Fresh Solutions to Social Problems through the
Systems Approach (New York: D. McKay, 1969).
Raphael, Bertram, “The Relevance of Robot Research to Artificial Intelligence,” in
Banerji, Ranan B., and Mesarovic, Mihajlo D., eds.. Theoretical Approaches to
Non-Numerical Problem Solving, proceedings of the Fourth Systems Sym¬
posium [at Case Western Reserve University] 1968 (New York: Springer-
Verlag, 1970), pp. 455-466.
Rapoport, Anatol, Operational Philosophy: Integrating Knowledge and Action
(New York: Harper and Row, 1953; 2nd printing, Wiley, 1967).
Rapoport, Anatol, “Foreword,” in Buckley, Walter, ed., Modern Systems Re¬
search for the Behavioral Scientist (Chicago: Aldine-Atherton, 1968a).
Rapoport, Anatol, “Systems Analysis—General Systems Theory, in Sills, David
L., ed.. International Encyclopedia of the Social Sciences (New York: Macmil¬
lan and Free Press, 1968b), vol. 15, pp. 452-458.
Rapoport, Anatol, “General Systems Theory: A Bridge Between Two Cultures
 716 CUMULATIVE LIST OF REFERENCES

[Third Annual Ludwig von Bertalanffy Memorial Lecture], in Report of the

SGSR [Society for General Systems Research] Annual Meeting, Feb. 18-21,
1976, pp. 9-16.

Ratliff, Floyd, Mach Bands: Quantitative Studies on Neural Networks in the Ret¬
ina (San Francisco: Holden-Day, 1965).
Rayward, W. Boyd, Systematic Bibliography in England: 1850-1895 (Urbana: Uni¬
versity of Illinois Graduate School of Library Science, 1967a). Occasional Paper
84.

Rayward, W. Boyd, “UDC and FID: A Historical Perspective,” in The Library

Quarterly, vol. 37 (July 1967(7), p. 259-278.
Rayward, W. Boyd, The Universe of Information: The Work of Paul Otletfor Docu¬
mentation and International Organization (Moscow: VINITI, 1975), Federation
Internationale de Documentation Publication No. 520.
Rayward, W. Boyd, “Subject Access to the Catalogue of Scientific Papers, 1800—
1900,” in Rayward, W. Boyd, ed., The Variety of Librarians hip: Essays in
Honour of John Wallace Metcalfe (Sydney: Library Association of Australia,
1976), pp. 146-170.

Rayward, W. Boyd, “Some Developments in Nineteenth-Century Bibliography:

Great Britain,” in Libri, vol. 27 (June 1977a), pp. 97-107.
Rayward, W. Boyd, “IFLA-FID: Is it Time for Federation?” in IFLA Journal, vol.
3, no. 3 (19776), pp. 278-280.

Rayward, W. Boyd, “International Library and Bibliographic Organization,” in

ALA World Encyclopedia of Library and Information Services (Chicago: Ameri¬
can Library Association, 1980a), pp. 264-268.

Rayward, W. Boyd, “The Problems of Professional Knowledge in Librarianship,”

in Changes and Exchanges; Australian Viewpoints (Sydney, Australia: Kuring-
gai College of Advanced Education, 19806), pp. 135-154.
Rayward, W. Boyd, “The Evolution of an International Library and Bibliographic
Community,” in Journal of Library History, vol. 16 (spring 1981a), pp. 449-462.
Rayward, W. Boyd, “The Perils of Bibliography: Four 19th-Century Experiences,”
in Wilson Library Bulletin, vol. 56 (Oct. 19816), pp. 110-115.

Reddy, D. Raj, Working Papers in Speech Recognition (IV): The HEARSAY II

System (Carnegie-Mellon University Computer Science Department Technical
Report, Feb. 1976).

Reilly, Park M., and Blau, G. E., “The Use of Statistical Methods to Build Mathe¬
matical Models of Chemical Reacting Systems,” in Canadian Journal of Chemi¬
cal Engineering, vol. 52 (June 1974), pp. 289-299.
Reitman, Walter R., Cognition and Thought: An Information-Processing Approach
(New York: Wiley, 1965).

Report of the Royal Commission Appointed to Inquire into the Constitution and
Government of the British Museum (London: His Majesty’s Stationery Office,
1850). Minutes of Evidence, Minute 9751.

Rescher, Nicholas, Cognitive Systematization: A System-Theoretic Approach to a

Coherentist Theory of Knowledge (Totowa, N.J.: Rowan and Littlefield, 1979).
Richardson, John, The Spirit of Inquiry in Library Science: The Graduate Library
CUMULATIVE LIST OF REFERENCES 717

School at Chicago, 1921-1951 (Chicago: American Library Association, 1982).

ACRL Monographs in Librarianship, no. 42.
Rinzel, John, “Integration and Propagation of Neuroelectric Signals,” in Levin,
Simon A., ed.. Studies in Mathematical Biology, Part 1: Cellular Behavior and
the Development of Pattern (Washington, D.C.: Mathematical Association of
America, 1978), pp. 1-66.
Rissanen, J., and Kailath, T., “Partial Realization of Random Systems,” in Auto-
matica, vol. 8 (1972), pp. 389-396.
Roberts, Lawrence G., “Machine Perception of Three-Dimensional Solids,” in Tip¬
pett, James T., et al., eds., Optical and Electro-Optical Information Processing.
Symposium on Optical and Electro-Optical Information Processing Technology
[Boston] 1964 (Cambridge, Mass.: MIT Press, 1965), pp. 159-197.
Robertson, George; McCracken, Daniel D.; and Newell, Allen, “The ZOG Ap¬
proach to Man-Machine Communication,” in International Journal of Man-
Machine Studies, vol. 14 (May 1981), pp. 461-488.
Robinson, Alan J., “A Machine-Oriented Logic Based on the Resolution Principle,”
in Journal of the Association for Computing Machinery, vol. 12 (Jan. 1965), pp.
23-41.
Rose, John, ed., Survey of Cybernetics—A Tribute to Dr. Norbert Wiener (London:
Iliffe Books, 1969).
Rosen, Saul, “Software,” in Encyclopedia of Computer Science (New York: Pet-
rocelli/Charter, 1976), pp. 1283-1285.
Rosenblatt, Frank, “The Perceptron: A Probabilistic Model for Information Stor¬
age and Organization in the Brain,” in Psychological Review, vol. 65 (Nov.
1958), pp. 386-408.
Rosenblatt, Frank, Principles of Neurodynamics: Perceptrons and the Theory of
Brain Mechanisms (Washington, D.C.: Spartan Books, 1962).
Rosenblueth, Arturo; Wiener, Norbert; and Bigelow, Julian, “Behavior, Pur¬
pose and Teleology,” in Philosophy of Science, vol. 10 (Jan. 1943), pp. 18-24.
Rosenbrock, H. H., State-Space and Multivariable Theory (London: Nelson and
Sons, 1970).
Rosenfeld, Azriel, and Kak, Avinash C., Digital Picture Processing (New York:
Academic Press, 1976).
Rothe, Carl F., “Regulations of Visceral Function—B. Homeostasis and Negative
Feedback-Control,” in Selkurt, Ewald E., ed., Physiology (Boston: Little,
Brown, 3rd ed., 1971).
Rouchaleau, Y., and Sontag, E., “On the Existence of Minimal Realizations of
Linear Dynamical Systems Over Neotherian Integral Domains,” in Journal of
Computer and System Sciences, vol. 18 (1979), pp. 65-75.
Ruch, Floyd Leon, Psychology and Life: A Study of the Thinking, Feeling, and
Doing of People (Glenview, Ill.: Scott, Foresman, 1937; 9th ed., rev., 1967).
Rugh, Wilson J., Mathematical Description of Linear Systems (New York: Marcel
Dekker, 1975).
Russell, Bertrand, Introduction to Mathematical Philosophy (New York: Simon
and Schuster, 1919).
 718 CUMULATIVE LIST OF REFERENCES

Russell, Bertrand, Human Knowledge: Its Scope and Limits (New York: Simon
and Schuster, 1948).

Russell, Bertrand, Bertrand Russell Speaks His Mind (London: Arthur Barker,
1960).

Ruyer, Raymond, La cybernetique et I’origine de I’information (Paris: Flammarion,

1954).

Sabloff, Jeremy A., Simulations in Archeology (Albuquerque: University of New

Mexico Press, 1981).

Sager, Naomi, “Information Structures in the Language of Science,” in Weiss,

Edward C., ed., The Many Faces of Information Science (Boulder, Colo.: West-
view Press, 1977), pp. 53-73.

Salton, Gerard, “Information Retrieval,” in Encyclopedia of Computer Science

(New York: Petrocelli/Charter, 1976), pp. 649-656.
Samuel, Arthur L., “Some Studies in Machine Learning Using the Game of Check¬
ers,” in IBM Journal of Research and Development, vol. 3 (July 1959), pp 210-
229.

Samuelson, Paul A., Foundations of Economic Analysis (Cambridge, Mass.: Har¬

vard University Press, 1948).

Samuelson, Paul A., Economics (New York: McGraw-Hill, 11th ed., 1980).
Saracevic, Tefko, ed., Introduction to Information Science (New York: R R Bow-
ker, 1970).

Sattinger, D. H., “Bifurcation and Symmetry Breaking in Applied Mathematics,”

in Bulletin (New Series) of the American Mathematical Society, vol. 3 (Sept
1980), pp. 779-819.

Saussure, Ferdinand de, Cours de linguistique generate (Lausanne and Paris, 1916).
English trans., Course in General Linguistics (New York: Philosophical Li¬
brary, 1959).

Schank, Roger C., “Identification of Conceptualizations Underlying Natural Lan¬

guage,” in Schank, Roger C., and Colby, Kenneth M., eds., Computer Models
of Thought and Language (San Francisco: W. H. Freeman, 1973), pp. 187-247.
Schank, Roger C., Dynamic Memory: A Theory of Reminding and Learning in
Computers and People (Cambridge: Cambridge University Press, 1982).
Schank, Roger, and Abelson, Robert P., Scripts, Plans, Goals, and Understand¬
ing: An Inquiry into Human Knowledge Structures (Hillsdale, N.J.: Lawrence
Erlbaum, 1977).

Schank, Roger C., and Nash-Webber, Bonnie, eds., Theoretical Issues in Natural
Language Processing, proceedings of the first conference on Theoretical Issues
in Natural Language Processing (prepared by Bolt, Beranek and Newman, Inc.,
Cambridge, Mass., 1975).

Schank, Roger C., and Wilensky, Robert, “Response to Dresher and Hornstein,”
in Cognition, vol. 5 (June 1977), pp. 133-145.

Scheibf.l, Madge E., and Scheibel, Arnold B., “Structural Substrates for Integra¬
tive Patterns in the Brain Stem Reticular Core,” in Jasper, H. H., et al., eds.,
Reticular Formation of the Brain (Boston: Little, Brown, and Co., 1958) pp
31-68.
CUMULATIVE LIST OF REFERENCES 719

Schell, Jonathan, “The Fate of Mankind,” in The New Yorker, Feb. 1, 1982, pp. 8,
15.
Schmidt, RobertF., “Somatovisceral Sensibility,” Section 3 of Schmidt, Robert F.,
ed.. Fundamentals of Sensory Physiology, trans. Marguerite A. Biederman-
Thorson (New York: Springer-Verlag, 1978; 2nd, corrected ed., 1981), pp. 81-
125.
Schrader, Alvin M., “In Search of a Name: Information Science and Its Conceptual
Antecedents,” in Library and Information Science Research, vol. 6 (spring
1984).
Schrodinger, Erwin, What is Life? The Physical Aspect of the Living Cell (Cam¬
bridge, England: The University Press; New York: Macmillan, 1945).
Schultz, Claire, K., and Garwig, Paul L., “History of the American Documenta¬
tion Institute—A Sketch,” in American Documentation, vol. 20 (1969), pp. 152—
160.
Schumpeter, JosephA., The Theory of Economic Development, trans. Redvers Opie
(Cambridge, Mass.: Harvard University Press, 1934).
Schutz, Alfred, and Luckmann, Thomas, The Structures of the Life-World, trans.
Richard M. Zaner and A. Tristram Engelhardt, Jr. (Evanston, Ill.: Northwestern
University Press, 1973).
Science, “Research News,” in Science, vol. 212 (1981), p. 313
“Research News,” in Science, vol. 214 (1981), p. 1334
“Research News,” in Science, vol. 216 (1982), p. 969
“Research News,” in Science, vol. 217 (1982), p. 1239
Searle, John R., “Minds, Brains, and Programs” [with peer commentaries], in The
Behavioral and Brain Sciences, vol. 3 (Sept. 1980), pp. 417-457.
Searle, John R., “Minds, Brains, and Programs,” in Hofstadter, Douglas R., and
Dennett, Daniel C., eds., The Mind's I: Fantasies and Reflections on Self and
Soul (New York: Basic Books, 1981), pp. 353-373.
Searle, John R., “The Myth of the Computer.” Book review of Hofstadter, Douglas
R., and Dennett, Daniel C., eds., The Mind’s I: Reflections on Self and Soul
[New York: Basic Books, 1981] in The New York Review of Books, vol. 29 (29
Apr. 1982a), pp. 3-6.
Searle, John R., “Response to Dennett” in the letters section of The New York
Review of Books, June 24, 1982b.
Sebeok, Thomas A., Contributions to the Doctrine of Signs (Bloomington: Indiana
University and Peter de Ridder Press, 1976).
Seiffert, Helmut, Information liber die Information; Verstandigung im Alltag—
Nachrichtentechnik—Wissenschaftliches Verstehen—Informationssoziologie—
Das Wissen des Gelehrten (Mtinchen: C. H. Beck, 1968).
Sellen, Betty-Carol, ed., What Else You Can Do With A Library Degree (Syra¬
cuse, New York: Gaylord, 1980).
Shannon, Claude E., “A Mathematical Theory of Communication,” in Bell System
Technical Journal, vol. 27 (July and Oct. 1948), pp. 379—423; 623-656.
Shannon, Claude E., “Communication Theory of Secrecy Systems,” in Bell Sys¬
tem Technical Journal, vol. 28 (1949), pp. 656—715.
720 CUMULATIVE LIST OF REFERENCES

Shannon, Claude E., Contribution to the discussion, in Foerster, Heinz M. von,

ed., Cybernetics, transactions of the Eighth Conference sponsored by the Josiah
Macy Jr. Foundation (New York: Josiah Macy Jr. Foundation, 1951), p. 219.
Shannon, Claude E., and McCarthy, John, eds., Automata Studies (Princeton,
N.J.: Princeton University Press, 1956).
Shannon, Claude E., and Weaver, Warren, The Mathematical Theory of Com¬
munication (Urbana: University of Illinois Press, 1949).
Shelley, Mary Wollstonecraft, Frankenstein; or The Modern Prometheus (1818
text), ed. James Rieger (Indianapolis: Bobbs-Merrill, 1974).
Shelly, Gary B., and Cashman, Thomas J., Introduction to Computers and Data
Processing (Fullerton, Calif.: Anaheim, 1980).
Shera, Jesse H., “Emergence of a New Institutional Structure for the Dissemination
of Specialized Information,” in American Documentation, vol. 4 (1953), pp.
163-173.
Shera, JesseH., “Of Librarianship, Documentation, and Information Science,” in
UNESCO Bulletin for Libraries, vol. 22 (Mar./Apr. 1968), pp. 58-65.
Shera, Jesse H., and Cleveland, Donald B., “History and Foundations of Informa¬
tion Science,” in Williams, Martha E., ed., Annual Review of Information Sci¬
ence and Technology (White Plains, N.Y.: Knowledge Industries Publications
[for the American Society for Information Science], 1977), vol. 12, pp. 249-275.
Shera, Jesse H., and Egan, Margaret E., “Introduction,” in Bradford, Samuel
Clement, Documentation (London: Crosby Lockwood, 2nd ed., 1953; 1st ed
1948).
Shera, Jesse H., and Egan, Margaret E., Bibliographic Organization, papers pre¬
sented before the fifteenth annual Conference of the Graduate Library School,
1950 (Chicago: University of Chicago Press, 1951).
Shortliffe, Edward H., “MYCIN: A Rule-Based Computer Program for Advising
Physicians Regarding Antimicrobial Therapy Selection” (Ph.D. diss., Stanford
University, 1974).
Shortliffe, Edward H., Computer-Based Medical Consultations: MYCIN (New
York: American Elsevier, 1976).
Silverman, L., “Realization of Linear Dynamical Systems,” in IEEE Transactions
on Automatic Control, vol. AC-16 (1971), pp. 554-567.
Simon, Herbert A., Models of Man, Social and Rational: Mathematical Essays on
Rational Human Behavior in a Social Setting (New York: Wiley, 1957).
Simon, Herbert A., “Modelling Human Mental Processes,” in Extending Man’s
Intellect, proceedings of the Western Joint Computer Conference, held [in Los
Angeles] by the Institute of Radio Engineers, the American Institute of Electri¬
cal Engineers, and the Association for Computing Machinery, May 9-11, 1961
vol. 19, pp. 111-120.
Simon, Herbert A., The Sciences of the Artificial [the Karl Taylor Compton Lec¬
ture, 1968] (Cambridge, Mass.: MIT Press, 1969; 2nd enlarged ed., 1981).
Simon, Herbert A., “Information Processing,” in Encyclopedia of Computer Sci¬
ence (New York: Petrocelli/Charter, 1976), pp. 647-649.
Simon, Herbert A., “Information Processing Models of Cognition,” in Annual Re¬
view of Psychology, vol. 30 (1979), pp. 363-396.
CUMULATIVE LIST OF REFERENCES 721

Simon, Herbert A., “Cognitive Science: The Newest Science of the Artificial,” in
Cognitive Science, vol. 4 (Jan./Mar. 1980), pp. 33-46.
Simon, Herbert A., “Studying Human Intelligence by Creating Artificial Intelli¬
gence,” in American Scientist, vol. 69 (May-June, 1981), pp. 300-309.
Simon, Herbert A., and Feigenbaum, Edward A., “An Information-Processing
Theory of Some Effects of Similarity, Familiarization, and Meaningfulness in
Verbal Learning,” in Journal of Verbal Learning and Verbal Behavior, vol. 3
(Oct. 1964), pp. 385-396.
Simon, Herbert A., and Kotovsky, Kenneth, “Human Acquisition of Concepts for
Sequential Patterns,” in Psychological Review, vol. 70 (Nov. 1963), pp. 534-
546.
Singer, Edgar A., Jr., In Search of a Way of Life (New York: Columbia University
Press, 1948).
Skinner, Burrhus F., The Behavior of Organisms (New York: Appleton-Century-
Crofts, 1938).
Skinner, Burrhus F., Verbal Behavior (New York: Appleton-Century-Crofts, 1957).
Slagle, JamesR., “A Heuristic Program that Solves Symbolic Integration Problems
in Freshman Calculus,” in Feigenbaum, Edward A., and Feldman, Julian, eds.,
Computers and Thought (New York: McGraw-Hill, 1963), pp. 191-203.
Slagle, JamesR.; Dixon, J. K.; and Jones, T. L., “List Processing Languages,” in
Encyclopedia of Computer Science (New York: Petrocelli/Charter, 1976), pp.
778-785.
Slamecka, Vladimir, “The Science and Engineering of Information,” in Finerman,
Aaron, ed.. University Education in Computing Science, proceedings of a con¬
ference on Graduate Academic and Related Research Programs in Computing
Science, held [at Stony Brook] by the State University of New York, 1967 (New
York: Academic Press, 1968), pp. 81-92.
Slamecka, Vladimir, and Gehl, John, eds., “Information Sciences at Georgia Insti¬
tute of Technology: The Formative Years 1963-1978,” in Information Process¬
ing and Management, vol. 14, no. 5 (1978), pp. 320-361.
Slamecka, Vladimir, and Pearson, Charls, “Information Science,” in Encyclope¬
dia of Computer Science (New York: Petrocelli/Charter, 1976), pp. 656-657.
Slamecka, Vladimir, and Pearson, Charls, “The Portent of Signs and Symbols,” in
Weiss, Edward C., ed., The Many Faces of Information Science (Boulder,
Colo.: Westview Press, 1977), pp. 105-128.
Slater, John C., Introduction to Chemical Physics (New York: McGraw-Hill, 1st
ed. 1938).
Sleeman, Derek E., and Brown, John Seely, eds., Intelligent Tutoring Systems
(New York: Academic Press, 1982).
Slepian, David, ed., Key Papers in the Development of Information Theory (New
York: IEEE Press, 1974).
Sloan Foundation, Advisors of the Alfred P., “Cognitive Science 1978.” Report
of the State of the Art Committee (New York: Alfred P. Sloan Foundation, Oct.
1978).
Slobin, Dan I., Psycholinguistics (Glenview, Ill.: Scott, Foresman, and Co., 1971;
2nd ed., 1979).
722 CUMULATIVE LIST OF REFERENCES

Sloman, Aaron, The Computer Revolution in Philosophy: Philosophy, Science, and

Models of Mind (New York: Humanities Press, 1978a).
Sloman, Aaron, “Intuition and Analogical Reasoning,” in Sloman, Aaron, The
Computer Revolution in Philosophy: Philosophy, Science, and Models of Mind
(Hassocks, England: Harvester Press, 19786), pp. 144-176.
Small, Henry, and Griffith, Belver C., “The Structure of Scientific Literatures. I:
Identifying and Graphing Specialties,” in Science Studies, vol. 4 (Jan. 1974), pp.
17-40.
Smets, Philippe, “Probability of a Fuzzy Event: An Axiomatic Approach,” in Fuzzy
Sets and Systems, vol. 7 (1982), pp. 153-164.
Smith, Adam, An Enquiry into the Nature and Causes of the Wealth of Nations
(London: W. Strahan and T. Cadell, 1776; another ed., New York: Modern
Library, 1936).
Smith, Reid G., “A Framework for Problem Solving in a Distributed Processing
Environment” (Ph.D. diss., Stanford University, 1978).
Snow, Charles P., The Two Cultures and the Scientific Revolution [the Rede Lec¬
ture, 1959] (Cambridge, England: The University Press, 1959).
Snow, Charles P., The Two Cultures and a Second Look. An Expanded Version of
the Two Cultures and the Scientific Revolution (Cambridge, England: The Uni¬
versity Press, 1964).
Solomonoff, R. J., “A Formal Theory of Inductive Inference,” in Information and
Control, vol. 7 (Mar. 1964), pp. 1-22.
Sontag, Eduardo D., Polynomial Response Maps (Berlin and New York: Springer-
Verlag, 1979).
Sonuga, J. O., “Entropy Principle Applied to the Rainfall-Runoff Process,” in Jour¬
nal of Hydrology, vol. 30 (May 1976), pp. 81-94.
Sperling, George, “The Information Available in Brief Visual Presentations,” in
Psychological Monographs no. 11, vol. 74 (1960).
Sperry, Roger W., “Mind, Brain, and Humanist Values,” in Platt, John R., ed.,
New Views of the Nature of Man (Chicago: University of Chicago Press, 1965),
pp. 71-92.
Sperry, Roger W., “Mental Unity Following Surgical Disconnection of the Cerebral
Hemispheres” [Harvey Lectures], vol. 62 (1968), p. 293-322.
Sperry, Roger W., “Some Effects of Disconnecting the Cerebral Hemispheres”
[Nobel Prize Lecture, 1981], Science, vol. 217 (24 Sept. 1982), pp. 1223-1226.
Spinoza, Benedictus de, On the Improvement of the Understanding (New York:
Dutton, 1963; 1st English trans., 1883).
Sridharan, N., Prototype Deformation (Rutgers University Computer Science De¬
partment Technical Report, New Brunswick, N.J., 1980).
Stakgold, Ivar, “Branching of Solutions of Nonlinear Equations,” in SIAM Re¬
view, vol. 13 (July 1971), pp. 289-332.
Stanley-Jones, Douglas, “Cybernetics and General Systems—A Unitary Sci¬
ence?” in Kybernetes, vol. 8 (1979), pp. 25-32.
Stark, Lawrence, “The Control System for Versional Eye Movements,” in Bach-y-
Rita, Paul, and Collins, Carter C., eds., The Control of Eye Movements, pro-
CUMULATIVE LIST OF REFERENCES 723

ceedings of a Symposium on the Control of Eye Movements, held [in San Fran¬
cisco] Nov. 10-11, 1969 (New York: Academic Press, 1971), pp. 363-428.
After Babel: Aspects of Language and Translation (New York and
Steiner, George,
London: Oxford University Press, 1975).
Sternberg, Saul, “High-Speed Scanning in Human Memory,” in Science, vol. 153
(5 Aug., 1966), pp. 652-654.
Sternberg, Stan R., “Language and Architecture for Parallel Image Processing,” in
Grisema, E. S., and Kanal, L. N., eds., Proceedings of the Conference on
Pattern Recognition in Practice (Amsterdam: North-Holland, 1980).
Stigler, George J., “The Economics of Information,” in Journal of Political Econ¬
omy, vol. 69 (1961), pp. 213-225.
Stirling, James, Methodus differentials sive tractatus de summatione et inter¬
polation serierum infinitarum (London: 1730; another ed., London, 1764).
Strachey, Christopher, “Computer,” in Bullock, Alan, and Stallybrass, Oliver,
eds., The Harper Dictionary of Modern Thought (New York: 1st U.S. ed.,
Harper and Row, 1977), p. 122.
Strachey, Christopher, “Computing Science,” in Bullock, Alan, and Stallybrass,
Oliver, eds., The Harper Dictionary of Modern Thought (New York: 1st U.S.
ed., Harper and Row, 1977), p. 124.
Sussman, Gerald Jay, A Computer Model of Skill Acquisition (New York: American
Elsevier, 1975).
Sutherland, Ivan E., “SKETCHPAD: A Man-Machine Graphical Communication
System,” in AFIPS Conference Proceedings, Vol. 23: 1963 Spring Joint Com¬
puter Conference, organized by the American Federation of Information Pro¬
cessing Societies and held [in Detroit] May 1963 (Washington, D.C.: Spartan
Books, 1963), pp. 329-346.
Sutton, Richard S., and Barto, Andrew G., “Towards a Modern Theory of Adap¬
tive Networks: Expectation and Prediction,” in Psychological Review, vol. 88
(1981), pp. 135-170.
Swank, Raynard C., A Unifying Influence (Metuchen, N.J.: Scarecrow Press,
1981).
Swanson, Don R., “Introduction,” in Intellectual Foundations of Library Educa¬
tion, proceedings of the twenty-ninth annual Conference of the Graduate Li¬
brary School (Chicago: University of Chicago Press, 1965), p. 2.
Swanson, Don R., “Introduction to the Conference,” in The Role of Libraries in the
Growth of Knowledge, proceedings of the fortieth Conference of the Graduate
Library School (Chicago: University of Chicago Press, 1980a).
Swanson, Don R., “Libraries and the Growth of Knowledge,” in The Role of Librar¬
ies in the Growth of Knowledge, proceedings of the fortieth Conference of the
Graduate Library School (Chicago: University of Chicago Press, 19806), pp.
112-134.
Sweeney, James L., “The Economics of Depletable Resources: Market Forces and
Intertemporal Bias,” in Review of Economic Studies, vol. 64 (Feb. 1977), pp.
125-142.
Szentagothai, Janos, and Arbib, Michael A., “Conceptual Models of Neural Or-
724 CUMULATIVE LIST OF REFERENCES

ganization,” in Neurosciences Program Research Bulletin, vol. 12, no. 3 (1974),

pp. 310-479.
Szilard, Leo, “Uber die Entropieverminderung in einem thermodynamischen Sys¬
tem bei Eingriffen intelligenter Wesen,” in Zeitschrift fur Physik, vol. 53 (1929),
pp. 840-856.
Tannenbaum, A., Invariance and System Theory: Algebraic and Geometric Aspects,
Lecture Notes in Mathematics No. 845 (Berlin, Germany: Springer-Verlag,
1981).
Tate, Vernon D., “Introducing American Documentation,” in American Documen¬
tation, vol. 1 (Jan. 1950), pp. 3-7.
Taube, Mortimer, Computers and Common Sense: The Myth of Thinking Machines
(New York: Columbia University Press, 1961).
Taylor, Robert S., “Professional Aspects of Information Science and Technology,”
in Cuadra, Carlos A., ed., Annual Review of Information Science and Technol¬
ogy (New York: Wiley [for the American Documentation Institute], 1966), vol.
1, pp. 15-40.
Teichroew, D., “Information Systems,” in Encyclopedia of Computer Science
(New York: Petrocelli/Charter, 1978), pp. 657-660.
Terman, Lewis Madison, The Measurement of Intelligence: An Explanation of and a
Complete Guide for the Use of the Stanford Revision and Extension of the Binet-
Simon Intelligence Scale (Boston and New York: Houghton Mifflin, 1916).
Terrace, Herbert S.; Petitto, L. A.; Sanders, R. J.; and Bever, Thomas G., “Can
an Ape Create a Sentence?” in Science, vol. 206 (1979), p. 891.
Thom, Rene, Stabilite structurelle et morphogenese (Reading, Mass.: Addison-
Wesley, 1972; trans. as Structural Stability and Morphogenesis, 1975).
Thom, Rene, Modeles mathematiques de la morphogenese, series 10/18 (Paris:
Union Generate d’Editions, 1974).
Thom, Rene, Structural Stability and Morphogenesis: An Outline of a General
Theory of Models, trans. D. H. Fowler (Reading, Mass.: 1st English ed., W. A.
Benjamin, 1975).
Thom, Rene, “Structural Stability, Catastrophe Theory, and Applied Mathematics”
[the John von Neumann lecture, 1976], in SIAM Review, vol. 19 (Apr. 1977), pp.
189-201.
Thom, Rene, “Systemic versus Morphological Approach in General Systems
Theory,” in Pichler, Franz R., and Hanika, Francis de Paula, eds., Progress in
Cybernetics and Systems Research, Vol. VII: General Systems Methodology,
Organization and Management, Cognition and Learning, papers presented at a
symposium organized by the Austrian Society for Cybernetic Studies and held
[in Vienna] Apr. 1974 (Washington, D.C.: Hemisphere Publishing Company,
1980) .
Tinbergen, Jan, “The Use of Models: Experience and Prospects” [Nobel Prize
lecture delivered Dec. 1969], in The American Economic Review, vol. 71 (Dec.
1981) , pp. 17-22.
Torda, Clara, Information Processing by the Central Nervous System and the
Computer (A Comparison) (Berkeley, Calif.: Walters, 1982).
Trappl, Robert, and Hanika, Francis de Paula, eds., Progress in Cybernetics and
CUMULATIVE LIST OF REFERENCES 725

Systems Research, Vol. II: Socio-Economic Systems, Cognition and Learning,

Systems Education, Organization and Management, papers presented at a sym¬
posium organized by the Austrian Society for Cybernetic Studies and held [in
Vienna] Apr. 1974 (New York: Halsted Press, 1975).
Tremblay, Jean-Paul, and Sorenson, Paul G., An Introduction to Data Structures
with Applications (New York: McGraw-Hill, 1976).
Tribe, Laurence H., “Policy Science: Analysis or Ideology?” in Philosophy and
Public Affairs, vol. 2 (fall 1972), pp. 66-110.
Tribus, Myron, “Thermodynamics: A Survey of the Field,” in Recent Advances in
the Engineering Sciences: Their Impact on Engineering Education, proceedings
of a conference on Science and Technology for Deans of Engineering, Purdue
University, Lafayette, Ind., Sept. 9-12, 1957 (New York: McGraw-Hill, 1958).
Tribus, Myron, Thermostatics and Thermodynamics: An Introduction to Energy,
Information, and States of Matter, with Engineering Applications (Princeton,
N.J.: Van Nostrand, 1961a).
Tribus, Myron, “Information Theory as the Basis for Thermostatics and Thermody¬
namics,” in Journal of Applied Mechanics, ASME Transactions, series E, no.
83 (Mar. 19616), pp. 1-8.
Tribus, Myron, “The Use of the Maximum Entropy Estimate in Reliability En¬
gineering,” in Machol, Robert E., and Gray, Paul, edsRecent Developments in
Information and Decision Processes, proceedings of the Third Symposium on
Information and Decision Processes, held [at Purdue University] 1961 (New
York: Macmillan, 1962).
Tribus, Myron, Rational Descriptions, Decisions, and Designs (New York: Perga-
mon Press, 1969).
Tribus, Myron; Evans, Robert; and Crellin, Gary L., “The Use of Entropy in
Hypothesis Testing,” in Proceedings of the Tenth National Symposium on Reli¬
ability and Quality Control, Jan. 7-9, 1964.
Turchin, Valentin Fedorovich,The Phenomenon of Science, trans. Brand Frentz
(New York: Columbia University Press, 1977).
Turing, Alan M., “On Computable Numbers, with an Application to the Ent-
scheidungsproblem,” in Proceedings of the London Mathematical Society, vol.
42, no. 2 (1936), pp. 230-265.
Turing, Alan M., “Computing Machinery and Intelligence,” in Mind, vol. 59 (Oct.
1950), pp. 433-460.
Turing, Alan M., “The Chemical Basis of Morphogenesis,” in Philosophical Trans¬
actions of the Royal Society of London, vol. 237 (Aug. 1952), pp. 37-72.
Turnbull, George Henry, Hartlib, Drury and Comenius: Gleanings from Hartlib’s
Papers (Liverpool, England: University Press of Liverpool, 1947).
Tyldesley, John R., “A Thermodynamic Approach to Turbulence Phenomena”
private communication, Dept, of Mechanical Engineering, Glasgow University
(1962).
Uhr, Leonard M., “Computer Arrays and Networks” (1982).
Van der Waals, Johannes, D., “Uber die Erklarung der Naturgesetze auf statis-
tisch-mechanischer Grundlage,” in Physikalische Zeitschrift, no. 13 (July 1911),
pp. 547-549.
726 CUMULATIVE LIST OF REFERENCES

Van Gigch, John P., and Hill, Richard E., Using Systems Analysis to Implement
Cost-Effectiveness and Program Budgeting in Education (Englewood Cliffs,
N.J.: Educational Technology Publications, 1971).
Van Marlen, Geert, and Dijkstra, Auke, “Information Theory Applied to Selec¬
tion of Peaks for Retrieval of Mass Spectra,” in Analytical Chemistry, vol. 48
(1976), pp. 595-598.
Van Putten, C., and Van Schuppen, J. H., “On Stochastic Dynamical Systems,” in
Proceedings of the Fourth International Symposium on the Mathematical
Theory of Networks and Systems (Delft, Netherlands: North-Holland, 1979).
von Bertalanffy, Ludwig [see “Bertalanffy, Ludwig von”].
von Neumann, John [see “Neumann, John von”].
Wang, Hao, “Toward Mechanical Mathematics,” in IBM Journal of Research and
Development, vol. 4 (Jan. 1960), pp. 2-22.
Waples, Douglas, “The Graduate Library School at Chicago,” in The Library Quar¬
terly, vol. 1 (Jan. 1931), pp. 26-36.
Waterman, Donald A. “Generalization Learning Techniques for Automating the
Learning of Heuristics,” in Artificial Intelligence, vol. 1 (spring 1970), pp. 121—
170.
Waterman, Donald A., and Hayes-Roth, Frederick, eds., Pattern-Directed Infer¬
ence Systems. Report of a Workshop on P-D I Systems [Honolulu] 1977 (New
York: Academic Press, 1978).
Waterman, Talbot H., “Systems Theory and Biology—View of a Biologist,” in
Mesarovic, Mihajlo D., ed., Systems Theory and Biology, proceedings of the
Third Systems Symposium at Case Institute of Technology (New York: Sprin-
ger-Verlag, 1968), pp. 1-37.
Watson, James D., Molecular Biology of the Gene (New York: Benjamin, 1965).
Watson, John B., “Psychology as the Behaviorist Views It,” in Psychological Re¬
view, vol. 20 (Mar. 1913), pp. 158-177.
Weaver, Warren, “Science and Complexity,” in The American Scientist, vol. 36
(Oct. 1948), pp. 536-544.
Weaver, Warren, “The Mathematics of Information,” in Automatic Control (New
York: Simon and Schuster, 1955), pp. 100-104.
Webb, Eugene J.; Campbell, Donald T.; Schwartz, Richard D.; and Sechrest,
Lee, Unobtrusive Measures: Nonreactive Research in the Social Sciences
(Chicago: Rand McNally, 1966).
Weber, David C., “The Changing Character of the Catalog in America,” in The
Library Quarterly, vol. 34 (Jan. 1964), pp. 20-33.
Webster, Charles, ed., Samuel Hartlib and the Advancement of Learning (London:
Cambridge University Press, 1970).
Wegner, Peter, “Three Computer Cultures, Computer Technology, Computer
Mathematics, and Computer Science,” in Alt, Franz L., and Rubinoff, Morris,
eds., and Freiberger, Walter, guest editor, Advances in Computers (New York:
Academic Press, 1970), vol. 10, pp. 8-78.
Wegner, Peter, “ADA Education and Technology Transfer Activities,” in ADA
Letters (Aug. 1982).
CUMULATIVE LIST OF REFERENCES 727

Wegner, Peter, ed.; Dennis, Jack; Hammer, Michael; and Teichroew, D., assoc,
eds., Research Directions in Software Technology (Cambridge, Mass.: MIT
Press, 1979).
Weinreich, Uriel, “Semantics and Semiotics,” in International Encyclopedia of the
Social Sciences (New York: Macmillan and Free Press, 1968), vol. 14, pp. 164—
169.
Weiss, Edward C., ed., The Many Faces of Information Science (Boulder, Colo.:
Westview Press, 1977).
Weizenbaum, Joseph, Computer Power and Human Reason: From Judgment to
Calculation (San Francisco: W. H. Freeman, 1976).
Wellisch, Hans, “From Information Science to Informatics,” in Journal of Librari-
anship, vol. 4 (July 1972), pp. 164-187.
Wertheimer, Max, Drei Abhandlungen zur Gestalttheorie (Neuherausgabe, Er¬
langen: Philosophische Akademie, 1925).
Western Joint Computer Conference, Extending Man’s Intellect, proceedings of
the Western Joint Computer Conference, held [in Los Angeles] by the Institute
of Radio Engineers, the American Institute of Electrical Engineers, and the
Association for Computing Machinery, May 9-11, 1961, vol. 19.
Wheeler, William Morton, “The Ant Colony as an Organism,” in Journal of
Morphology, vol. 22, no. 2 (1911), pp. 307-325.
White, Carl M., A Historical Introduction to Library Education (Metuchen, N.J.:
Scarecrow Press, 1976).
Whitehead, Alfred North, and Russell, Bertrand, Principia Mathematica (Cam¬
bridge: Cambridge University Press, 1910-1913; 2nd ed. 1925-1927; paperback
ed. 1962).
Whittemore, Bruce J., and Yovits, Marshall C., “A Generalized Concept for the
Analysis of Information,” in Debons, Anthony, ed., Information Science:
Search for Identity, proceedings of the 1972 NATO Advanced Study Institute in
Information Science, held [in Seven Springs, Champion, Penn.] Aug. 12-20,
1972 (New York: Marcel Dekker, 1974), pp. 29-45.
Widrow, Bernard, “Pattern Recognition and Adaptive Control,” in IEEE Transac¬
tions on Applications and Industry, vol. 83 (1964), pp. 269-277.
Wiener, Norbert, Cybernetics, or Control and Communication in the Animal and
the Machine (Cambridge, Mass.: MIT Press, 1948; 2nd ed., MIT Press and
Wiley, 1961).
Wiener, Norbert, Extrapolation, Interpolation, and Smoothing of Stationary Time
Series with Engineering Applications (Cambridge, Mass.: Technology Press of
the Massachusetts Institute of Technology, 1949).
Wiener, Norbert, The Human Use of Human Beings—Cybernetics and Society
(Boston: Houghton Mifflin, 1950 and 1954; 2nd ed., Boston: Avon, 1967).
Wiener, Norbert, God and Golem, Inc.: A Comment on Certain Points Where
Cybernetics Impinges on Religion (Cambridge, Mass.: MIT Press, 1964).
Wijesinghe,O., “The Effect of Varying the Rate of Presentation on the Information
Transmission of Schizophrenic and Control Group,” in British Journal of Psy¬
chiatry, vol. 130 (May 1977), pp. 509-513.
 728 CUMULATIVE LIST OF REFERENCES

Willems, J. C., “System-Theoretic Models for the Analysis of Physical Systems,” in

Ricerche di automatica, vol. 10 (Dec. 1979), pp. 71-105.
Williamson, Charles Clarence, “The Place of Research in Library Service,” in
The Library Quarterly, vol. 1 (Jan. 1931), pp. 1-19.
Williamson, Oliver E., “Transaction-cost Economics: The Governance Contrac¬
tual Relations,” in Journal of Law and Economics, vol. 22, no. 2 (1979), pp
233-261.
Williamson, Oliver E., “The Organization of Work,” in Journal of Economic Be¬
havior and Organization, vol. 1 (1980), pp. 5-38.
Wilson, Alan G., “A Statistical Theory of Spatial Distribution Models,” in Trans¬
portation Research (New York: Pergamon Press, 1967), vol. 1, pp. 253-269.
Wilson, Alan G., “The Use of Entropy-Maximizing Models in the Theory of Trip
Distribution, Mode Split, and Route Split,” in Journal of Transport Economics
and Policy, vol. 3 (Jan. 1969), pp. 108-126.
Wilson, Edward Osborne, The Insect Societies (Cambridge, Mass.: Harvard Uni¬
versity Press, 1971).
Wilson, Patrick, “Limits to the Growth of Knowledge: The Case of the Social and
Behavioral Sciences,” in The Library Quarterly, vol. 50 (Jan. 1980), pp. 4-21.
Wilson, T. D., “Sociological Aspects of Information Science,” in International
Forum on Information and Documentation (Moscow), vol. 6 (1981), pp. 13-18.
Windelband, Wilhelm, “The Principles of Logic,” in Ruge, Arnold, et al., Encyclo¬
paedia of the Philosophical Sciences, Vol. 1, Logic (London: Macmillan, 1913),
pp. 7-66.
Winograd, Shmuel, and Cowan, Jack D., Reliable Computation in the Presence of
Noise (Cambridge, Mass.: MIT Press, 1963).
Winograd, Terry, “Procedures as a Representation for Data in a Computer Program
for Understanding Natural Language” (Ph.D. diss., Massachusetts Institute of
Technology, 1971).

Winograd, Terry, Understanding Natural Language (New York: Academic Press

1972a).

Winograd, Terry, “Understanding Natural Language,” in Cognitive Psychology,

vol. 3 (Jan. 1972b), pp. 1-191. [Special issue]
Winograd, Terry, “On Some Contested Suppositions of Generative Linguistics
About the Scientific Study of Language,” in Cognition, vol. 5 (1977), pp 151-
179.

Winograd, Terry, “Beyond Programming Languages,” in Communications of the

ACM, vol. 22 (July 1979), pp. 391-401.
Winston, Patrick H., “Learning Structural Descriptions from Examples” (Ph.D.
diss., Massachusetts Institute of Technology, 1970).
Winston, Patrick H., ed., The Psychology of Computer Vision (New York:
McGraw-Hill, 1975).

Wisdom, John Oulton, “The Hypothesis of Cybernetics,” in British Journal for the
Philosophy of Science, vol. 2 (May 1951), pp. 1-24. Reprinted in Pylyshyn,
Zenon W., ed., Perspectives on the Computer Revolution (Englewood Cliffs,
N.J.: Prentice-Hall, 1970).
CUMULATIVE LIST OF REFERENCES 729

Wittgenstein, Ludwig von, Tractatus logico-philosophicus (London: K. Paul,

Trench, Trubner, 1922; another ed., trans. D. F. Pears and B. F. McGuinness;
London: Routledge and Kegan Paul, 1969).
Wonham, W. Murray, “On Pole Assignment in Multi-Input Controllable Linear
Systems,” in IEEE Transactions on Automatic Control, vol. AC-12 (1967), pp.
660-665.
Wonham, W. Murray, Linear Multivariable Control: A Geometric Approach (Ber¬
lin, Germany: Springer-Verlag, 1974, 1979).
Woods, William A., “Transition Network Grammars for Natural Language Analy¬
sis,” in Communications of the ACM, vol. 13 (Oct. 1970), pp. 591-606.
Woods, William A., “What’s in a Link: Foundations for Semantic Networks,” in
Bobrow, Daniel G., and Collins, Allan, eds., Representation and Understand¬
ing: Studies in Cognitive Science (New York: Academic Press, 1975), pp. 35-82.
Woodworth, Robert Sessions, and Marquis, Donald G., Psychology; A Study of
Mental Life (New York: Holt, 1921; 20th ed., 1949; reprinted Methuen and
Company, 1967).
Wright, H. Curtis, “The Wrong Way To Go,” in Journal of the American Society
for Information Science, vol. 30 (Mar. 1979), pp. 67-76.
Wright, H. Curtis, “Professionalism and the Socratic Paradox,” in Scholar and
Educator (spring 1980), pp. 5-15.
Wright, H. Curtis, “The Instrumentality of Data,” in National Library Association
Newsletter, vol. 6 (May 1981), pp. 4-5.
Wundt, Wilhelm, Sprachgeschichte und Sprachpsychologie, mit Rucksicht auf B.
Delbriicks “Grundfragen der Sprachforschung” (Leipzig, Germany: W. Engel-
mann, 1901).
Yockey, Hubert P., ed., Proceedings of a Symposium on Information Theory in
Biology, held [in Gatlinburg, Tenn.] Oct. 29-31, 1956 (New York: Pergamon
Press, 1958).
Young, John Frederick, Cybernetics (New York: American Elsevier, 1969).
Young, Robert M., Mind, Brain, and Adaptation in the Nineteenth Century: Cere¬
bral Localization and its Biological Context from Gall to Ferrier (Oxford, En¬
gland: Clarendon Press, 1970).
Yovits, Marshall C., and Abilock, Judith G., A Semiotic Framework for Infor¬
mation Science Leading to the Development of a Quantitative Measure of Infor¬
mation,” in Zunde, Pranas, ed., Information Utilities, proceedings of the thirty-
seventh ASIS Annual Meeting, vol. 11 (Washington, D.C.: American Society for
Information Science, 1974), pp. 163-168.
Yovits, Marshall C.; Jacobi, George T.; and Goldstein, Gordon D., eds., Self-
Organizing Systems, proceedings of a Conference on Self-Organizing Systems
[Chicago] 1962 (Washington, D.C.: Spartan, 1962).
Yovits, Marshall C.; Rose, Lawrence; and Abilock, Judith, Development of a
Theory of Information Flow and Analysis, in Weiss, Edward C., ed., The
Many Faces of Information Science (Boulder, Colo.: Westview Press, 1977), pp.
19-51.
Zadeh, Lofti A., “Fuzzy Sets,” in Information and Control, vol. 8 (June 1965), pp.
338-353.
730 CUMULATIVE LIST OF REFERENCES

Zadeh, Lofti A., and Desoer, Charles A., Linear System Theory (New York:
McGraw-Hill, 1963).
Zadeh, Lofti A., and Polak, E., System Theory (New York: McGraw-Hill, 1969).
Zeleny, Milan, “Cybernetics and General Systems—A Unitary Science?” in Ky-
bernetes, vol. 8 (1979), pp. 17-23.
Zeleny, Milan, Autopoiesis: A Theory of Living Organization (New York: Elsevier,
1981), General Systems Research Series, vol. 3.
Ziff, Paul, “The Feelings of Robots,” in Analysis, vol. 19 (Jan. 1959), pp. 64-68.
Zimmermann, Manfred, “Neurophysiology of Sensory Systems: The Sensory Sys¬
tem in the Light of Information Theory,” in Schmidt, Robert F., ed., Funda¬
mentals of Sensory Physiology, 3rd ed. of Grundriss der Sinnesphysiologie,
trans. Marguerite A. Biederman-Thorson (New York: Springer-Verlag, 1978),
pp. 68-80.
Zloof, Moshe M., “Query-by-Example: A Data Base Language,” in International
Business Machines Systems Journal, vol. 16, no. 4 (1977), pp. 324-343.
Zmud, Robert W., “Individual Differences and MIS Success: A Review of the
Empirical Literature,” in Management Science, vol. 25 (Oct. 1979), pp 966-
979.
Zunde, Pranas, and Gehl, John, “Empirical Foundations of Information Science,”
in Williams, Martha E., ed., Annual Review of Information Science and Tech¬
nology (White Plains, N.Y.: Knowledge Industries Publications [for the Ameri¬
can Society for Information Science], 1979), vol. 14, pp. 67-92.
 INDEX

Abelson, Robert P., 108, 194, 216, 249-251 Aristotle (384-322 B.C.), 10, 125, 130, 140,
Abilock, Judith, 140n, 649n 418-419, 561n, 662
Ackoff, Russell L., xv, 448n, 550, 555, 558, Armer, Paul, 255, 262
56In, 576, 603 Amol’d, V.I., 454
ADA, 168-169, 175 Arrow, Kenneth J., 583n, 630n, 649
Akaike, Hirotugu, 538n ASCII, 276, 293
Allen, Garland E., 654n Ashby, W. Ross, 442, 470, 619
Allende, President Salvador, 41, 437 Asratyan, G.E., 454
Amarel, Saul, 25, 26, 134, 233 Association of American Library Schools
Amari, Shun-ichi, 461, 464 (AALS), viii
American Association for the Advancement of Association for Computing Machinery (ACM),
Science (AAAS), vi, 373 viii, 20, 23, 164
American Association for Artificial Intelligence Association of Special Libraries and
(AAAI), vii, 101 Information Bureaus (ASLIB), 348, 399
American Association of University Professors Atkinson, Anthony A., 630n
(AAUP), vii Atlas, Ronald M., 484
American Bibliographical Society (formerly Attneave, Fred, 482, 500
Bibliographical Society of Chicago), 346 Augustinus, Aurelius [St Augustine]
American Documentation Institute (ADI), (354-430 A.D.), 561n
351-353, 372, 380-382, 399 Austrian Society for Cybernetic Studies, 429n
American Economic Association (AEA), vi, vii
American Library Association (ALA), 21, Babbage, Charles, 239, 670
345-347, 359, 367, 375-376, 399 Bach-y-Rita, Paul, 425
American Psychological Association) APA), vii Backus, John, 153, 203n
American Society for Cybernetics, 39, 457 BACON (computer system), 271-272
American Society for Information Science Bacon, Francis, 125, 665
(ASIS), 19, 352-353, 359, 372, 376, Bacon, Roger, 125
381-382, 386, 399, 403 Bailey, Herbert S., Jr., xv
Ampere, Andre-Marie, 39, 409, 459 Baiman, Stanley, 629
Anderson, James A., 102 Baker, Carl, 74n
Anderson, John R., 102, 108, 210, 289 Baker, Jeffrey J. W., 654n
Anderson, Norman H., 102 Ballard, Dana, 85
Anderson, Richard C., 102 Baralt-Torrijos, Jorge, 141
Acki, Masanao, 596n Bar-Hillel, Yehoshua, 37, 48, 53n, 495,
Aphrodite, 552 539-542, 592n, 650
APL, 168 Barker, Roger C., 64
Aplenc, A., 479 Bamea, Amir, 630n
Aquinas, Saint Thomas, 668n Barr, Avron, v, 28-31,237,249,259,264,267,
Arbib, Michael A., v, 30, 40, 74n, 81, 84-86, 274, 276, 280, 287, 289-290, 671
90, 116, 248, 434,452, 459, 463-467, 471, Bartlett, Frederic C., 86
525, 527n, 529, 653 Barto, Andrew G., 461
Archimedes (287-212 B.C.), 125 Barzun, Jacques, 584
Arden, Bruce, 66-67, 156, 165 Bateson, Gregory, 39n, 413, 561n

731
 732
INDEX

Batty, David, v, 365, 399-400 Bolt, Beranek, and Newman, Inc., 74n
Bauer, Walter F., 666n Boltzmann, Ludwig von, 49,433,476,478,489,
Baumol, William J., xvi, 393, 634 500, 504-505, 510
Bayes, Thomas, 478 Bongard, Mikhail Moiseevich, 267, 269-270
Beadle, George W., 654 Bookstein, Abraham, 392
Beam, Robert D., v, 607, 612, 616-617, Boole, George, 195, 285
631-632 Booth, A. Donald, 197
Bearman, Toni Carbo, v, 365, 399^400 Borden, H., 260
Beer, Stafford, 39n, 41n, 437, 561n Borelli, Giovanni Alfonso, 420
Bell Laboratories, 17, 49, 74n, 175, 223, 381, Boulding, Kenneth E., vi, xi, 8,43n, 46,49,547,
433, 547 555,574,576, 627,634,637,643n, 659
Bellman, Richard, 433, 527n, 575 Bower, Gordon H., 108, 210, 225
Bellugi, Ursula, 308 Bowler, T. Downing, 617
Benhabib, Jess, 620 Brachman, Ronald J., 216
Bennett, James S., 259 Bradford, Samuel Clement, 392, 395
Bense, Max, 470 Branscomb, Lewis M., 50n
Bensoussan, Alain, 527n Breal, Michel Jules Alfred, 143
Bentham, Jeremy, lOn, 561n, 666-667 Bresnan, Joan, 74n, 306
Berger, J., 484 Briet, Suzanne, 362
Berlekamp, Elwyn R, 501 Brigham Young University, 356
Berlinski, David, 416n, 435, 524 Brillouin, Leon, 478
Bernard, Claude, 81-82, 193, 635, 646n British Society for International Bibliography,
Bernoulli, Daniel, 508 348
Bernstein, Nikolai, 88, 464-465 Broadbent, Donald E., 100
Bertalanffy, Ludwig von, xiii, 8,39n, 42n, 41 In, Brockett, Roger W., 525, 527n
431-432, 446, 450, 527, 555-558, 561n, Broglie, Louis de, 41, 665-666
598n
Brookes, Bertram C., 548, 645, 654n
Berwick, Robert C., v, 335, 340 Brown, Christopher, 85
Beta Phi Mu, viii Brown, John Seely, 34, 235, 647n
Bever, Thomas G., v, 297, 306, 310, 328, Brown, Robert R., 611
330, 336 Brown University, viii, 74n
Bibliographical Retrieval Services (BRS), 355 Brunei University (U.K.), 429n
Bibliographical Society, The (U.K.), 346 Bruner, Jerome S., 100, 423-424
Bibliographical Society of Chicago, 346, 347 Brunovsky, P., 527n, 539, 541
Bibliography, International Institute of, Bryson, Arthur, 596n
347-348, 352n, 399 Buchanan, Bruce G., 217, 257
Bigelow, Julian, 65, 82, 192, 410, 416, 468 Buchanan, James, 636n
Binet, Alfred, 270 Buckland, Michael K., 22
Birkel, Paul, 187n
Buckley, Walter, vi, 598, 601, 612, 616,
Bitzer, Donald, 175 631-637
Black, Max, 13n Bucy, Richard, 461-462
Blaivas, Alexander, 454
Bunge, Mario, 448, 451-452, 555, 557-559,
Blau, G.E., 484 576,585-586, 625-626
Bliss, Henry Evelyn, 368 Burks, Arthur W., 63, 69
Blum, Jay, 463 Burmeister, Edwin, 596n
Blumer, Herbert, 386 Bush, Vannevar, 373, 410, 653n
Bobrow, Daniel G., 33n, 34, 73n, 206, 215 Butler, Pierce, 350
Bobrow, Robert J., 34, 647n Butterfield, Herbert, 63
Bode, Hendrik, 607 Buxton, John N., 166
Boden, Margaret A., v, 5, 29, 229, 233, 235, Byrnes, C., 527n
287-289
Bogdanov, Aleksandr A., 446, 556-558, 626 Cadwallader, M., 612
Bohm, Stephan, 635 Caesar, Caius Julius (101-44 B.C.), 662
Bohnert, Herbert, 373 Caldwell, Samuel, 410
INDEX 733

Callimachus (ca. 305-240 B.C.), 366 CONNIVER, 215, 221

Campise, Rose Ann, xvi Converse, Alvin O., 483
Cannon, Walter B., 41 In, 426n, 436, 596n, Cooke, Kenneth, 620
619,635 Cooley, Charles H., 386
Capek, Karel, 194 Copernicus, Nicholas, 163
Caplan, David, 84 Copinger, Walter Arthur, 346
Carbonell, Jaime G., 219 Cornell Medical School (New York), vi, 74n
Card, Stu, 187n Cornell University, ix
Carnap, Rudolf, 48, 53n, 82, 127, 142, 495, Costa, Serpil de, 348
539, 592n, 650, 665 Costanza, Robert, 633
Camegie-Mellon University, vii, 74n, 270 Counelis, James S., 452n
Carroll, C. Edward, 350 Covacich, Eliana, xv
Case Western Reserve University, vii, viii, Cowan, Jack D., 454-455, 486
354-355, 376, 380 Cowan, W. Maxwell, 102
Cashman, Thomas J., 647 Craik, Kenneth J. W., 82, 462
Casti, John L., 518 Crellin, Gary L., 478
Catronix Corporation, viii Crestadoro, Andrea, 367
CDB Enterprises, Inc., v Croce, Benedetto, 663, 665
Center for Advanced Study in the Behavioral Cuatrecasas, Pedro, 426-427
Sciences, 102 Curry, Haskell Brooks, 131
Cemy, J., 539 Cutter, Charles A., 365, 367-368, 385
Chaitin, Gregory J., 501 Cyert, Richard M., 629
Chandler, Robert E., 506 Cyranski, John F., 539-541
Chase, William G., 274
Checkland, Peter, 56In Daedalus, 179
Cherry, Colin, xv, 50-51, 55n, 434, 442, 470, D’Andrade, Roy, 74n
540,657-659 Dana, John Cotton, 381
Chervany, Norman L., 630n Dalton, John, 125
Dartmouth Summer Research Project on
Chiaraviglio, Lucio, 141
Chomsky, Noam, 37, 39, 100-101, 131, 154, Artificial Intelligence, 28n
208, 211, 297, 300, 302, 307,311-312, Darwin, Charles R-, 599
321-325, 330, 335, 494 Davis, Martin D., 64
Christian Church Fathers, 10 Davis, Randall, 259, 462
Church, Alonso, 28, 64, 131, 152, 238-239 Davis, Watson, 351-352, 373, 400
Churchland, Paul M., 234 Dawkins, Richard, 281
Churchman, C. West, vi, 44, 552n, 555, 558, Day, Richard H., vi, 619-620, 623, 631, 634
561, 576, 631, 635n, 636n, 647n Debons, Anthony, xvi, 22n, 139
Defense Advanced Research Projects Agency
Churchwell, Charles D., 350
(DARPA), 99n, 187n, 218, 223, 287n
Cigno, A., 620
Clancey, William J., 259 Delfour, M., 527n
Clark, Margaret, 288 Demski, Joel S., 627, 630n
Clausius, Rudolf, 476-477, 489 Denbigh, Kenneth G., 478
DENDRAL, 217, 257-258
Cleveland, Donald B., 352
Dennett, Daniel C., 69, 233, 263n, 273
Coase, Ronald, 636
Coates, J.F., 484 Denning, Peter J., 154
COBOL, 26, 30n, 168 Dennis, Jack B., 154
Cognitive Science Society, 72-73, 101, 267 Descartes, Rene, 58, 101, 118, 192, 216,
419—420, 451-452, 459, 663
Cohen, Morris R., 665
Desoer, Charles A., 527n
Cohen, Paul R., 254
Collins, Allan, 32-33, 73n Deutsch, Karl, 69, 457
Dev, Parvati, 464
Collins, Carter C., 425
Dewey, John, 124, 128, 143n, 349, 386
Columbia University, v, 324
Comenius, Johann Amos, 401 Dewey, Melvil, 367-368, 380
Condillac, Etienne Bonnot, Abbe de, 136 DIALOG, 355
 734 INDEX

Dickson, Gary R., 630n Ewert, Jorg-Peter, 85

Dijkstra, Auke, 484 Exxon Corporation, 74n
Dijkstra, Edsger, 135, 157, 164
Diodes (2nd Century B.C.), 418 Fairchild Industries, 74n
Dixon, J.K., 30n Fairleigh Dickinson University, viii
Dobell, A. Rodney, 596n Fairthome, Robert A., 384
Dolan, F.T., 646n Falb, Peter L., 434, 452, 525, 527n, 529
Dormer, Marc, 187n Fama, Eugene F., 630n
Dougherty, Ray C., vi, 321, 335-336 Fano, Robert M., 497
Dowds, John P., 483 Farkas-Conn, Irene, 35In
Dray, William H., 626n Farradane, Jason E.L., 19
Dresher, B. Elan, 249 Faurre, P., 538n
Drexel University, 381 Faustus, 179
Dreyfus, Hubert L., 222, 232, 597 Feigenbaum, Edward A., 28, 31, 67, 171, 179,
Driesch, Hans, 193 200, 210, 217, 244, 249, 255, 257, 290
Drury, John, 401 Feigenbaum, Mitchell J., 620
Dunlap, Knight, 420^122 Feinstein, Amiel, 51
Dworkin, Ronald, 636n Feldman, Jerome A., 165
Feldman, Julian, 67, 179n, 200, 255
Earhart Foundation, xv Feltham, Gerald A., 627, 629, 630n
Earl of Chesterfield, The, ix Financial Accounting Standards Board of the
Eastern Psychological Association, vii United States, 630
Eastman Kodak, 380 Fischer, Emile, 427n
Ebenfield, Helene, xv Fisher, Ronald A., 52, 434,487-488, 497-498
Eccles, John C., 91 Fiske, Susan, 288
Eckhom, R., 484 Flat Earth Society, 193
Eddington, Arthur Stanley, 547 Flexner, Abraham, 384
Edelman, Gerald M., 454 Flood, Merrill M., 355
Eden, Murray, vi, 40, 41n, 409, 441-442, 445, Flower, Richard, 501
448,450-454,457, 459-462, 465, 467, 471 Fodor, Jerry A., 69, 131, 233, 305-306, 313
Edison, Thomas Alva, 261 Ford Foundation, 102
Edwards, Edward, 365, 367 Fordham University, ix
Egan, Margaret E., 354 Forrester, Jay W., 69, 568-569, 620
Ehrlich, Paul, 426n Forte, B., 539
Eigen Manfred, 448, 454 FORTRAN, 26, 30n, 166, 168, 203
Einstein, Albert, 128, 268, 451-452, 505, 548 Frank, Philipp, 666, 669n
Elias, Peter, vi, 40, 4In, 47-48, 55, 441, 470, Frankenstein, 179, 194
497,501 Frederiks, J.A.M., 86
El Sayed, Y„ 479 Fredholm, (Erik) Ivar, 444
Emery, Fred E., 555, 558, 56In, 576, 603 Free University of Berlin, 484
Emmet, Boris, 508, 510 Frege, Friedrich Ludwig Gottlob, 28, 64, 238,
Emonds, Joseph E., 302 298
England, James W., 500 Frick, Frederick C., 481
EPAM (Elementary Perceiver and Memorizer), Fried, Charles, 636n
210 Fromkin, Victoria A., 309
Erasistratus (b. ca. 304 B.C.), 418 Fuhrman, P., 527n
Erman, Lee, 218 Furth, Hans G., 140n
Ermentrout, G.B., 454
Ernst, George, 31, 73n, 206 Gabor, Dennis, 52, 487-488
Euclid (b. ca. 295 B.C.), 170 Galanter, Eugene, 247
Euler, Leonhard, 454 Galen (129-199 A.D.), 419
Evans, Robert B., 478-479 Galilei, Galileo, 63, 136
Evans, Thomas G., 269 Gallagher, Robert G., 497, 501
INDEX 735

Galvani, Luigi, 81 Haken, Hermann, 453

Galvin, Thomas J., 376 Hall, Thomas S., 418-420
Gardner, Beatrice T., 308 Halle, Morris, 300, 327n
Gardner, R Allen, 308 Hanika, Francis de Paula, 429, 436
Garfield, Eugene, 11, 373 Hanson, Allen R, 464
Gamer, Wendell R, 483, 649 Hardin, Garrett, 161, 597, 635
Garrett, Merrill F., 305 Harter, Stephen P., 21
Garwig, Paul L., 351 Hartley, Ralph V. L., 17, 49-50, 65, 433, 505,
Gaughan, James T., xvi 659
Gauld, Alan, 233 Hartlib, Samuel, 401
Gazis, Denos C., 506 Hartmanis, Juris, 134, 154
Gazzaniga, Michael S., vi, 93, 116, 653 Harvard College, 366
Gehl, John, 141, 151, 392 Harvard University, ix
Gel’fand, Izrail’ Moiseevich, 465 Harvey, William, 419—420
Gell-Mann, Murray, 451 Hastings, John Woodland (Woody), xv
General Electric Corporation, 380 Haugeland, John, 69
General Motors Corporation, 257, 506 Haugen, Robert A., 630n
Georgia Institute of Technology, viii, 141 Hawking, Stephen, 451
Gergely, Tamas, 452n Hayek, Friedrich A., 583—585,586n, 599,632,
Gibbs, Josiah Willard, 146, 411, 478-479 633n
Gigley, Helen, 84 Hayes, Patrick J., 29, 234
Givol, David, 426 Hayes, Robert M., 647n
Glushkov, Viktor M., 416, 442 Hayes-Roth, Frederick, 219
Godel, Kurt, 64, 131, 452, 548 Hazewinkel, M., 527n
Godot, 548 HEARSAY-II, 218, 270
Goguen, Jack, 455 Hegel, Georg Wilhelm Friedrich, 124, 585
Golden Gate Bridge, 142 Heilprin, Laurence B., 140n
Goldfeld, Stephen M., xv Held, Richard, 426
Goldschmidt, Robert, 348 Helmholtz, Hermann von, 146, 182, 426n
Goldstein, Gordon D., 201 Hempel, Carl G., 127-128, 611
Goldstein, Ira, 206 Herbrand, Jacques, 180
Goldstein, Kurt, 309 Herman, E. Edward, 617
Good, Irving J., 478 Herman, Robert, 506, 508
Goodwin, Brian C., 426 Hermann, Robert T., 527n
Gorelik, George, 452n Hero of Alexandria (ca. 62 A.D.), 419, 459
Gom, Saul, vi, xv, 20, 23, 26, 36,95,115,117, Herophilus (4th Century B.C.), 418
121,124,129,141-146,149-150,155-156, Herschel, John, 663
160-164, 177, 669n Hess, W.R, 41 In
GPS (General Problem Solver), 428 Hesse, Hermann, 128
Greaves, Melvyn F., 426—427 Hewitt, Carl E., 251
Gregory, Richard L., 86 Hewlett-Packard Laboratories, 260
Greibach, Sheila A., 134 Hick, W. Edmund, 490
Grenander, Ulf, 454, 468 Higginbotham, James, 327n, 333n
Griffith, Belver C., 457 Hilbert, David, 64
Groot, Adriaan D. de, 274 Hilgard, Ernest R, 225, 425
Grossberg, Stephen, 461 Hill, Richard E., 56In
Gunderson, Keith, 217 Hillman, Donald J., 140n
Gurfinkel’, V.A., 465 Hillman, James, 562
Guzman, A., 268 Hilton, Ronald W., 630n
Hinton, G.E., 231-233
Habermas, Jurgen, 56In Hirshleifer, Jack, 594, 629
Haberstroh, Chadwick H., 612 Ho, Y.C., 596n
Hain, Ludwig [Friedrich Theodor], 346 Hobbes, Thomas, 56In
 736
INDEX

Hofstadter, Douglas R., vi, 29, 31, 130n, Jewett, Charles C., 365, 367-368, 400
263-264, 277, 283, 285, 288,291-294 Johns, Ada Winifred, 349
Holst, Erich von, 41 In
Johns Hopkins University, 420
Holstrom, Bengt R., 630n Johnson-Laird, Philip, 234
Holzner, Burkhart, 393 Jones, T.L., 30n
Homans, George C., 611
Josiah Macy Jr. Foundation, 82, 411,
Hooke, Robert, 420 441-442
Hoos, Ida R., 552n, 56In
Jouvenel, Bertrand de, 136n
Hopf, Eberhard, 454 Julesz, Bella, 464
Homstein, Norbert, 249 Juncosa, Mario L., 666n
Howard, Ronald A., 589
Hoyle, Frederick, 386 Kailath, T., 538n
Huber, Mary Taylor, xv Kak, Avinash C., 454
Hume, David, 58, 599n
Kalman, Rudolf E., 433-434, 452, 461-462,
Huyghens, Christiaan, 126
521, 525, 527n, 529, 533-539, 552-553,
557, 568-569, 576
Icarus, 179
Kamien, Ralph, 596n
Imperial College, London, 442 Kampe de Feriet, J., 539
Indiana University, vi Kaneko, Tatsuo, 484
Ingarden, R., 539
Kant, Elaine, 187n
Inhelder, Barbel, 229
Kant, Immanuel, 38, 58, 561n, 562,598,636n,
Institute of Electrical and Electronics Engineers 663
(IEEE), 430-431, 441-442
Kaplan, Abraham, 383, 403
Institute of Information Scientists, The (UK) Kaplan, Ronald, 74n
19
Karetsky, Stephen, 400
Institute for Intellectual Cooperation, The, 401 Karmiloff-Smith, Annette, 229
Institute of Management Sciences, The (TIMS), Katz, Jerrold J., 298
vi
Kelvin, Lord, 52, 146
International Association of Cybernetics, The
Kendall, D.G., 500
429
Kent, Allen, 380, 647 n
Institut National de Recherche en Informadque
Kepler, Johannes, 557, 577
et en Automatique (ENRIA), 517n
Keyser, Samuel Jay, vi, 74n, 327, 329n, 336
International Business Machines Corporation Khinchin, Aleksandr L, 47n, 443-444
(IBM), 58, 74n, 372
Khrushchev, Nikita, 444
International Economic Association, vii Kilmer, William L., 463
International Federation for Documentation Kim, Scott, 279
(FID), 352, 380 King, Donald W., xvi, 19, 22n, 381-382
International Institute for Applied Systems King, Gilbert W., 373
Analysis (IIASA), 450 Kirchoff, Gustav Robert, 533
Itami, Hiroyuki, 630n Kirzner, Israel M., 599n
Kistiakowski, George, 156
Jackson of Burnley, Willis Jackson, Baron, 442 Kleene, Stephen C., 64, 131
Jacob, Francois, 448 Klima, Edward S., 308
Jacobi, George T., 201 Kling, Robert, 155
Jakobson, Roman, 36n, 103 Klir, George Jiri, 42n, 432, 438, 451, 527n
James, William, 225, 386 Klivington, Kenneth A., xv
Jamison, Dean, 48, 651 Knuth, Donald E„ xv, 26, 135, 156, 164, 265,
Jantsch, Erich, 56In 271,501,667
Jaspers, Karl, 621 Kochen, Manfred, vi, 47n, 140n, 371, 373n,
Jaynes, Edwin T„ 478^181, 483, 506, 513 376,401-403,453,455,457,467-468,471,
Jeffrey, Richard C., xv 643n, 647n
Jekyll and Hyde, 179 Koestler, Arthur, 63
Jensen, Michael C., 629 Koffka, Kurt, 270
Jeuck, John E., 508, 510 Kohler, Wolfgang, 207, 270
INDEX 737

Kolmogorov, A.N., 443, 500, 501 Lindsay, Robert, 217, 257

Korchak, Robert, xv Linnaeus, Carl, 170
Komfeld, W.A., 251 Lions, Jacques Louis, 527n
Kosslyn, Stephen M., 223 LISP, 30n, 109, 221, 277, 279-281
Kotelnikov, Vladimir A., 443 Llinas, Rudolfo, 462
Kotovsky, Kenneth, 269 Locke, John, 36n, 142n
Kowalski, Robert, 221 Locke, William Nash, 197
Krauser, Cheri, 19 Loebner, Egon E., 260
Krichevsky, Micah, 484 Loew, Rabbi, 193
Kristol, Irving, 597 LOGIC THEORIST, 202
Kuhn, Alfred, 483, 605, 608n, 611-613, LOGO, 322, 324-325
615-617 Lonberg-Holm, Knud, 373n
Kuhn, Thomas S., 97, 104,122,125-126,128, Lorenz, Edward, 620
136-137, 153-154, 163, 169,182,188,226 Lorenz, Konrad, 309
Lotke, Alfred J., 619
LADSIRLAC (Liverpool and District Scientific, Lowen, R., 455
Industrial, and Research Libraries Advisory Luckmann, Thomas, 7, 656n
Council), 368 Lucretius, Titus, Cams (98-55 B.C.), 561n
Laferriere, Daniel, 103 Luenberger, David G., 525, 527n, 532n
Lakatos, Imre, 104, 188, 226, 553n Luhn, Hans Peter, 372
Lakoff, George, 13n, 302 Lukes, Steven, 626
Lamberton, Donald M., 627
Lange, Oskar, 447n MacDougall, William, 423
Langefors, Boije, 151 Machlup, Fritz, vii, ix, x, xi, xiii, xv, xvi, 3, 9,
Langer, Susanne, 63 lOn, 1 In, 13,20, 48,57,126,155,169, 333,
Langley, Patrick, 272 393, 584, 598, 627, 629n, 641, 643, 644n,
Langlois, Richard N., vii, xv, 551, 559, 561 n, 645n, 656n, 662-664, 666-667, 668n
563,575-576,581,585,597, 601-605,607, Machlup, Mitzi, xvi
611, 613-615, 618-619, 622, 625-626, 631 Machlup, Stefan, xv
Larson, C.T., 373n MacKay, Donald M., vii, 5, 47n, 52-53,
Lashley, Karl S., 89 485-492, 513, 587n, 591, 593, 598,
Lavoisier, Antoine-Laurent, 125 602-603, 609-610, 614, 658n
Lederberg, Joshua, 217, 257 Macko, Donald, 452, 527n, 556, 569
Lee, Yuk Wing, 410 MACSYMA, 203, 217n
Leeson, Kenneth, 393 Majumder, D. Dutta, 452n
Leibniz, Gottfried Wilhelm, 401, 411, 439, Malin, Morton V., 11
446n Manes, Ernest, 463
Leibowitz, Herschel W., 426 Mannheim, Karl, 393
Lenat, Douglas, 252, 462 Mansfield, Una, vii, x, xi, xvi, 3, 22n, 64In,
Lenneberg, Eric H., 309 666n
Leontief, Wassily, 604n Manutius, Aldus, Pius, 366
Lesser, Victor, 218, 459 Marais, Eugene N., 278
MARC (Machine Readable Cataloging), 358
Levine, Lewis, xv
Levine, Raphael D., 475n, 484 March, James G., 629
Levi-Strauss, Claude, 64, 69 Marcus, Matityahu, 393
Levy, Jerre, 310 Marcus, Mitchell P., xv, 69
Lewin, Kurt, 422 Markov, Andrei Andreevich, 131, 198
Lewis, Gilbert N., 478, 660 Marquis, Donald G., 422^123
Library Association (U.K.), 345-346, 348, Marr, David C., 69, 85, 220, 230-231, 337,

367,399 464
Library of Congress, 172, 358, 372-373 Marschak, Jacob, 627, 629
Licklider, Joseph C. R., xv, 205 Marshall, John C., 84
Lieberman, Philip, 308 Martin, Clyde F., 527n
Lindsay, Peter H., 270, 425 Martin, Nathaniel F.G., 500
 738 INDEX

Martin, Paul E., 41 n 200, 215-216,246-247,260-261,287,307,

Maruyama, Magoroh, 599 428-429, 454, 471
Marx, John H., 393 Misunas, David P., 154
Marx, Karl, 124, 584, 597, 635 Mitchell, Thomas, 219
Massachusetts Institute of Technology, v, vi, vii, Mittelmann, H.D., 454
viii, 3On, 32, 74n, 157, 208, 211, 217n, Mitter, S.K., 527n
268-269, 442, 495, 505 Moivre, Abraham de, 503, 505, 510
Mattessich, Richard, vii, 40-41, 445, 449, Monod, Jacques, 281, 448
451-452, 468^169, 471, 555, 557-558, Montroll, Elliott W., vii, 503, 506-508
576-577, 625-628, 630n, 632-634 Mooers, Calvin C., 372, 380
Maunsell, Andrew, 366 Moore, Edward F., 455
Maxwell, James Clerk, 47n, 81, 146, 433, 505, Morris, Charles William, 36n, 64, 128, 132,
635 142-144
May, Robert, 620 Mortazavian, Hassan, vii, 447, 449, 451, 517,
May, Rollo, 454 547-559, 563, 566-570, 573, 575,577,
Mayne, A.J., 452n 58In, 585, 592n, 594n, 625, 631n, 665
McCarthy, John, 28n, 30n, 206-207, 283,428, Moser, Jtingen, 620
465,622 Moses, Joel, vii, 24n, 45,95,157,177-178,203
McCorduck, Pamela, 28, 187, 222, 240, 670n Mosteller, Frederick, 607
McCracken, Daniel D., 173 Mountcastle, Vernon B., 454, 463
McCulloch, Warren, 39n, 82, 84-85,103-104, Musgrave, Alan, 553n
195, 199, 239, 255, 421, 463, 465, 622 MYCIN, 217, 258
McEliece, Robert J., 497, 502
McGill, Michael, xv Nagel, Thomas, 263
McGill, William J., 482, 500 Nalimov, Vasilii V., 37
McKerrow, Ronald Brunlees, 366 Napoleon, 136
McRae, Robert, 401 Nash-Webber, Bonnie, 214n
Mead, George Herbert, 143n, 319, 386, 603 National Aeronautics and Space Administration
Mead, Margaret, 142n, 413 (NASA), 449, 557
Meadows, Donnella H., 561 n, 566 National Commission on Libraries and
Meckling, William H., 629 Information Science (NCLIS), v
Mendel, Johann Gregor, 268 National Institutes of Health, Bethesda, vi
Meredith, Marsha J., 263n National Microfilm Association, 380
Merton, Robert K., 152 National Physical Laboratory (U.K.), 67
Mesarovic, Mihajlo D., vii, xv, 44-45, 56, 432, National Science Foundation, xiii, xv
450-452, 523-525, 527n, 556, 567, 569, Naur, Peter, 166
573-574, 585-588, 594n, 597-598 Neisser, Ulric, 9, 100, 209-210, 248
Mettrie, Offray de la, 81 Neufeld, Richard W.J., 484
Meyer, Jean, 278 Neumann, John von, xiii, 48, 49n, 63-64,
Meyer-Eppler, Werner, 470 66-67, 104, 153-154, 196, 239, 246, 251,
Michalski, Ryszard S., 219 413, 426n, 433, 455, 469n, 476
Michelangelo [Buonarroti], 385 Nevitt, Barrington, 614
Mikulak, Christina, xvi Newell, Allen, vii, 12, 23, 26, 29-32, 37, 57,
Mill, John Stuart, 667 63-64, 69, 72, 73n, 99-100, 115-117, 134,
Miller, George A., vii, xi, xv, 34, 57, 59, 74n, 152, 164, 173, 181, 187, 189, 190, 198n,
100, 104, 111, 116, 131, 234, 247, 272, 200, 202, 203n, 206, 216n, 232, 238-240,
304-305, 319, 481, 483, 493,613,654 248,250, 252-258,263,273,275-278,287,
Miller, James G., 52n, 56, 446, 484, 555, 558, 289, 293, 335-336, 428, 671
561n, 576, 618, 654-656 Newmeyer, Frederick J., 299
Miller, Jonathan, 449 Newton, Isaac, 47n, 126, 163, 268, 439, 455,
Miller, Laurence, 250, 287 533,557,577, 597
Milsum, John H., 461 New York University, vi, vii, x, xv
Minsky, Marvin L., 27-29, 32, 57,67, 86,153, Ng, David S., 630n
INDEX 739

Nicolis, George, 454 PLANNER, 215, 221

Nilsson, Nils J., 212, 221, 237, 462 Plato (427-347 B.C.), 116,127,154,178,551,
Nirenberg, Marshall W., 654 553-554, 561n
Nisbett, Richard, 288 PLATO (computer system), 175
Nishimura, Kazuo, 620 Poggio, T., 337
Norman, Donald A., 35, 74n, 216, 256-257, Poincare, Jules Henri, 455
270, 425 Polak, E., 527n
Nozick, Robert, 636n Pollack, Irwin, 482
Nyquist, Harry, 17, 65, 433, 505 Pontryagin, L.S., 527n
Poole, William Frederick, 368
Oates, Wallace, 634 Poore, Jesse H., 141
Ochs, Sidney, 652n Popel, B., 484
OCLC (Online Computer Library Center), 355 POPLAR, 215
Office of Scientific Research and Development Popov, V.M., 527n
(U.S.), 441 Popper, Karl R., 91, 126, 137, 154, 233, 556,
Ogden, Charles K., 142 635n, 636n
O’Neil, Wayne, 329n Porat, Marc Uri, x
Otlet, Paul, 347-348, 351, 368, 373, 379-380, Posner, Michael I., 100
400-401 Post, Emil Leon, 64, 131
Otten, Klaus, 139, 140n Potts, Renfrey B., 506
Oxford University, 662 Praxagoras (b. ca. 340 B.C.), 418
Premack, David, 308
PAC-MAN, 261 PRESTEL, 361
Panizzi, Antonio, 365, 367-368 Pribram, Karl H., 247
Papert, Seymour, 206,235,247,255,287,322, Price, Derek J. de Solla, 419
324-325, 454n, 471 Priestley, Joseph, 125
Parsons, Talcott, 655 Prigogine, Ilya, 454, 599, 620
PASCAL, 30n, 168 Princeton University, vii, x, xv, xvi, 74n
Pask, Gordon, 39n Princeton University Press, x, xv
Pearson, Charls, viii, 20, 27-28, 36, 134, Procaccia, I., 484
140-142, 144, 181-183, 668n PROLOG, 221, 222
Peck, A.L., 418 Prometheus, 179
Peirce, Charles Sanders, 36n, 126, 128, 132, Ptolemy (100-170 A.D.), 163
142-145, 181, 386 Public Libraries Act (U.K.), 367
Pellionisz, Andras, 462 Putnam, Hilary, 102
Peritz, Bluma C., 396 Pylyshyn, Zenon W., viii, 12, 20, 27, 32—33,
Perlis, Alan J., viii, xv, 23-26, 149, 152, 164, 66, 79, 74n, 81, 84-87, 90-104, 107-108,
178, 181,666n 112,115, 289,294,459,467,671n
Perry, J.W., 380
Pestel, Eduard, 597 QA4, 215
Petty, William, 401 Quastler, Henry, 501
Phillips, Almarin, 600n Quillian, M. Ross, 108, 211
Phillips, Dennis C., 582, 584-585, 633n Quine, Willard van Orman, 311
Piaget, Jean, 39, 64, 86, 124, 136, 322-325
Piatelli-Palmarini, Massimo, 39 Radner, Roy, 627, 629
Picavet, Francois Joseph, 124, 136 Rail, Wilfred, 465
Picci, G., 538n Ralston, Anthony, 23-24
Pierce, John R., 47n Ramo, Simon, 552n
Pinker, Steven, 223 Ramon y Cajal, 82
Pinsker, M.S., 501 Rand Corporation, 74n, 380
Pitts, Walter, 82, 84-85, 195, 199, 412, 421, Randell, Brian, 166
465,622 Ranganathan, Shiyali Ramamrita, 368, 400
Planck, Max, 505, 547 Raphael, Bertram, 31-32, 215, 462
 740 INDEX

Rapoport, Anatol, 42n, 127, 450, 451n, 601, Sattinger, D.H., 454
633n Saussure, Ferdinand de, 36n, 64
Ratliff, Floyd, 463 Sayers, W.C.B., 368, 400
Rayward, W. Boyd, viii, 22,343,346,348,350, Scalapino, D., 481
365-369, 374-376, 379, 382-383, 389, Schank, Roger C., 108, 214n, 216, 249-251,
394-397, 399-400 269-271, 281-283
Reddy, D. Raj, 270 Scheele, Carl Wilhelm, 125
Reilly, Park M., 484 Scheibel, Arnold B., 463
Reitman, Walter, 289 Scheibel, Madge E., 463
Rescher, Nicholas, 557 Schell, Jonathan, 457
Resnikoff, Howard L., xv Schlumberger-Doll, Research, 74n
Rich, Robert F., xv Schmidt, Robert F., 425, 652n
Richards, Ivor A., 142 Schrader, Alvin M., 6n
Richardson, John, 400 Schrodinger, Erwin, 111
Riemann, Georg Friedrich Bernhard, 548 Schultz, Claire K., 351
Riley, John G., 594, 629 Schumpeter, Joseph A., 599n
Rinzel, John, 465 Schupbach, Rosa, xv
Riseman, Edward M., 464 Schuster, Peter, 454
Rissanen, J., 538n Schutz, Alfred, 7n, 656n
Roberts, Lawrence G., 210, 268 Schwager, Carole, xvi
Robertson, George, 173 Schwartz, Nancy, 596n
Robinson, Alan J., 212 Science Service, 351
Rockefeller University, 454 Searle, John R., 84, 233, 235, 241, 263, 273,
Rose, John, 437 277, 280, 282
Rose, Lawrence, 140n Sears, Roebuck and Company, 507-510
Rosen, Saul, 25, 27 Sebeok, Thomas A., 36n
Rosenblatt, Frank, 199, 454n, 471 Seiffert, Helmut, 38, 48
Rosenblueth, Arturo, 65, 82, 192, 410, 416, Selkurt, Ewald E., 426
438, 453, 456, 468 Sellen, Betty-Carol, 357
Rosenbrock, H.H., 527n Senbet, Lemma W., 630n
Rosenfeld, Azriel, 454 Senn, James A., 630n
Ross, John R, 302 Shakespeare, William, 471
Ross, Lee, 288 SHAKEY (the robot), 462
Rosser, John Barkley, 64 Shannon, Claude E., xiii, xv, 5, 17, 41, 47-52,
Rothe, Carl F., 426 56, 65-67, 156, 371-372, 381, 383, 385,
Rouchaleau, Y., 527n 421, 433-434, 441-444, 455, 465, 469n,
Royal Society, 366, 399, 401, 488 475-480,483-491,493-501,505-506,510,
Ruch, Floyd Leon, 424 513,540,547,570,573,587,601,613,622,
Rugh, Wilson J., 531 646n, 658-660
Rumelhart, David E., 270 Shaw, John C., 202, 203n, 248
Russell, Bertrand, 28, 64, 66, 82, 195, 227, Shaw, Ralph, 380
238, 520, 548, 571,643, 656n, 665 Shelley, Mary Wollenstonecroft, 194
Rutgers University, 74n, 270 Shelly, Gary B., 647
Ruyer, Raymond, 59In Shera, Jesse Hauk, viii, xv, 352, 354, 379,
402-403, 649
Sabloff, Jeremy A., 620 Sherrington, Charles Scott, 82
Sager, Naomi, 140n Shik, M.L., 465
Sague, Virginia M., 19 Shils, Edward Albert, 655n
Salton, Gerard, 134 Shimoni, Y., 484
Samuel, Arthur L., 219 Shirey, Donald L., 22 n
Samuelson, Paul A., 583n, 620, 621 Shortliffe, Edward H., 217, 258, 462
San Jose State University, 430 SHRDLU, 280
Saracevic, Tefko, 155 Shwartz, Steven P., 223
Samoff, David, 260 Sigma Xj, 264
INDEX 741

Simon, Herbert A., xiii, xv, 5, 23, 25-26, 29, Steiner, George, 283
32-36, 57, 63-65, 68, 73n, 100, 104, 134, Sternberg, Saul, 100, 210
152, 164, 179n, 181, 200, 202, 210, 232, Sternberg, Stan R., 454
238-240,248,252-258,260,263-264,267, Stigleitz, Kenneth, xv
269-278,285,293-294,307,428,555,558, Stigler, George J., 627
577, 629 Stirling, James, 503-504
Simonov, P., 454 Stoeckenius, Jan, 630n
Sinai, iAkov, Grigor’evich, 500 Stoics, 36n, 139-140
Singer, Edgar A., Jr., 56In Strachey, Christopher, 23n, 152
SINTO (Sheffield Interchange Organization), Sussman, Gerald Jay, 219, 268
368 Sutherland, Ivan E., 205
SKETCHPAD, 205 Sutherland, William R., 165
Skinner, Burrhus F., 297, 319, 597 Sutton, Richard S., 461
Slagle, James R., 30n, 203 Swank, Raynard C., 388
Slamecka, Vladimir, viii, 20, 27-28, 36, 134, Swanson, Don R., xv, 344, 355, 360
140n, 141, 151, 181-183, 668n Sweeney, James L., 596n
Slater, John C., 478, 505 Swets, John Arthur, 100
Sleeman, Derek, 235 Swire, Peter, xv
Slepian, David, xv, 497 Syracuse University, 21
Sloan Foundation, 33-34, 74-75, 95, 99, 102 Szentagothai, Janos, 465
Slobin, Daniel I., 312 Szilard, Leo, 426n, 433, 469n
Sloman, Aaron, 69, 233
Small, Henry, 11, 457 Takahara, Yasuhiko, 452, 525, 527n, 556,
Smets, Philippe, 455 568-569
Smith, Adam, 599-600, 605, 635-636 Tanimoto, Taffee T., 372
Smith, Brian C., 216 Tannenbaum, A., 527n, 539
Smith, George E., 223 Tanner, J.M., 100
Smith, Reid G., 251 Tarski, Alfred, 28, 238
Snow, Charles P., 3 Tate, Vernon D., 352
Soames, Scott, 74n Taube, Mortimer, 146, 222, 380
Social Sciences and Humanities Research Technical University of Berlin, 429n
Council of Canada, 445n, 555n, 625n Teichroew, D., 648 n
Society for General Systems Research, 42,548, TEIRESIAS, 259
605 TELENET, 358
Society for Philosophy and Psychology, viii Terman, Lewis Madison, 270
Socrates (469-401 B.C.), 384 Terrace, Herbert S., 308
Solomonoff, R.J., 501 Teuber, Hans-Lukas, 426
Sommerfeld, Arnold [Johannes Wilhelm], 505 Thom, Rene, 45In, 454, 527-529, 535n,
Sontag, E., 527n 542-543, 547, 599-600n
Sonuga, J.O., 484 Thompson, Nina R., 416n
Sorenson, Paul G., 131 Thompson, Richard, 74n
Special Libraries Association, 359, 381, 399 Thue, 131
Spencer Foundation, xv Tinbergen, Jan, 537
Sperber, Daniel, 319 Tonsor, Stephen J., 386n
Sperry, Roger W., 280, 653 Torda, Clara, 252
Spinoza, Benedictus de, 56In Toussaint, Godfried, 266
Sridharan, N.S., 270, 283 Toynbee, Arnold, 655n
Stakgold, Ivar, 454 Tracy, Destutt de, 96, 124, 136-137
Stalin, Josef, 417, 444 Trappl, Robert, 429, 436
Stanford Research Institute, 32, 74n, 211 Tremblay, Jean-Paul, 131
Stanford University, v, 32, 74n, 211, 258, 388 ' Tribe, Laurence H., 552n
Stanley-Jones, Douglas, 452n Tribus, Myron, viii, 48, 372n, 475-482, 485,
Stark, Lawrence A., 425 490, 492-493, 497, 500-503, 513
Stein, Gertrude, 292 Tsetlin, M.L., 465
 742 INDEX

Tukey, John, 607 Van Schuppen, J.H., 538n

Turchin, Valentin F., 430 VIEWDATA, 361
Turing, Alan M., 28, 64, 67, 152, 195,
238-239, 280, 454, 468, 667 Wagner, Richard, 41 In
Turing Machine, 57-58, 66-67, 82, 131, 239, Walker, Edward, 74n
293, 314, 501 Walras, Leon, 627
Turnbull, George Henry, 401 Walton, Aline, xvi
Tyldesley, John R, 483 Waltz, David L., 268
TYMNET, 358 Wang, Hao, 202
Waples, Douglas, 349
Uhr, Leonard M., 454 Waterman, Donald A., 219
United States Congress Office of Technology Waterman, Talbot H., 45, 435^136
Assessment, 484 Watson, James D., 564
United States National Committee for FID Watson, John B., 194
[Federation for International Documentation], Watt, James, 81
viii Weaver, Warren, 49, 65, 156, 381, 383, 385,
University of British Columbia, vii, 452n 433-434, 476, 540, 547, 582, 587, 613,
University of California, Berkeley, vi, viii, 74n, 646n, 660
356,388 Webb, Eugene W., 392
University of California, Irvine, 74n Weber, H., 454
University of California, Los Angeles, vii, 430 Weber, Max, 665
University of California, San Diego, 74n Webster, Charles, 401
University of California, Santa Barbara, 74n Wegner, Peter, viii, 20, 25,156,163-164,167,
University of Chicago, viii, 74n, 349-350, 169,177,179, 182-183
35In, 355-356, 365, 368, 399-400 Weinberg, A., 340
University of Colorado, vi, 74n, 356 Weinberg, Alvin, 373
University of Connecticut, Storrs, vii Weinreich, Uriel, 38
University of Illinois, 74n, 350, 356 Weiss, Edward C., xvi, 392
University of Keele, Staffordshire (U.K.), vii Weizenbaum, Joseph, 135, 222
University of Maryland, College Park, vii Wellisch, Hans, 19
University of Massachusetts, Amherst, v, 74n, Wells, H.G., 373, 400-401
459, 463 Wertheimer, Max, 270
University of Michigan, vi, 74n, 350, 375, 386 Wheeler, John A., 451
University of New Hampshire, Durham, vi Wheeler, William Morton, 278
University of Pennsylvania, vi, 18-19, 74n White, Carl M., 350
University of Pittsburgh, 74n Whitehead, Alfred North, 28, 64, 82,195, 238,
University of Rochester, 74n 446
University of Southern California, vi, 620 Whittemore, Bruce J., 140n
University of Sussex (U.K.), v Widrow, Bernard, 461
University of Texas, 74n Wiener, Norbert, xiii, 5, 7n, 8, 39, 41,65-66,
University of Vienna Medical School, 429n 82, 91, 137, 192, 239, 371, 409^117,
University of Western Ontario, London 421-449,453-456,459^163,467-471,481,
(Canada), viii, 74n 488, 491, 527, 591n, 615, 635, 656-657
University of Wisconsin, Superior, v Wijesinghe, O., 484
UNIX, 175 Wilensky, Robert, 249
Uppsala University (Sweden), 429n Willems, J.C., 527n
Urbanik, K., 539 Williamson, Charles C., 349-350
Williamson, Oliver E., 632
Wilson, Alan G., 483
Van der Waals, Johannes D., 478 Wilson, Edward Osborne, 277
Van Doren, Charles, 373 Wilson, Patrick, viii, 389, 392, 404-405
Van Gigch, John P., 56In Wilson, T.D., 393
Van Marlen, Geert, 484 Windelband, Wilhelm, 17
Van Putten, C., 538n Winograd, Shmuel, 455, 486
 743
INDEX

Winograd, Terry, xv, 73n, 153, 213, 244, 249, Yockey, Hubert P., 501
268, 280-281, 315, 462 Young, John F., 416-417
Winsor, Charles, 607 Young, Robert M., 91
Winsor, Justin, 367 Yovits, Marshall C., 140n, 201, 649n
Winston, Patrick H., 219, 268
Wisdom, John Oulton, 65 Zadeh, Lofti A., 455, 527n
Wittgenstein, Ludwig von, 265, 270, 592 Zeleny, Milan, 452n, 599n
Wohler, Friedrich, 193 Zermelo, Ernst Friedrich Ferdinand, 64
Wonham, W. Murray, 527n Ziff, Paul, 216
Woods, William A., 34, 213 Zimmermann, Manfred, 425, 652n
Woodworth, Robert Sessions, 422, 423 Zloof, Moshe M., 153
Wright, H. Curtis, 140, 383-385 Zmud, Robert W., 630n
Wright, Herbert F., 64 Znaniecki, Florian, 152
Wundt, Wilhelm, 194 ZOG, 173
Zunde, Pranas, 392
Xerox Corporation, 74n Zurif, Edgar, 74n
Zwass, Vladimir, viii, 151, 179-181
Yale University, viii, 74n, 269, 281, 435 Zworykin, Vladimir, 260
 date due
r
--- T"

---
O-1£-la- - -
J- - ^ t.-rVK

ncn 1 \ nm
——--

CARR McLEAN, TORONTO FORM #38-297

 Z 665 .S826 1983
010101 000
The Study of information : int

0 1163 0073209 O'

TRENT UNIVERSITY

Z665 -S826 1983

The Study of information:

R£S£RV£ 374493

Reserve: Comp St. 320; 10/21/85

374493
ISBN 0-471-88717-X