Developments in Business Simulation & Experiential Exercises, Volume 15, 1988

EXPERT SYSTEMS - THE NEW BUSINESS SIMULATION TOOL

Norman E. Sondak, San Diego State University

Robert O. Briggs, San Diego State University

ABSTRACT
Chess programs sere developed that played well enough to
Expert systems are receiving a tremendous amount of attention challenge expert players. The checkers program defeated the
in the data processing and business communities. But few world champion. But, despite these initial successes intrinsic
business educators have been exposed to the base concepts of problems remained. A! researchers discovered it was very
expert systems or can appreciate the effect they will have on difficult to generalize the experience gained in these early game
business. This presentation covers the fundamental principles of playing programs into handling real world situations.
the design and development of expert systems in business.
Special emphasis is placed ability of an expert system to Al research then focused on natural language processing.
simulate the behavior of a manager or executive in a decision However, processing complex linguistic and semantic structures
making situation. The structure and operation of a typical expert proved to be far more formidable a task than originally thought.
systems is presented. The use of expert system development While simple cases could be managed with little difficulty,
tools and expert system shells is discussed. An example is given automatically handling the ambiguity that humans routinely
of how an expert system can be used to capture and simulate encounter was a far different story. It was soon realized that a
management knowledge and the decision making process. As staggering amount of information is utilized by a person in
expert system development tools become commonplace, digesting even the most rudimentary prose.
business educators must become as proficient with expert
systems as they are now with spreadsheets. Pressures of the economy forced Al research into more
commercial areas. Development of industrial robots became an
important research goal. Again, initial success was encouraging.
INTRODUCTION But, the real AI breakthrough was to come in a far different area
of study.
Expert systems are receiving a tremendous amount of attention
in the data processing and business communities. Settanni Medicine always seemed like an ideal Al application area.
(1987) Nearly every computer magazine or journal contains a Medical Artificial Intelligence (MAT) was defined as the use of
report on expert systems. The article often begins by describing Artificial Intelligence techniques and computational support to
expert systems as the exciting new frontier of computer science. simulate the mental processes a physician exerts in the
However, the facts are very different. Work on the first expert treatment of patients. This definition allowed computer
system, MYCIN, began in the early 70s and there were several scientists to consider what was required to capture and simulate
well established expert systems in place more than 10 years the expertise of the physician. Clancey (1994), Buchanan
ago. Why, then are expert systems receiving so much attention (1984), Fischer (1987), Smith (1996)
today And, why is it so critical for business educators to become
informed about Expert Systems? Finkel (1985), Guterl (1986).
KNOWLEDGE-BASED SYSTEMS
There are three important reasons. First, expert systems permit
the computer to simulate human reasoning and expertise. The seminal idea needed to simulate human expertise came
Therefore, expert systems can be applied to a very wide range from a source far removed from either game playing or
of real life problems. If the problem solution can be expressed medicine. A program called DENDRAL, used to assist chemists
in terms of a knowledge base of rules and facts, then an expert in the determination of the structure of organic compounds from
system can be used. Second, modern expert system mass spectrometer data, derived its power not from automating
development tools allow even data processing novices to logical functions, as so many Al programs had done before, but
construct professional quality expert systems, provided they are from processing tremendous volumes of information about the
familiar with the basic principles of expert system development. spectra of thousands of different chemical compounds
Finally, expert systems sill be common in forthcoming business embedded within the program. Mishkoff (1995)
software as embedded “intelligent front ends.’ And, the
business person must understand the strengths and weaknesses The program’s developer, Edward Feigenbaum’s of Stanford,
of these embedded expert systems. Malone (1986) contended that “intelligence’ in real world situations comes
from the proper employment of enormous amounts of specific
THE ORIGIN OF EXPERT SYSTEMS knowledge about a subject, not from sophisticated logical
operations. In short, Feigenbaum’s maintained that a human
expert uses compiled data and experience encapsulated as “rules
Expert systems are a natural outgrowth of Artificial Intelligence of thumb” or heuristics to solve problems. And that intelligent
research. The stated goal of Artificial Intelligence (Al) research programs would have to process this type “knowledge’ to
is to duplicate the functioning of the human brain by computer successfully emulate human experts.
programs. Al was pioneered and developed at major university
laboratories in £111, Stanford, Cal Berkeley, Carnegie- Mellon, EXPERT SYSTEMS
and Yale. Mishkoff (1995)
Another research team at Stanford carried Feigenbaum’s
Initially, Al was concerned with the computerization of logic, as concept of knowledge-based systems forward with the
it s used in problem solving. Board games like chess and development of MYCIN. Clancey (1994),Buchanan (1994)
checkers, along with the proof of mathematical theorems, MYCIN was designed to act as an “intelligent consultant” to
provided the type of highly structured environment where early physicians, advising them on the diagnosis
AI researchers could test their theories and programming skills.
Mishkoff (1985)

72
 Developments in Business Simulation & Experiential Exercises, Volume 15, 1988
and treatment of bacterial infections. In essence, MYCIN was to EXPERT SYSTEM SHELLS
simulate the decision-making powers of an expert physician.
The MYCIN team realized that the user interface and inference
The “knowledge” about bacterial infections was stored within engine portions of the program could stand alone and form an
MYCIN as a series of IF/THEN/ELSE “production” rules. “expert system shell.” Now, new expert systems could be
These rules appeared in the form: constructed by subject matter experts merely by creating a new
knowledge base. They called their expert system shell
IF 1)the stain of the organism is gramneg. And EMYCIN.
2)the morphology of the organism is rod, and The potential of production rule-based expert systems using
expert system shells was soon recognized by business. A
3)the aerobicity of the organism is aerobic number of commercially successful expert systems were
developed in the early ‘50s. However, the initial expert systems
THEN there is strongly suggestive evidence (0.9) that the class were written in a specialized Al programming language called
of the organism is enterobateriaceae. LISP. LISP programs were slow in execution. Schutzer (1997)
Expert systems with large knowledge bases like MYCIN
The IF portion of the rule is called the “premise” and the THEN required large and expensive mainframe computers to complete
portion of the rule is called the “conclusion.” If the data in the consultations in any reasonable time period.
premise is true, then the facts contained in the conclusion can
be use to find a solution to a problem. To improve performance and reduce development time for
expert systems, special LISP machines and workstations were
Because medicine rarely has clear cut ‘yes or no” cases, developed. These devices used the LISP language for basic
MYCIN rules also allowed a Certainty Factor. The Certain machine instructions and could execute Al programs much
Factor (0.8 in rule above) indicates the degree of brief in the faster than general purpose computers. But, LISP machines
applicability of the rule to the given situation. The knowledge were costly and few organizations could commit to specialized
base for MYCIN contained about 550 such rules. computers that were restricted to expert systems applications.
MYCIN also permitted the rules to be displayed so that PERSONAL COMPUTER EXPERT SYSTEM
physicians could "trace" the “logic behind MYCIN’s “advice” DEVELOPMENT TOOLS
to the user.
The business computer environment changed radically with the
When tested against physicians with an actual set of case introduction of the IBM PC. Expert system shells were
studies, MYCIN performed admirably. And, since the program developed for personal computers. However, the initial
simulated an expert in the field, it was naturally called an packages were still costly and the PC was slow in executing
‘expert system. A powerful new area of Al was established as a expert system programs. When the PC/AT and AT compatibles
result of the Stanford became the standard for business personal computers full-sized
research effort. desk top expert systems were practical.
THE STRUCTURE OF AN EXPERT SYSTEM In addition, the cost for expert system development tools
dropped steadily. Currently, dozens of sophisticated expert
The MYCIN program could logically be broken down into three systems development tools (ESDT) are available in a price
separate parts. the first of these was the knowledge base that range from several thousand to several hundred dollars.
contained the rules for diagnosis and treatment. This section Packages which fall into this price range include GURU,
contained the “expert” knowledge the program needed to solve Personal Consultant Plus, Ml, Microexpert, ESIE, Expert Ease,
problems. The second section was the mechanism required to and EXSYS. But the real price break in software occurred in
find the proper rules for a particular case. This part was named early 1987 when Paperback Software delivered its VP-EXPERT
the “inference engine.” The last section was the “user system.
interface.” It collected the information needed by inference
engine from the user, displayed results, and facilitated updating VP-EXPERT offers tools and capabilities of top of the line PC
and modifying the knowledge base by the system developers. ESOT packages at a price in the $100 range.
Mishkoff (l985(,Clancey (1984), Buchanan 1998
The package received outstanding reviews Siegel (19871, and
The inference engine carried out the logical processes needed to within three months of availability, Paperback Software
offer sound advice. This was done by “backward” chaining indicated that more units of VP- EXPERT were delivered than
through the knowledge base. With backward chaining the all of the other PC packages put together’
inference enqine assumed a specific “goal state’ or final
diagnosis. This diagnosis is contained in a rule conclusion. The AN EXPERT SYSTEM SIMULATING MANAGEMENT
inference engine would examine the premise of that rule, and DECISION MAKING
then find a rule where that premise appeared in a rule
conclusion. The inference engine could then “chain” through A common problem facing sales executives is the proper
the knowledge base verifying that the assumed diagnosis was strategy for deploying a sales force in the field. The types of
consistent with the “initial state” --the symptoms and test results strategies available to select selling effort allocation methods
for the patient. If the symptoms matched the disease, than the are discussed in detail by LaForge, Cravens, and Young
diagnosis was correct. If not a new diagnosis was assumed and LaForge (1996). The type of allocation strategy recommended
the process repeated. depends on a number of factors including customer variations
among the territories, the effort required to make a sale, the
Backward chaining is like first looking at the answers on a mathematical skills of the executive, the computers available to
multiple choice test and then testing each to determine if they the organization, and even the reason why this strategy review
meet the question data. Backward chaining works best if there is being performed.
are few goal states and many initial states. It is very efficient in
many practical situations and many of us use backward The authors have taken the knowledge presented in LaForge,
chaining In everyday problem solving. Cravens. and Young LaForge (1996) and con-

73
 Developments in Business Simulation & Experiential Exercises, Volume 15, 1988
structed a expert system using VP-EXPERT to offer advice on
sales force allocation methodology. The goal of this expert
system is to illustrate how an expert system can be constructed
to simulate and automate the reasoning process proposed in the
article. indeed, the popular literature already describes the
potential of computerized marketing expertise McCann (1987).
VP-EXPERT offers the designer of an expert system a coherent
set of tools to construct the knowledge base. The knowledge
base can be though of as containing the rules that capture the
expertise to solve the problem. Rules can be augmented with
WHY clauses that explain the reason each rule is used for a
given situation. ASK statements are also utilized to that provide
information to the user on how to enter data. In addition, there
are a number of procedural statements and declarations to assist
the program managing the display of questions and results.
Table 1 shows an edited version of the rules portion of
knowledge base for DEPLOY. These rules were derived from
the information contained in the article. The rule format is
similar to that described for the MYCIN program. The rule
premises contain conditional logical expressions. The rule
conclusions indicate the action to be taken if the ;remise is true.
The CNF values in the rule conclusions indicate the Certainty
Factors associated with that particular rule. The semicolon [;] is
a delimiter used by VP- EXPERT to indicate the end of rule.
Table 2 shows some of the 4V statements used in the DEPLOY
knowledge base. The CHOICES statement allows VP-EXPERT
to restrict the user resources to a given set of replies.
Figures 1, 2. 3, and 4 show a topical consultation dialog with
VP-EXPERT. The program can be run in two modes, the
production or RUNTIME mode and the testing or TRACE
mode. These first four figures show the RUNTIME user
interface.
Figure 1 illustrates a typical opening screen used to introduce
the export system to the user. Figure 2 illustrates the results of
the consultation. In this case, two different strategies were
plausible, with one, the Empirical model, showing a higher
degree of confidence. Figure 3 shows the questions asked by the
system of the user, this screen shows three questions. Normally
a single question would be displayed and then the screen
cleared before the next one is asked. Figure ‘i shows what
happens if the user inquired why a question was asked. In this
case, the user wanted to know why the expert system asked
about the computer facilities of the company
In the TRACE mode, the inner workings of the inference engine
are made visible as it chains through the knowledge base to
select various rules. This mode is used to debug the knowledge
case. The screen is divided into three windows, The too window
shows the user dialog, the one the bottom two windows show
the rule generating the Question while the other shows the
values the user entered as answers to the questions. Figure 5
shows the internal status of the expert system during a
consultation. While Figure shows a typical TRACE display at
the conclusion of consultation.
CONCLUSION

Expert system development tools are now in the class of

electronic spreadsheets. Any business persons with access to a
personal computer can avail themselves of this powerful
instrument for capturing and automating complex business
decisions. Reinstein (1986), Keim (l986), Deitz (1996) VP-
EXPERT broke the price barrier by offering as impressive
ESDT at cost within the reach of any company. Now, business
educators must be in a position to understand expert system
applications and illustrate their use so that the next generation
of executives and managers are as comfortable with automating
complex decision processes as they are now at automating
complex calculations.

74
 Developments in Business Simulation & Experiential Exercises, Volume 15, 1988
REFERENCES

Settanni, Joseph Andrew, (19971 ‘Artificial intelligence: next

evolution of software. The Office, v105, February, p40-
42.
Finkel, James 1., 1985) “Few people are knowledgeable about
expert systems and fewer still are aware of the strictures
on how to get information into the computer.” School
Product News, v24, December, p14.
Mishkoff, H., (1985) Understanding Artificial Intelligence,
Texas instruments Learning Center, Dallas, TX
Clancey, W., and Shortliffe, E., Eds., (1984) Readings in
Medical Artificial Intelligence, Addison-Wesley
Buchanan, B. and Shortliffe, E., Eds., (1984) Rule- Based
Expert Systems, Addison-Wesley
Fischer, M., and Firschcein, D. (1987), Intelligence: The Eye,
the Brain, and the Computer Addison- Wesley
Smith. A., “Bibliography, Artificial Intelligence,” Journal of
Computers in Mathematics and Science Teaching, vS
n3, p67-68, So.
Schutzer, D. 1987, Artificial intelligence: An Applications-
oriented Approach, Van Nostrand Reinhold
GURU, MDBS, PD Box EL*8, Lafayette, IN 47902
Personal Consultant Plus 1e<as Instruments, PD Box 8090a3.
Dallas, TX 75380
Nil, Teknowledge, 525 University Aye, Suite 200, Palo Alto,
CA 94301
TABLE 2 -- ASK AND CHOICES STATEMENTS USED 10 Micro Expert, McGraw-Hill. NY. NY
COLLECT INFORMATION FROM THE USER DURING
THE CONSULTATION ESIE, Lightwave Consultants, P.O. Box 290539, Tampa, FL
33617
ASK Change In Sales:
If your sales people called on customers half as frequently Expert Ease, Human Edge Software, 2445 Faber Place, Palo
as they currently do would your sales drop dramatically Alto, CA 94303
CHOICES Change In Sales” yes/no: EXSYS, EXSYS, PO Box 75158. Contract Station 14,
Albuquerque, NM 87194
ASK Market Concentration:
Is the concentration of customers quite similar from VP-EXPERT, Paperback Software International, 2930 Ninth
sales territory to sales Street, Berkeley, CA 94710.
territory or are customer concentrations
different in different territories?; CHOICES Market
Concentration: similar, different; Siegel, P. (1987), Inexpensive Program Designs Expert Systems
for Novices, INFOWORLD (Sept. 21), p67
ASK Computer:
What is the most powerful Computer to which your LaForge, Raymond W. and David W. Cravens, and Clifford E
company has access?; Young. (1986), Using Contingency analysis to select
CHOICES Computer: MicroComputer, MiniComputer, selling effort allocation methods.” Journal of Personal
MainFrame, None; Selling & Sales Management (August). p19-28.

49K Use_Other_Models: McCann, John M.(1987), “First steps toward the expert
Does your company use computerized mathematical marketing computer,” Marketing & Media Decisions,
models to aid in the decision-making process?”; v22 (February), p36.
CHOICES Use_Other_Models: Yes/No;
Malure, Thomas W. and Michael S. Scott Morton., (1996)
ASK Purpose_Of_Analysis: Expert systems: the newt challenge for managers. Sloan
Are you planning to make small, incremental adjustments Management Review, v27 (Summer), p3-14.
to your sales force or a total reorganization or your selling
effort?’; Reinstein, Harry C. 1996), Expert systems generate applications
CHOICES Purpose_Of_Analysis: Small Adjustment, Iota without magic.” Data Management, v24 (December,
1 Reorganization; p26-29.

75
 Developments in Business Simulation & Experiential Exercises, Volume 15, 1988

Guterl, Fred V. and Leah Nathans., (1996) “Computers think

for Business; after years En the laboratory, expert
systems are becoming key decision-making tools for
companies across the industrial spectrum.’ Dun’s
Business Month, v128, October, p30-36.
Gold, Jordan., “Do-it-yourself expert systems. Computer
Decisions, v18, January, p301-321.

Keim, Robert 1. and Sheila Jacobs (1996), ‘Expert systems: the

DSS of the future?” Journal of Systems Management,
v37, December, p6-14.

Dietz, P.W. (1996), “Artificial Intelligence: building rule-based

expert systems.” Design News, v42, March, p137-142.

76
Developments in Business Simulation & Experiential Exercises, Volume 15, 1988

77