THE PHILOSOPHICAL BASIS OF

INTUITIONISTIC LOGIC

Michael DUMMETT
University of Oxford, U.K .

The question with which I am here concerned is: What plausible

rationale can there be for repudiating, within mathematical reasoning, the
canons of classical logic in favour of those of intuitionistic logic? I am,
thus, not concerned with justifications of intuitionistic mathematics from
an eclectic point of view, that is, from one which would admit intuitionis-
tic mathematics as a legitimate and interesting form of mathematics
alongside classical mathematics: I am concerned only with the standpoint
of the intuitionists themselves, namely that c!assical mathematics emp-
loys forms of reasoning which are not valid on any legitimate construal of
mathematical statements (save, occasionally, by accident, as it were,
under a quite unintended reinterpretation). Nor am I concerned with
exegesis of the writings of Brouwer or of Heyting: the question is what
forms of justification of intuitionistic mathematics will stand up, not what
particular writers, however eminent, had in mind. And, finally, I am
concerned only with the most fundamental feature of intuitionistic
mathematics, its underlying logic, and not with the other respects (such as
the theory of free choice sequences) in which it differs from classical
mathematics. It will therefore be possible to conduct the discussion
wholly at the level of elementary number theory. Since we are, in effect,
solely concerned with the logical constants-with the sentential operators
and the first-order quantifiers-our interest lies only with the most general
features of the notion of a mathematical construction, although it will be
seen that we need to consider these in a somewhat delicate way.
Any justification for adopting one logic rather than another as the logic
for mathematics must turn on questions of meaning. It would be impos-
6 MICHAEL DUMMETT

sible to contrive such a justification which took meaning for granted, and
represented the question as turning on knowledge or certainty. We are
certain of the truth of a statement when we have conclusive grounds for it
and are certain that the grounds which we have are valid grounds for it
and are conclusive. If classical arguments for mathematical statements
are called in question, this cannot possibly be because it is thought that we
are, in general, unable to tell with certainty whether an argument is
classically valid, unless it is also intuitionistically valid: rather, it must be
that what is being put in doubt is whether arguments which are valid by
clas~icalbut not by intuitionistic criteria are absolutely valid, that is,
whether they really conclusively establish their conclusions as true. Even
if it were held that classical arguments, while not in general absolutely
valid, nevertheless always conferred a high probability on their conclu-
sions, it would be wrong to characterise the motive for employing only
intuitionistic arguments as lying in a desire to attain knowledge in place of
mere probable opinion in mathematics, since the very thesis that the use
of classical arguments did not lead to knowledge would represent the
crucial departure from the classical conception, beside which the question
of whether or not one continued to make use of classical arguments as
mere probabilistic reasoning is comparatively insignificant. (In any case,
within standard intuitionistic mathematics, there is no reason whatever
why the existence of a classical proof of it should render a statement
probable, since if, e.g., it is a statement of analysis, its being a classical
theorem does not prevent it from being intuitionistically disprovable.)
So far as I am able to see, there are just two lines of argument for
repudiating classical reasoning in mathematics in favour of intuitionistic
reasoning. The first runs along the following lines. The meaning of a
mathematical statement determines and is exhaustively determined by its
use. The meaning of such a statement cannot be, or contain as an
ingredient, anything which is not manifest in the use made of it, lying
solely in the mind of the individual who apprehends that meaning: if two
individuals agree completely about the use to be made of the statement,
then they agree about its meaning. The reason is that the meaning of a
statement consists solely in its role as an instrument of communication
between individuals, just as the powers of a chesspiece consist solely in
it5 role in the game according to the rules. An individual cannot
communicate what he cannot be observed to communicate: if one
individual associated with a mathematical symbol or formula some mental
content, where the association did not lie in the use he made of the symbol
 T H E PHILOSOPHICAL BASIS OF lNTUITIONlSTIC LOGIC 7

or formula, then he could not convey that content by means of the symbol
or formula, for his audience would be unaware of the association and
would have no means of becoming aware of it.
The argument may be expressed in terms of the knowledge of meaning,
i,e. of understanding. A model of meaning is a model of understanding,
i.e. a representation of what it is that is known when an individual knowc
the meaning. Now knowledge of the meaning of a particular symbol or
expression is frequently verbalisable knowledge, that is, knowledge
which consists in the ability to state the rules in accordance with which
the expression or symbol is used or the way in which it may be replaced
by an equivalent expression or sequence of symbols. But to suppose that,
in general, a knowledge of meaning consisted in verbalisable knowledge
would involve an infinite regress: if a grasp of the meaning of an
expression consisted, in general, in the ability to state its meaning, then it
would be impossible for anyone to learn a language who was not already
equipped with a fairly extensive language. Hence that knowledge which,
in general, constitutes the understanding of the language of mathematics
must be implicit knowledge. Implicit knowledge cannot, however, mean-
ingfully be ascribed to someone unless it is possible to say in what the
manifestation of that knowledge consists: there must be an observable
difference between the behaviour or capacities of someone who is said to
have that knowledge and someone who is said to lack it. Hence it follows,
once more, that a grasp of the meaning of a mathematical statement m u 4 ,
in general, consist of a capacity to use that statement in a certain way, or
to respond in a certain way to its use by others.
Another approach is via the idea of learning mathematics. When we
learn a mathematical notation, or mathematical expressions, or, more
generally, the language of a mathematical theory, what we learn to do ic to
make use of the statements of that language: we learn when they may be
established by computation, and how to carry out the relevant computa-
tions, we learn from what they may be inferred and what may be inferred
from them, that is, what role they play in mathematical proofs and how
they can be applied in extra-mathematical contexts, and perhaps we learn
also what plausible arguments can render them probable. These things are
all that we are shown when we are learning the meanings of the
expressions of the language of the mathematical theory in qiiestion,
because they are all that we can be shown: and, likewise, our proficiency
in making the correct use of the statements and expressions of the
language is all that others have from which to judge whether or not we
8 M I<')+ A EL DUMME T I

have acquired a grasp of their meanings. Hence it can only be in the

capacity to make a correct use of t h e statements of the language that a
grasp of their meanings, and those of the symbols and expressions which
they contain, can consist. To suppose that there is an ingredient of
meaning which transcends the use that is made of that which carries the
meaning is to suppose that someone might have learned all that is directly
taught when the language of a mathematical theory is taught to him, and
might then behave in every way like someone who understood that
language, and yet not actually understand it, or understand it only
incorrectly. But to suppose this is to make meaning ineffable, that is, in
principle incommunicable. If this is possible, then no one individual ever
has a guarantee that he is understood by any other individual; for all he
knows, or can ever know, everyone else may attach to his words or to the
symbols which he employs a meaning ,luite different from that which he
attaches to them. A notion of meaning so private to the individual is one
that has become completely irrelevant to mathematics as it is actually
practised, namely as a body of theory on which many individuals are
corporately engaged, an enquiry within which each can communicate his
results to others.
I t might seem that an approach to meaning which regarded it as
exhaustively determined by use would rule out any form of revisionism.
If use constitutes meaning, then, it might seem, use is beyond criticism:
there can be no place for rejecting any established mathematical practice,
such as the use of certain forms of argument or modes of proof, since that
practice, together with all others which are generally accepted, is simply
constitutive of the meanings of our mathematical statements, and we
surely have the right to make our statements mean whatever we choose
that they shall mean. Such an attitude is one possible development of the
thesis that use exhaustively determines meaning: it is, however, one
which can, ultimately, be supported only by the adoption of a holistic
view of language. On such a view, it is illegitimate to ask after the content
of any single statement, or even after that of any one theory, say a
mathematical or a physical theory; the significance of each statement or
of each deductively systematised body of statements is modified by the
multiple connections which it has, direct and remote, with other state-
ments in other areas of our language taken as a whole, and so there is no
adequate way of understanding t h e statement short of knowing the entire
language. Or. rather, even this image is false to the facts: it is not that a
statement or even a theory has, as it were, a primal meaning which then
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 9

gets modified by the interconnections that are established with other

statements and other theories; rather, its meaning simply consists in the
place which it occupies in the complicated network which constitutes the
totality of our linguistic practices. The only thing to which a definite
content may be attributed is the totality of all that we are, at a given time,
prepared to assert; and there can be no simple model of the content which
that totality of assertions embodies; nothing short of a complete know-
ledge of the language can reveal it.
Frequently such a holistic view. is modified to the extent of admitting a
class of observation statements which can be regarded as more or less
directly registering our immediate experience, and hence as each carrying
a determinate individual content. These observation statements lie, in
Quine’s famous image of language, at the periphery of the articulated
structure formed by all the sentences of our language, where alone
experience impinges. To these peripheral sentences, meanings may be
ascribed in a more or less straightforward manner, in terms of the
observational stimuli which prompt assent to and dissent from them. No
comparable model of meaning is available for the sentences which lie
further towards the interior of the structure: an understanding of them
consists solely in a grasp of their place in the structure as a whole and
their interaction with its other constituent sentences. Thus, on such a
view, we may accept a mathematical theory, and admit its theorems as
true, only because we find in practice that it serves as a convenient
substructure deep in the interior of the complex structure which forms the
total theory: there can be no question of giving a representation of the
truth-conditions of the statements of the mathematical theory under
which they may be judged individually as acceptable, or otherwise. in
isolation from the rest of language.
Such a conception bears an evident analogy with Hilbert’s view of
classical mathematics; or, more accurately, with Boole’s view of his
logical calculus. For Hilbert, a definite individual content, according to
which they may be individually judged as correct or incorrect, may
legitimately be ascribed only to a very narrow range of statements of
elementary number theory: these correspond to the observation state-
ments of the holistic conception of language. All other statements of
mathematics are devoid of such a content, and serve only as auxiliaries,
though psychologically indispensable auxiliaries, t o the recognition as
correct of the finitistic statements which alone are individually meaning-
ful. The other mathematical statements are not, on such a view, devoid of
10 MICHAEL DUMMETT

significance: but their significance lies wholly in the role which they play
within the mathematical theories to which they belong, and which are
themselves significant precisely because they enable us to establish the
correctness of finitistic statements. Boole likewise distinguished, amongst
the formulas of his logical calculus, those which were interpretable from
those which were uninterpretable: a deduction might lead from some
interpretable formulae as premisses, via uninterpretable formulae as
intermediate steps, to a conclusion which was once more interpretable.
The immediately obvious difficulty about such a manner of construing a
mathematical, or any other, theory is to know how it can be justified. How
can we be sure that the statements or formulae to which we ascribe a
content, and which are derived by such a means, are true? The difference
between Hilbert and Boole, in this respect, was that Hilbert took the
demand for justification seriously, and saw the business of answering it as
the prime task for his philosophy of mathematics, while Boole simply
ignored the question. Of course, the most obvious way to find a
justification is to extend the interpretation to all the statements or
formulae with which we are concerned, and, in the case of Boole’s
calculus, this is very readily done, and indeed yields a great simplification
of the calculus. Even in Hilbert’s case, the consistency proof, once found,
does yield an interpretation of the infinitistic statements, though one
which is relative to the particular proof in which they occur, not one
uniform for all contexts. Without such a justification, the operation of the
mechanism of the theory or the language remains quite opaque to us; and
it is because the holist is oblivious of the demand for justification, or of
the unease which the lack of one causes us, that I said that he is to be
compared to Boole rather than to Hilbert. In his case, the question would
become: With what right do we feel an assurance that the observation
statements deduced with the help of the complex theories, mathematical,
scientific and otherwise, embedded in the interior of the total linguistic
structure, are true. when these observation statements are interpreted in
terms of their stimulus meanings? To this the holist attempts no answer,
save a generalised appeal to induction: these theories have ‘worked’ in the
past, in the sense of having for the most part yielded true observation
statements, and so we have confidence that they will continue to work in
the future.
The path of thought which leads from the thesis that use exhaustively
determines meaning to an acceptance of intuitionistic logic as the correct
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 11

logic for mathematics is one which rejects a holistic view of mathematics,

and insists that each statement of any mathematical theory must have a
determinate individual content. A grasp of this content cannot, in general,
consist of a piece of verbalisable knowledge, but must be capable of being
fully manifested by the use of the statement: but that does not imply that
every aspect of its existing use is sacrosanct. An existing practice in the
use of a certain fragment of language is capable of being subjected to
criticism if it is impossible to systematise it, that is, to frame a model
whereby each sentence carries a determinate content which can, in turn,
be explained in terms of the use of that sentence. What makes it possible
that such a practice may prove to be incoherent and therefore in need of
revision is that there are different aspects to the use of a sentence; if the
whole practice is to be capable of systematisation in the present sense,
there must be a certain harmony between these different aspects. This is
already apparent from the holistic examples already cited. One aspect of
the use of observation statements lies in the propensities we have
acquired to assent to and dissent from them under certain types of
stimuli; another lies in the possibility of deducing them by means of
non-observational statements, including highly theoretical ones. If the
linguistic system as a whole is to be coherent, there must be harmony
between these two aspects: it must not be possible to deduce observation
statements from which the perceptual stimuli require dissent. Indeed, if
the observation statements are to retain their status as observation
statements, a stronger demand must be made: of an observation state-
ment deduced by means of theory, it must hold that we can place
ourselves in a situation in which stimuli occur which require assent to it.
This condition is thus a demand that, in a certain sense, the language as a
whole be a conservative extension of that fragment of the language
containing only observation statements. In just the same way, Hilbert’s
philosophy of mathematics requires that classical number theory, or even
classical analysis, be a conservative extension of finitistic number theory.
For utterances considered quite generally, the bifurcation between the
two aspects of their use lies in the distinction between the conventions
governing the occasions on which the utterance is appropriately made and
those governing both the responses of the hearer and what the speaker
commits himself to by making the utterance: schematically, between the
conditions for the utterance and the consequences of it. Where, as in
mathematics, the utterances with which we are concerned are statements,
12 MICHAEL D U M M E T T

that is, utterances by means of which assertions can be effected, this

becomes the distinction between the grounds on which the statement can
be asserted and its inferential consequences, the conclusions that can be
inferred from it. Plainly, the requirement of harmony between these in
respect of some type of statement is the requirement that the addition of
statements of that type to the language produces a conservative extension
of the language; i.e., that it is not possible, by going via statements of this
type as intermediaries, to deduce from premisses not of that type
conclusions, also not of that type, which could not have been deduced
before. In the case of the logical constants, a loose way of putting the
requirement is to say that there must be a harmony between the
introduction and elimination rules; but, of course, this is not accurate,
since the whole system has to be considered (in classical logic, for
example, it is possible to infer a disjunctive statement, say by double
negation elimination, without appeal to the rule of disjunction introduc-
tion). An alternative way of viewing the dichotomy between the two
principal aspects of the use of statements is as a contrast between direct
and indirect means of establishing them. So far as a logically complex
statement is concerned, the introduction rules governing the logical
constants occurring in the statement display the most direct means of
establishing the statement, step by step in accordance with its logical
structure; but the statement may be accepted on the basis of a compli-
cated deduction which relies also on elimination rules, and we require a
harmony which obtains only if a statement that has been indirectly
established always could (in some sense of ‘could’) have been established
directly. Here again the demand is that the admission of the more
complex inferences yield a conservative extension of the language. When
only introduction rules are used, the inference involves only statements
of logical complexity no greater than that of the conclusion: we require
that the derivation of a statement by inferences involving statements of
greater logical complexity shall be possible only when its derivation by
the more direct means is in some sense already possible.
On any molecular view of language-any view on which individual
sentences carry a content which belongs to them in accordance with the
way they are compounded out of their own constituents, independently of
other sentences of the language not involving those constituents-there
must be some demand for harmony between the various aspects of the
use of sentences, and hence some possibility of criticising or rejecting
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC L.OGIC 13

existing practice when it does not display the required harmony. Exactly
what the harmony is which is demanded depends upon the theory of
meaning accepted for the language, that is, the general model of that in
which the content of an individual sentence consists; that is why I
rendered the above remarks vague by the insertion of phrases like ‘in
some sense’. It will always be legitimate to demand, of any expression or
form of sentence belonging to the language, that its addition to the
language should yield a conservative extension; but, in order to make the
notion of a conservative extension precise, we need to appeal to some
concept such as that of truth or that of being assertible or capable in
principle of being established, or the like; and just which concept is to be
selected, and how it is to be explained, will depend upon the theory of
meaning that is adopted.
A theory of meaning, at least of the kind with which we are mostly
familiar, seizes upon some one general feature of sentences (at least of
assertoric sentences, which is all we need be concerned with when
considering the language of mathematics) as central: the notion of the
content of an individual sentence is then to be explained in terms of this
central feature. The selection of some one such feature of sentences as
central to the theory of meaning is what is registered by philosophical
dicta of the form, ‘Meaning is ...’- e.g. ‘The meaning of a sentence is the
method of its verification’, ‘The meaning of a sentence is determined by
its truth-conditions’, etc. (The slogan ‘Meaning is use’ is, however, of a
different character: the ‘use’ of a sentence is not, in this sense, a single
feature; the slogan simply restricts the kind of feature that may legiti-
mately be appealed to as constituting or determining meaning.) The
justification for thus selecting some one single feature of sentences as
central-as being that in which their individual meanings consist-is that it
is hoped that every other feature of the use of sentences can be derived,
in a uniform manner, from this central one. If, e.g., the notion of truth is
taken as central to the theory of meaning, then the meanings of individual
expressions will consist in the manner in which they contribute to
determining the truth-conditions of sentences in which they occur; but
this conception of meaning will be justified only if it is possible, for an
arbitrary assertoric sentence whose truth-conditions are taken as known,
to describe, in terms of the notion of truth, our actual practice in the use
of such a sentence; that is, to give a general characterisation of the
linguistic practice of making assertions, of the conditions under which
14 MICHAEL DUMMETT

they are made and the responses which they elicit. Obviously, we are very
far from being able to construct such a general theory of the use of
sentences, of the practice of speaking a language; equally obviously, it is
likely that, if we ever do attain such an account, it will involve a
considerable modification of the ideal pattern under which the account
will take a quite general form, irrespective of the individual content of the
sentence as given in terms of whatever is taken as the central notion of
the theory of meaning. But it is only to the extent that we shall eventually
be able to approximate to such a pattern that it is possible to give
substance to the claim that it is in terms of some one feature, such as truth
or verification, that the individual meanings of sentences and of their
component expressions are to be given.
It is the multiplicity of the different features of the use of sentences,
and the consequent legitimacy of the demand, given a molecular view of
language, for harmony between them, that makes it possible to criticise
existing practice, to call in question uses that are actually made of
sentences of the language. The thesis with which we started, that use
exhaustively determines meaning, does not, therefore, conflict with a
revisionary attitude to some aspect of language: what it does d o is to
restrict the selection of the feature of sentences which is to be treated as
central to the theory of meaning. On a platonistic interpretation of a
mathematical theory, the central notion is that of truth: a grasp of the
meaning of a sentence of the language of the theory consists in a
knowledge of what it is for that sentence to be true. Since, in general, the
sentences of the language will not be ones whose truth-value we are
capable of effectively deciding, the condition for the truth of such a
sentence will be one which we are not, in general, capable of recognising
as obtaining whenever it obtains, or of getting ourselves into a position in
which we can so recognise it. Nevertheless, on the theory of meaning
which underlies platonism, an individual’s grasp of the meaning of such a
sentence consists in his knowledge of what the condition is which has to
obtain for the sentence to be true, even though the condition is one which
he cannot, in general, recognise as obtaining when it does obtain.
Such a conception violates the principle that use exhaustively deter-
mines meaning; or, at least, if it does not, a strong case can be put up that
it does, and it is this case which constitutes the first type of ground which
appears to exist for repudiating classical in favour of intuitionistic logic
for mathematics. For, if the knowledge that constitutes a grasp of the
meaning of a sentence has to be capable of being manifested in actual
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 15

linguistic practice, it is quite obscure in what the knowledge of the

condition under which a sentence is true can consist, when that condition
is not one which is always capable of being recognised as obtaining. In
particular cases, of course, there may be no problem, namely when the
knowledge in question may be taken as verbalisable knowledge, i.e. when
the speaker is able to state, in other words, what the condition is for the
truth of the sentence; but, as we have already noted, this cannot be the
general case. An ability to state the condition for the truth of a sentence
is, in effect, no more than an ability to express the content of the sentence
in other words. We accept such a capacity as evidence of a grasp of the
meaning of the original sentence on the presumption that the speaker
understands the words in which he is stating its truth-condition; but at
some point it must be possible to break out of the circle: even if it were
always possible to find an equivalent, understanding plainly cannot in
general consist in the ability to find a synonymous expression. Thus the
knowledge in which, on the platonistic view, a grasp of the meaning of a
mathematical statement consists must, in general, be implicit knowledge,
knowledge which does not reside in the capacity to state that which is
known. But, at least on the thesis that use exhaustively determines
meaning, and perhaps on any view whatever, the ascription of implicit
knowledge to someone is meaningful only if he is capable, in suitable
circumstances, of fully manifesting that knowledge. (Compare Wittgens-
tein’s question why a dog cannot be said to expect that his master will
come home next week.) When the sentence is one which we have a
method for effectively deciding, there is again no problem: a grasp of the
condition under which the sentence is true may be said to be manifested
by a mastery of the decision procedure, for the individual may, by that
means, get himself into a position in which he can recognise that the
condition for the truth of the sentence obtains or does not obtain, and we
may reasonably suppose that, in this position, he displays by his linguistic
behaviour his recognition that the sentence is, respectively, true or false.
But, when the sentence is one which is not in this way effectively
decidable, as is the case with the vast majority of sentences of any
interesting mathematical theory, the situation is different. Since the
sentence is, by hypothesis, effectively undecidable, the condition which
must, in general, obtain for it to be true is not one which we are capable of
recognising whenever it obtains, or of getting ourselves in a position to do
so. Hence any behaviour which displays a capacity for acknowledging the
sentence as being true in all cases in which the condition for its truth can
16 MICHAEL DUMMETT

be recognised as obtaining will fall short of being a full manifestation of

the knowledge of the condition for its truth: it shows only that the
condition can be recognised in certain cases, not that we have a grasp of
what, in general, it is for that condition to obtain even in those cases when
we are incapable of recognising that it does. It is, in fact, plain that the
knowledge which is being ascribed to one who is said to understand the
sentence is knowledge which transcends the capacity to manifest that
knowledge by the way in which the sentence is used. The platonistic
theory of meaning cannot be a theory in which meaning is fully deter-
mined by use.
If to know the meaning of a mathematical statement is to grasp its use;
if we learn the meaning by learning the use, and our knowledge of its
meaning is a knowledge which we must be capable of manifesting by the
use we make of it: then the notion of truth, considered as a feature which
each mathematical statement either determinately possesses or determi-
nately lacks, independently of our means of recognising its truth-value,
cannot be the central notion for a theory of the meanings of mathematical
statements. Rather, we have to look at those things which are actually
features of the use which we learn to make of mathematical statements.
What we actually learn to do, when we learn some part of the language of
mathematics, is to recognise, for each statement, what counts as estab-
lishing that statement as true or as false. In the case of very simple
statements, we learn some computation procedure which decides their
truth or falsity: for more complex statements, we learn to recognise what
is to be counted as a proof or a disproof of them. That is the practice of
which we acquire a mastery: and it is in the mastery of that practice that
our grasp of the meanings of the statements must consist. We must,
therefore, replace the notion of truth, as the central notion of the theory
of meaning for mathematical statements, by the notion of proof: a grasp
of the meaning of a statement consists in a capacity to recognise a proof
of it when one is presented to us, and a grasp of the meaning of any
expression smaller than a sentence must consist in a knowledge of the
way in which its presence in a sentence contributes to determining what is
to count as a proof of that sentence. This does not mean that we are
obliged uncritically to accept the canons of proof as conventionally
acknowledged. On the contrary, as soon as we construe the logical
constants in terms of this conception of meaning, we become aware that
certain forms of reasoning which are conventionally accepted are devoid
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 17

of justification. Just because the conception of meaning in terms of proof

is as much a molecular, as opposed to holistic, theory of meaning as that
of meaning in terms of truth-conditions, forms of inference stand in need
of justification, and are open to being rejected as unjustified. Our
mathematical practice has been disfigured by a false conception of what
our understanding of mathematical theories consisted in.
This sketch of one possible route to an account of why, within
mathematics, classical logic must be abandoned in favour of intuitionistic
logic obviously leans heavily upon Wittgensteinian ideas about language.
Precisely because it rests upon taking with full seriousness the view of
language as an instrument of social communication, it looks very unlike
traditional intuitionist accounts, which, notoriously, accord a minimum of
importance to language or to symbolism as a means of transmitting
thought, and are constantly disposed to slide in the direction of solipsism.
However, I said at the outset that my concern in this paper was not in the
least with the exegesis of actual intuitionist writings: however little it may
gibe with the view of the intuitionists themselves, the considerations that
I have sketched appear to me to form one possible type of argument in
favour of adopting an intuitionistic version of mathematics in place of a
classical one (at least as far as the logic employed is concerned), and,
moreover, an argument of considerable power. I shall not take the time
here to attempt an evaluation of the argument, which would necessitate
enquiring how the Platonist might reply to it, and how the debate between
them would then proceed: my interest lies, rather, in asking whether this
is the only legitimate route to the adoption of an intuitionistic logic for
mathematics.
Now the first thing that ought to strike us about the form of argument
which I have sketched is that it is virtually independent of any considera-
tions relating specifically to the mathematical character of the statements
under discussion. The argument involved only certain considerations
within the theory of meaning of a high degree of generality, and could,
therefore, just as well have been applied to any statements whatever, in
whatever area of language. The argument told in favour of replacing, as
the central notion for the theory of meaning, the condition under which a
statement is true, whether we know or can know when that condition
obtains, by the condition under which we acknowledge the statement as
conclusively established, a condition which we must, by the nature of the
case, be capable of effectively recognising whenever it obtains. Since we
I8 M 1C H A E I ~ I) U M M ETI

were concerned with mathematical statements, which we recognise as

true by means of a proof (or, in simple cases, a computation), this meant
replacing the notion of truth by that of proof: evidently, the appropriate
generalisation of this. for statements of a.1 arbitrary kind, would be the
replacement of the notion of truth, as the central notion of the theory of
meaning, by that of verification; t o know the meaning of a statement is, on
such a view. t o be capable of recognising whatever counts as verifying the
statement. i.e. as conclusively establishing it as true. Here, of course, the
verification woiild not ordinarily consist in the bare occurrence of some
v q u e n c e o f sense-experiences. as on the positivist conception of the
verification of ;L statement. In the mathematical case, that which estab-
lishes ;I statement 21s true is the production of a deductive argument
terniinsting in that statement as conclusion; in the general case, a
statement will. i n general, also be established as true by a process of
reiisoning, though here the reasoning will not usually be purely deductive
in character, and the p r e m i s e s of the argument will be based on
obwxvation; only for a restricted class of statements-the observation
\tatements-will their verification be of a purely observational kind,
without the mediation of any chain of reasoning or any other mental,
linguistic or symbolic proce
It follows that. in s o far as an intuitionist position in the philosophy of
mathematics (or, at least, the acceptance of an intuitionistic logic for
mathematics) is supported by an argument of this first type, similar,
though not necessarily identical, revisions must be made in the logic
accepted for statements of other kinds. What is involved is a thesis in the
theory of meaning of the highest possible level of generality. Such a thesis
is vulnerable in many places: if it should prove that it cannot be
coherently applied t o any one region of discourse, t o any one class of
statements, then the thesis cannot be generally true, and the general
argument in favour o f it must be fallacious. Construed in this way,
therefore, a position in the philosophy of mathematics will be capable of
being undermined by considerations which have nothing directly t o d o
with mat he in at ic s at al I .
Is there, then. any alternative defence of the rejection, for mathema-
tics, o f classical in favour of intuitionistic logic? Is there any such defence
which turns o n the fact that we are dealing with nzatheniatical statements
in particular, and leaves it entirely open whether or not we wish t o extend
the argument t o statements of any other general class?
 THE PHILOSOPHICAL BASIS OF I N T U l T l O N l S T I C IX)GIC 19

Such a defence must start from some thesis about mathematical

statements the analogue of which we are free to reject for statements of
other kinds. It is plain what this thesis must be: namely, the celebrated
thesis that mathematical statements do not relate to an objective
mathematical reality existing independently of us. The adoption of such a
view apparently leaves us free either to reject or to adopt an analogous
view for statements of any other kind. For instance, if we are realists
about the physical universe, then we may contrast mathematical state-
ments with statements ascribing physical properties to material objects:
on this combination of views, material-object statements do relate to an
objective reality existing independently of ourselves, and are rendered
true or false, independently of our knowledge of their truth-values or of
our ability to attain such knowledge or the particular means, if any, by
which we do so, by that independently existing reality: the assertion that
mathematical statements relate to no such external reality gains its
substance by contrast with the physical case. Unlike material objects,
mathematical objects are, on this thesis, creations of the human mind:
they are objects of thought, not merely in the sense that they can be
thought about, but in the sense that their being is to be thought of: for
them, esse est concipi.
On such a view, a conception of meaning as determined by truth-
conditions is available for any statements which do relate to an indepen-
dently existing reality, for then we may legitimately assume. of each such
statement, that it possesses a determinate truth-value, true or false,
independently of our knowledge, according as it does or does not agree
with the constitution of that external reality which it is about. But, when
the statements of some class do not relate to such an external reality, the
supposition that each of them possesses such a determinate truth-value is
empty, and we therefore cannot regard them as being given meanings by
associating truth-conditions with them; we have, in such a case, faute de
rnieux, to take them as having been given meaning in a different way.
namely by associating with them conditions of a different kind-
conditions that we are capable of recognising when they obtain-namely,
those conditions under which we take their assertion or their denial a s
being conclusively justified.
The first type of justification of intuitionistic logic which we considered
conformed to Kreisel’s dictum, ‘The point is not the existence of
mathematical objects, but the objectivity of mathematical truth’: it bore
20 MICHAEL DUMMET’I’

directly upon the claim that mathematical statements possess objective

truth-values, without raising the question of the ontological status of
mathematical objects or the metaphysical character of mathematical
reality. But a justification of the second type violates the dictum: it makes
the question whether mathematical statements possess objective truth-
values depend upon a prior decision as to the being of mathematical
objects. And the difficulty about it lies in knowing on what we are to base
the premiss that mathematical objects are the creations of human thought
in advance of deciding what is the correct model for the meanings of
mathematical statements or what is the correct conception of truth as
relating to them. It appears that, on this view, before deciding whether a
grasp of the meaning of a mathematical statement is to be considered as
consisting in a knowledge of what has to be the case for it to be true or in a
capacity to recognise a proof of it when one is presented, we have first to
resolve the metaphysical question whether mathematical objects-natural
numbers, for example-are, as on the constructivist view, creations of the
human mind, or, as on the Platonist view, independently existing abstract
objects. And the puzzle is to know on what basis we could possibly
resolve this metaphysical question, at a stage at which we do not even
know what model to use for our understanding of mathematical state-
ments. We are, after all, being asked to choose between two metaphors,
two pictures. The Platonist metaphor assimilates mathematical enquiry to
the investigations of the astronomer: mathematical structures, like galax-
ies, exist, independently of us, in a realm of reality which we do not
inhabit but which those of us who have the skill are capable of observing
and reporting on. The constructivist metaphor assimilates mathematical
activity to that of the artificer fashioning objects in accordance with the
creative power of his imagination. Neither metaphor seems, at first sight,
especially apt, nor one more apt than the other: the activities of the
mathematician seem strikingly unlike those either of the astronomer or of
the artist. What basis can exist for deciding which metaphor is to be
preferred? How are we to know in which respects the metaphors are to be
taken seriously, how the pictures are t o be used?
Preliminary reflection suggests that the metaphysical question ought
not to be answered first: we cannot, as the second type of approach would
have u s do, first decide the ontological status of mathematical objects,
and then, with that as premiss, deduce the character of mathematical truth
or the correct model of meaning for mathematical statements. Rather, we
have first to decide on the correct model of meaning-either an intuitionis-
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 21

tic one, on the basis of an argument of the first type, or a platonistic one,
on the basis of some rebuttal of it; and then one or other picture of the
metaphysical character of mathematical reality will force itself on us. If
we have decided upon a model of the meanings of mathematical state-
ments according to which we have to repudiate a notion of truth
considered as determinately attaching, or failing to attach, to such
statements independently of whether we can now, or ever will be able to,
prove or disprove them, then we shall be unable to use the picture of
mathematical reality as external to us and waiting to be discovered.
Instead, we shall inevitably adopt the picture of that reality as being the
product of our thought, or, at least, as coming into existence only as it is
thought. Conversely, if we admit a notion of truth as attaching objectively
to our mathematical statements independently of our knowledge, then,
likewise, the picture of mathematical reality as existing, like the galaxies,
independently of our observation of it will force itself on us in an equally
irresistible manner. But, when we approach the matter in this way, there
is no puzzle over the interpretation of these metaphors: psychologically
inescapable as they may be, their non-metaphorical content will consist
entirely in the two contrasting models of the meanings of mathematical
statements, and the issue between them will become simply the issue as to
which of these two models is correct. If, however, a view as to the
ontological status of mathematical objects is to be treated as a premiss for
deciding between the two models of meaning, then the metaphors cannot
without circularity be explained solely by reference to those models; and
it is obscure how else they are to be explained.
These considerations appear, at first sight, to be reinforced by reflection
upon Frege’s dictum, ‘Only in the context of a sentence does a name stand
for anything’. We cannot refer to an object save in the course of saying
something about it. Hence, any thesis concerning the ontological status of
objects of a given kind must be, at the same time, a thesis about what
makes a statement involving reference to such objects true, in other
words, a thesis about what properties an object of that kind can have.
Thus, to say that fictional characters are the creations of the imagination
is to say that a statement about a fictional character can be true only if it is
imagined as being true, that a fictional character can have only those
properties which it is part of the story that he has; to say that something is
an object of sense-that for it esse est percipi-is to say that it has only
those properties it is perceived as having: in both cases, the ontological
thesis is a ground for rejecting the law of excluding middle as applied to
22 MICHAEL DUMMETT

statements about those objects. Thus we cannot separate the question of

t h e ontological status of a class of objects from the question of the correct
notion of truth for statements about those objects, i.e. of the kind of thing
in virtue of which such statements are true, when they are true. This
conclusion corroborates the idea that an answer to the former question
cannot serve as a premiss for an answer to the latter one.
Nevertheless, the position is not so straightforward as all this would
make it appear. From the possibility of an argument of the first type for
the use of intuitionistic logic in mathematics, it is evident that a model of
the meanings of mathematical statements in terms of proof rather than of
truth need not rest upon any particular view about the ontological
character o f mathematical objects. There is no substantial disagreement
between the two models of meaning so long as we are dealing only with
decidable statements: the crucial divergence occurs when we consider
ones which are not effectively decidable, and the linguistic operation
which first enables us t o frame effectively undecidable mathematical
statements is that of quantification over infinite totalities, in the first place
over the totality of natural numbers. Now suppose someone who has, on
whatever grounds, been convinced by the Platonist claim that we do not
create the natural numbers, and yet that reference to natural numbers is
not a mere f a t ~ ) t i de purler, but is a genuine instance of reference to
objects: he believes. with tne Platonist, that natural numbers are abstract
objects. existing timelessly and independently of our knowledge of them.
Such a person may, nevertheless, when he comes to consider the meaning
of existential and universal quantification over the natural numbers, be
convinced by ;I line of reasoning such as that which I sketched as
constituting t h e first type of justification for replacing classical by
intuitionistic logic. He may come to the conclusion that quantification
over a denumerable totality cannot be construed in terms of our grasp of
t h e conditions under which a quantified statement is true, but must,
rather. be understood in terms of our ability to recognise a proof or
disproof of such a stztement. He will therefore reject a classical logic for
number-theoretic statements in general, admitting only intuitionistically
valid arguments involving them. Such a person would be accepting a
platonistic view of the existence of mathematical objects (at least the
objects o f number theory), but rejecting a platonistic view of the
object i v it y of mat he m at ical statements.
Our question is, rather. whether the opposite combination of views is
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC 1.0GIC 23

possible: whether one may consistently hold that natural numbers are the
creations of human thought, but yet believe that there is a notion of truth
under which each number-theoretic statement is determinately either
true or false, and that it is in terms of our grasp of their truth-conditions
that our understanding of number-theoretic statements is to be explained.
If such a combination is possible, then, it appears, there can be n o route
from the ontological thesis that mathematical objects are the creations of
our thought to the model of the meanings of mathematical statements
which underlies the adoption of an intuitionistic logic.
This is not the only question before us: for, even if these two views
cannot be consistently combined, it would not follow that the ontological
thesis could serve as a premiss for the constructivist view of the meanings
of mathematical statements; our difficulty was to understand how the
ontological thesis could have any substance if it were not merely a picture
encapsulating that conception of meaning. The answer is surely this: that,
while it is surely correct that a thesis about the ontological status of
objects of a given kind, e.g. natural numbers, must be understood as a
thesis about that in which the truth of certain statements about those
objects consists, it need not be taken as, in the first place, a thesis about
the entire class of such statements; it may, instead, be understood as a
thesis only about some restricted subclass of such statements, those
which are basic to the very possibility of making reference to those
objects. Thus, for example, the thesis that natural numbers are creations
of human thought may be taken as a thesis about the sort of thing which
makes a numerical equation or inequality true, or, more generally, a
statement formed from such equations by the sentential operators and
bounded quantification. To say that the only notion of truth we can have
for number-theoretic statements generally is that which equates truth
with our capacity to prove a statement is to prejudge the issue about the
correct model of meaning for such statements, and therefore cannot serve
as a premiss for the constructivist view of meaning. But to say that, for
decidable number-theoretic statements, truth consists in provability, is
not in itself to prejudge the question in what the truth of undecidable
statements, involving unbounded quantification, consists: and hence the
possibility is open that a view about the one might serve as a premiss for a
view about the other. Our problem is to discover whether it can do so in
fact: whether there is any legitimate route from the thesis that natural
numbers are creations of human thought, construed as a thesis about the
24 MICHAEL DUMMETT

sort of thing which makes decidable number-theoretic statements true, to

a view of the meanings of number-theoretic statements generally which
would require the adoption for them of an intuitionistic rather than a
classical logic.
In order to resolve this question, it is necessary for us to take a rather
closer look at the notion of truth for mathematical statements, as
understood intuitionistically. The most obvious suggestion that comes to
mind in this connection is that the intuitionistic notion of truth conforms,
just as does the classical notion, to Tarski’s schema:
S is true iff A,
where an instance of the schema is t o be formed by replacing ‘ A ’ by some
number-theoretic statement and ‘S’by a canonical name of that sentence,
as, e.g., in:
‘There are infinitely many twin primes’
is true iff there are infinitely many twin primes.
It is necessary to admit counter-examples to the schema (T) in any case in
which we wish to hold that there exist sentences which are neither true
nor false: for if we replace ‘A ’ by such a sentence, the left-hand side of
the biconditional becomes false (on the assumption that, if the negation of
a sentence is true, that sentence is false), although, by hypothesis, the
right-hand side is not false. But, in intuitionistic logic, that semantic
principle holds good which stands to the double negation of the law of
excluded middle as the law of bivalence stands to the law of excluded
middle itself: it is inconsistent to assert of any statement that it is neither
true nor false; and hence there seems no obstacle to admitting the
correctness of the schema (T). Of course, in doing so, we must construe
the statement which appears on the right-hand side of any instance of the
schema in an intuitionistic manner. Provided we do this, a truth-definition
for the sentences of an intuitionistic language, say that of Heyting
arithmetic, may be constructed precisely on Tarski’s lines, and will yield,
as a consequence, each instance of the schema (T).
However, notoriously, such an approach leaves many philosophical
problems unresolved. The truth-definition tells us, for example, that
‘598017 + 246532 = 844549’ is true
just in the case in which 598017 + 246532 = 844549. We may perform the
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 25

computation, and discover that 598017 + 246532 does indeed equal

844549: but does that mean that the equation was already true before the
computation was performed, or that it would have been true even if the
computation had never been performed? The truth-definition leaves such
questions quite unanswered, because it does not provide for inflections of
tense or mood of the predicate ‘is true’: it has been introduced only as a
predicate as devoid of tense as are all ordinary mathematical predicates;
but its role in our language does not reveal why such inflections of tense
or even of mood should be forbidden.
These difficulties raise their heads as soon as we make the attempt to
introduce tense into mathematics, as intuitionism provides us with some
inclination to do; this can be seen from the problems surrounding the
theory of the creative subject. These problems are well brought out in
Troelstra’s discussion of the topic. It is evident that we ought to admit as
an axiom
(a) (k,,A) -+ A;
if we know that, at any stage, A has been (or will be) proved, then we are
certainly entitled to assert A. But ought we to admit the converse in the
form

(P 1 A ---f 3 n @,A) ?
Its double negation
(Y) A + 1 1 3 n (t,A)

is certainly acceptable: if we know that A is true, then we shall certainly

never be able to assert, at least on purely mathematical grounds, that it
will never be proved. But can we equate truth with the obtaining of a
proof at some stage, in the past or in the future, as the equivalence:
(6) A c-, 3 n (k,A)
requires us to do? (To speak of ‘truth’ here seems legitimate, since, while
Tarski’s truth-predicate is a predicate of sentences, the sentential
operator to which it corresponds is a redundant one, which can be
inserted before or deleted from in front of any clause without change of
truth-value.)
If we accept the axiom ( p ) , and hence the equivalence (a), we run
into certain difficulties, on which Troelstra comments. The operator
26 MICHAEL DUMMETT

‘ 3 n (t,l...)’ becomes a redundant truth-operator, and hence may be

distributed across any logical constant, as in

(E 1 (Fk VirZ A ( W t ) ) + Vm EIn (FnA(m)).

A s Troelstra observes, this appears to have the consequence that, if we
have once proved a universally quantified statement, we are in some way
committed to producing, at some time in the future, individual proofs of
all its instances, whereas, palpably, we are under no such constraint. The
solution to which he inclines is that proposed by Kreisel, namely that the
operator ‘t, ’ must be so construed that a proof, at stage n, of a universally
quantified statement counts as being, at the same time, a proof of each
instance, so that we could assert the stronger thesis

(i) ( b k t/m A(m)) 4t l m (kkA(m)).

(Troelstra in fact recommends this interpretation on separate grounds, as
enabling u s to escape a paradox about constructive functions; he himself
points out, however, that this paradox can alternatively be avoided by
introducing distinctions of level which seem intrinsically plausible.) The
difficulty about this solution is that it must be extended to every
recognised logical consequence. From

(rl) (ni 5 n & ( F A ) ) -+ (FA)

we have

(0) (ti = max ( m , k ) & @,,,A) & (FkC)) -+ ( ( F A ) & (tnC)>,

while from ( 6 ) we obtain

( 1 ) ( t , , , A ) & (kc(A -+ B)) - 3 n (F,,B).

We could in the same way complain that this committed us, whenever we
had proved a statement A and had recognised some other statement B as
being a consequence of A, to actually drawing that consequence some
time in the future: and, if our interpretation of the operator ‘ t n ’is to be
capable of dealing with this difficultyin the same way as with the special
case of instances of a universally quantified statement, we should have to
allow that a proof that a theorem had a certain consequence was, at the
same time, a proof of that consequence, and, likewise, that a proof of a
statement already known to have a certain consequence was, at the same
time, a proof of that consequence; we should, that is, have to accept the
 T H E PHILOSOPHICAL B A S I S OF INTUITIONISTIC LOGIC 27

law
( K) (n = max (m,k ) & ( k m A )& ( k k ( A-+ B ) ) )+ (k,,B).
We should thus so have to construe the notion of proof that a proof of a
statement is taken as simultaneously constituting a proof of anything that
has already been recognised as a consequence of that statement. We can,
no doubt, escape having to say that it is simultaneously a proof of
whatever, in a platonistic sense, as a matter of fact is an intuitionistic
consequence of the statement: but when are we to be said to have
recognised that one statement is a consequence of another? If a proof of a
universally quantified statement is simultaneously a proof of all its
instances, it is difficult to see how we can avoid conceding that a
demonstration of the validity of a schema of first-order predicate logic is
simultaneously a demonstration of the truth of all its instances, or an
acceptance of the induction schema simultaneously an acceptance of all
cases of induction. The resulting notion of proof would be far removed
indeed from actual mathematical experience, and could not be explained
as no more than an idealisation of it.
The trouble with all this is that, as a representation of actual mathemati-
cal experience, we are operating with too simplified a notion of proof. The
axiom (7)is acceptable in the sense that, prescinding from the occasional
accident, once a theorem has been proved, it always remains uuuilahle to
be subsequently appealed to: but the idea that, having acknowledged the
two premisses of a modus ponens, we have thereby recognised the truth
of the conclusion, is plausible only in a case in which we are simultane-
ously bearing in mind the truth of the two premisses. To have once proved
a statement is not thereafter t o be continuously aware of its truth: if it
were, then we should indeed always know the logical consequences of
everything which we know, and should have no need of proof.
Acceptance of axiom (p)leads to the conclusion that we shall eventually
prove every logical consequence of everything we prove. This, as a
representation of the intuitionist notion of proof, is an improvement upon
Beth trees, as normally presented: for these are set up in such a way that, at
any stage (node), every logical consequence of statements true at that stage
is already true; the Beth trees are adapted only to situations, such as those
involving free choice sequences, where new information is coming in that
is not derived from the information we have at earlier stages. But the idea
that we shall eventually establish every logical consequence of everything
28 MICHAEL DUMMETT

we know is implausible and arbitrary: and it cannot be rescued by

construing each proof as, implicitly, a proof also of the consequences of the
statement proved, save at the cost of perverting the whole conception. If
we wish to do so, there seems no reason why we should not take the stages
represented by the numerical subscripts as punctuated by proofs, however
short the stages thereby become, and the notion of proof as relating only to
what is quite explicitly proved, so that, at each stage, one and only one new
statement is proved, and consider what axioms hold under the resulting
interpretation of the symbol ‘k,,’. It thus appears that, under this interpreta-
tion. the axiom ( p ) must be rejected in favour of the weaker axiom ( 7 ) .
Looked at another way, however, the stronger axiom ( p ) seems
entirely acceptable. If, that is, we interpret the implication sign in its
intuitionistic sense, the axiom merely says that, given a proof of A, we
can effectively find a proof that A was proved at some stage; and this
seems totally innocuous and banal. But, if axiom ( p ) is innocuous, how
did we arrive at our earlier difficulties? The only possibility seems to be
that our logical laws are themselves at fault. For instance, the law
(A 1 V x A ( x ) -+ A ( m )

leads, via axiom ( B ) , to the conclusion

(P) Vx A (x 1 + 3 n ( F A ( m I),
which appears, on the present interpretation of ‘I,,’, to say that we shall
explicitly prove every instance of every universally quantified statement
which we prove; so perhaps the error lies in the law ( A ) itself. A law such
as ( A ) is ordinarily justified by saying that, given a proof of Vx A(x), we
can, for each m, effectively find a proof of A ( m ) . If this is to remain a
sufficient justification of ( p ) , then ( p , ) must be construed as saying that,
given a proof of Vx A ( x ) , we can effectively find a proof that A ( m ) will
be proved at some stage. How can we do this, for given m ? Obviously, by
proving A ( m ) and noting the stage at which we do so. This means, then,
that the existentially quantified statement
(v) 3 n (k,,A(m))
is to be so understood that its assertion does not amount to a claim that we
shall, as a matter of fact, prove A ( m ) at some stage n, but only that we are
capable of bringing it about that A ( m ) is proved at some stage. Our
difficulties thus appear to have arisen from understanding the existential
quantifier in ( p ) in an excessively classical or realistic manner, namely as
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 29

meaning that there will in fact be a stage n at which the statement is

proved rather than as meaning that we have an effective means, if we
choose to apply it, of making it the case that there is such a stage. The
point here is that it is not merely a question of interpreting the existential
quantifier intuitionistically rather than classically in the sense that we can
assert that there is a stage n at which a statement will be proved only if we
have an effective means for identifying a particular such stage. Rather, if
quantification over temporal stages is to be introduced into mathematical
statements, then it must be treated like quantification over mathematical
objects and mathematical constructions: the assertion that there is a stage
n at which such-and-such will hold is justified provided that we possess
an enduring capability of bringing about such a stage, regardless of
whether we ever exercise this capability or not.
The confusions concerning the theory of the creative subject which we
have been engaged in disentangling arose in part from a perfectly
legitimate desire, to relate the intuitionistic truth of a mathematical
statement with a use of the logical constants which is alien to intuitionistic
mathematics. Troelstra’s difficulties sprang from his desire to construe the
expression ‘ 3 n (t,A)’ as meaning that A would in fact be proved at some
stage: but, whether we interpret the existential quantifier classically or
constructively, such a construal of it fails to gibe with the way it and the
other logical constants are construed within ordinary mathematical state-
ments, and hence, however we try to modify our notion of a statement’s
being proved, we shall not obtain anything equivalent to the mathematical
statement A itself. Nevertheless, the desire to express the condition for
the intuitionistic truth of a mathematical statement in terms which do not
presuppose an understanding of the intuitionistic logical constants as used
within mathematical statements is entirely licit. Indeed, if it were impossi-
ble to d o so, intuitionists would have no way of conveying to Platonist
mathematicians what it was that they were about: we should have a
situation quite different from that which in fact obtains, namely one in
which some people found it natural to extend basic computational
mathematics in a classical direction, and others found it natural to extend
it in an intuitionistic direction, and neither could gain a glimmering of
what the other was at. That we are not in this situation is because
intuitionists and platonists can find a common ground, namely statements,
both mathematical and non-mathematical, which are, in the view of both,
decidable, and about whose meaning there is therefore no serious dispute
and which both sides agree obey a classical logic. Each party can,
30 MICHAEL DUMMETT

accordingly, by use of and reference to these unproblematic statements,

explain to the other what his conception of meaning is for those
mathematical statements which are in dispute. Such an explanation may
not be accepted as legitimate by the other side (the whole point of the
intuitionist position is that undecidable mathematical statements cannot
legitimately be given a meaning by laying down truth-conditions for them
in the platonistic manner): but at least the conception of meaning held by
each party is not wholly opaque to the other.
This dispute between platonists and intuitionists is a dispute over
whether or not a realist interpretation is legitimate for mathematical
statements: and the situation I have just indicated is quite characteristic
for disputes concerning the legitimacy of a realist interpretation of some
class of statements, and is what allows a dispute to take place at all.
Typically, in such a dispute there is some auxiliary class of statements
about which both sides agree that a realist interpretation is possible
(depending upon the grounds offered by the anti-realists for rejecting a
realist interpretation for statements of the disputed class, this auxiliary
class may or may not consist of statements agreed to be effectively
decidable); and typically, it is in terms of the truth-conditions of state-
ments of this auxiliary class that the anti-realist frames his conception of
meaning, his non-classical notion of truth, for statements of the disputed
class, while the realist very often appeals to statements of the auxiliary
class as providing an analogy for his conception of meaning for state-
ments of the disputed class. Thus, when the dispute concerns statements
about the future. statements about the present will form the auxiliary
class; when it concerns statements about material objects, the auxiliary
class will consist of sense-data statements; when the dispute concerns
statements about character-traits, the auxiliary class will consist of
statements about actual or hypothetical behaviour; and so on.
If the intuitionistic notion of truth for mathematical statements can be
explained only by a Tarski-type truth-definition which takes for granted
the meanings of the intuitionistic logical constants, then the intuitionist
notion of truth, and hence of meaning, cannot be so much as conveyed t o
anyone who does not accept it already, and no debate between intuition-
ists and platonists is possible, because they cannot communicate with one
another. It is therefore wholly legitimate, and, indeed, essential, to frame
the condition for the intuitionistic truth of a mathematical statement in
 THE PHILOSOPHICAL BASIS OF; INTUITIONISTIC mwc 31

terms which are intelligible to a Platonist and do not beg any questions,
because they employ only notions which are not in dispute.
The obvious way to do this is to say that a mathematical statement is
intuitionistically true if there exists an (intuitionistic) proof of it, where
the existence of a proof does not consist in its platonic existence in a
realm outside space and time, but in our actual possession of it. Such a
notion of truth, obvious as it is, already departs at once from that supplied
by the analogue of the Tarski-type truth-definition, since the predicate ‘is
true’, thus explained, is significantly tensed: a statement not now true may
later become true. For this reason, when ‘true’ is so construed, the
schema (T) is incorrect: for the negation of the right-hand side of any
instance will be a mathematical statement, while the negation of the
left-hand side will be a non-mathematical statement, to the effect that we
do not as yet possess a proof of a certain mathematical statement, and
hence the two sides cannot be equivalent. We might, indeed, seek to
restore the equivalence by replacing ‘is true’ on the left-hand side by ‘is or
will be true’: but this would lead us back into the difficulties we
encountered with the theory of the creative subject, and I shall not further
explore it.
What does require exploration is the notion of proof being appealed to,
and that also of the existence of a proof. It has often, and, I think,
correctly been held that the notion of proof needs to be specialised if it is
to supply a non-circular account of the meanings of the intuitionistic
logical constants. It is possible to see this by considering disjunction and
existential quantification. The standard explanation of disjunction is that
a construction is a proof of A v B just in case it is a proof either of A or
of B. Despite this, it is not normally considered legitimate to assert a
disjunction, say in the course of a proof, only when we actually have a
proof of one or other disjunct. For instance, it would be quite in order to
assert that
1 is either prime or composite
without being able to say which alternative held good, and to derive some
theorem by means of an argument by cases. What makes this legitimate,
on the standard intuitionist view, is that we have a method which is in
principle effective for deciding which of the two alternatives is correct: if
we were to take the trouble to apply this method, the appeal to an
32 MICHAEI- DUMMETI

argument by cases could be dispensed with. Generally speaking, there-

fore, if we take a statement as being true only when we actually possess a
proof of it, an assertion of a disjunctive statement will not amount to a
claim that it is true, but only to a claim that we have a means, effective in
principle, for obtaining a proof of it. This means, however, that we have
to distinguish between a proof proper, a proof in the sense of ‘proof’ used
in the explanations of the logical constants, and a cogent argument. In the
course of a cogent argument for the assertibility of a mathematical
statement, a disjunction of which we do not possess an actual proof may
be asserted, and an argument by cases based upon this disjunction. This
argument will not itself be a proof, since any initial segment of a proof
must again be a proof: it merely indicates an effective method by which
we might obtain a proof of the theorem if we cared to apply it. We thus
appear to require a distinction between a proof proper-a canonical
proof-and the sort of argument which will normally appear in a
mathematical article or textbook, an argument which we may call a
‘demonstration’. A demonstration is just as cogent a ground for the
assertion of its conclusion as is a canonical proof, and is related to it in
this way: that a demonstration of a proposition provides an effective
means for finding a canonical proof. But it is in terms of the notion of a
canonical proof that the meanings of the logical constants are given.
Exactly similar remarks apply to the existential quantifier.
There is some awkwardness about this way of looking at disjunction
and existential quantification, namely in the divorce between the notions
of truth and of assertibility. It might be replied that the significance of the
act of assertion is not, in general, uniquely determined by the notion of
truth: for instance, even when we take the notion of truth for mathemati-
cal statements as given, it still needs to be stipulated whether the assertion
of a mathematical statement amounts to a claim to have a proof of it, or
whether it may legitimately be based on what Polya calls a ‘plausible
argument’ of a non-apodeictic kind. (We can imagine people whose
mathematics wholly resembles ours, save that they do not construe an
assertion as embodying a claim to have more than a plausible argument.)
It nevertheless remains that, if the truth of a mathematical statement
consists in our possession of a canonical proof of it, while its assertion
need be based on possession of no more than a demonstration, we are
forced to embrace the awkward conclusion that it may be legitimate to
assert a statement even though it is known not to be true. However, if the
 THE PHILOSOPHICAI< BASIS OF I N T U I T I O N I S T I C LOGIC 33

sign of disjunction and the existential quantifier were the only logical
constants whose explanation appeared to call for a distinction between
canonical proofs and demonstrations, the distinction might be avoided
altogether by modifying their explanations, to allow that a proof of a
disjunction consisted in any construction of which we could recognise
that it would effectively yield a proof of one or other disjunct, and
similarly for existential quantification: we should then be able to say that
a statement could be asserted only when it was (known to be) true.
However, the distinction is unavoidable if the explanations of universal
quantification, implication and negation are to escape circularity. The
standard explanation of implication is that a proof of A -+ B is a
construction of which we can recognise that, applied to any proof of A, it
would yield a proof of B. It is plain that the notion of proof being used
here cannot be one which admits unrestricted use of modus ponens: for, if
it did, the explanation would be quite empty. We could admit anything we
liked as constituting a proof of A -+ B, and it would remain the case that,
given such a proof, we had an effective method of converting any proof of
A into a proof of B, namely by adding the proof of A -+ B and
performing a single inference by modus ponens. Obviously, this is not
what is intended: what is intended is that the proof of A -+ B should
supply a means of converting a proof of A into a proof of B without
appeal to modus ponens, at least, without appeal to any modus ponens
containing A + B as a premiss. The kind of proof in terms of which the
explanation of implication is being given is, therefore, one of a restricted
kind. On the assumption that we have, or can effectively obtain, a proof of
A + B of this restricted kind, an inference from A -+ B by modus
ponens is justified, because it is in principle unnecessary. The same must,
by parity of reasoning, hold good for any other application of modus
ponens in the main (though not in any subordinate) deduction of any
proof. Thus, if the intuitionistic explanation of implication is to escape,
not merely circularity, but total vacuousness, there must be a restricted
type of proof-canonical proof-in terms of which the explanation is
given, and which does not admit modus ponens save in subordinate
deductions. Arguments employing modus ponens will be perfectly valid
and compelling, but they will, again, not be proofs in this restricted sense:
they will be demonstrations, related to canonical proofs as supplying a
means effective in principle for finding canonical proofs. Exactly simi!ar
remarks apply to universal quantification vis-8-vis universal instantiation
34 MICIiAHL DUMMETT

and to negation vis-a-vis the rule ex falso quodlibet: the explanations of

these operators presuppose a restricted type of proof in which
the corresponding elimination rules do not occur within the main deduc-
tion.
What exactly the notion of a canonical proof amounts to is obscure.
The deletion of elimination rules from a canonical proof suggests a
comparison with the notion of a normalised deduction. On the other hand,
Brouwer’s celebrated remarks about fully analysed proofs in connection
with the bar theorem do not suggest that such a proof is one from which
unnecessary detours have been cut out-the proof of the bar theorem
consists in great part in cutting out such detours from a proof taken
already to be i n ‘fully analysed’ form. Rather, Brouwer’s idea appears to
be that, in a fully analysed proof, all operations on which the proof
depends will actually have been carried out. That is why such a proof may
be an infinite structure: a proof of a universally quantified statement will
be an operation which, applied to each natural number, will yield a proof
of the corresponding instance; and, if this operation is carried out for each
natural number, we shall have proofs of denumerably many statements.
The conception of the mental construction which is the fully analysed
proof as being an infinite structure must, of course, be interpreted in the
light of the intuitionist view that all infinity is potential infinity: the mental
construction consists of a grasp of general principles according to which
a n y finite segment of the proof could be explicitly constructed. The
direction of analysis runs counter to the direction of deduction; while one
could not be convinced by an actually infinite proof-structure (because
one would never reach the conclusion), one may be convinced by a
potentially infinite one, because its infinity consists in our grasp of the
principles governing its analysis. Indeed, it might reasonably be said that
the standard intuitionistic meanings of the universal and conditional
quantifiers involve that a proof is such a potentially infinite structure.
Nevertheless, the notion of a fully analysed proof, that is, of the result of
applying every operation involved in the proof, is far from clear, because
it is obscure what the effect of the analysis would be on conditionals and
negative statements. We can systematically display the results of applying
the operation which constitutes a proof of a statement involving universal
quantification over the natural numbers, because we can generate each
natural number in sequence. But the corresponding application of the
operation which constitutes the proof of a statement of the form A + B
 THE PHILOSOPHICAL BASIS OF INTUITION!STIC LOGIC 35

would consist in running through all putative canonical proofs of A and

either showing, in each case, that it was not a proof of A, or transforming
it into a proof of B : and, at least without a firm grasp upon the notion of a
canonical proof, we have no idea how to generate all the possible
candidates for being a proof of A.
The notion of canonical proof thus lies in some obscurity; and this state
of affairs is not indefinitely tolerable, because, unless it is possible to find
a coherent and relatively sharp explanation of the notion, the viability of
the intuitionist explanations oi the logical constants must remain in doubt.
But, for present purposes, it does not matter just how the notion of
canonical proof is to be explained; all that matters is that we require some
distinction between canonical proofs and demonstrations, related to one
another in the way that has been stated. Granted that such a distinction is
necessary, there is no motivation for refusing to apply it to the case of
disjunctions and existential statements.
Let us now ask whether we want the intuitionistic truth of a mathemati-
cal statement to consist in the existence of a canonical proof or of a
demonstration. If by the ‘existence’ of a proof or demonstration we mean
that we have actually explicitly carried one out, then either choice leaves
us with certain counter-intuitive consequences. On either view, naturally,
a valid rule of inference will not always lead from true premises to a true
conclusion, namely if we have not explicitly drawn the inference: this will
always be so on any view which equates truth with our actual possession
of some kind of proof. If we take the stricter line, and hold a statement to
be true only when we possess a canonical proof of it, then, as we have
seen, we shall have to allow that a statement may be asserted even though
it is known not to be true. If, on the other hand, we allow that a statement
is true when we possess merely a demonstration of it, then truth will not
distribute over disjunction: we may possess a demonstration of A v B
without having a demonstration either of A or of B . Now, admittedly,
once we have admitted a significant tense for the predicate ‘is true’, then,
as we have noted, the schema (T) cannot be maintained as in all cases
correct: but our instinct is to permit as little divergence from it as
possible, and it is for this reason that we are uneasy about a notion of
truth which is not distributive over disjunction or existential quantifica-
tion.
A natural emendation is to relax slightly the requirement that a proof or
demonstration should have been explicitly given. The question is how far
36 MICHAEL DUMMETT

we may consistently go along this path. If we say merely that a

mathematical statement is true just in case we are aware that we have an
effective means of obtaining a canonical proof of it, this will not be
significantly different from equating truth with our actual possession of a
demonstration. It might be allowed that there would be some cases when
we had demonstrated the p r e m i s e s of, say, an inference by modus
ponens in which we were aware that we could draw the conclusion,
though we had not quite explicitly done so; but there will naturally be
others in which we were not aware of this, i.e. had not noticed it; if it were
not so, we could never discover new demonstrations. It is therefore
tempting to go one step further, and say that a statement is true provided
that we are in fact in possession of a means of obtaining a canonical proof
of it, whether or not we are aware of the fact. Would such a step be a
betrayal of intuitionist principles?
In which cases would it be correct to say that we possess an effective
means of finding a canonical proof of a statement, although we do not
know that we have such a means? Unless we are to suppose that we can
attain so sharp a notion of a canonical proof that it would be possible to
enumerate effectively all putative such proofs of a given statement (the
supposition whose implausibility causes our difficulty over the notion of a
fully analysed proof), there is only one such case: that in which we
possess a demonstration of a disjunctive or existential statement. Such a
demonstration provides us with what we recognise as an effective means
(in principle) for finding a canonical proof of the disjunctive or existential
statement demonstrated. Such a canonical proof, when found, will be a
proof of one or other disjunct, or of one instance of the existentially
quantified statement: but we cannot, in general, tell which. For example,
when A ( x ) is a decidable predicate, the decision procedure constitutes a
demonstration of the disjunction ‘ A ( i i ) v l A ( i i ) ’ ,for specific n ; but,
until we apply the procedure, we d o not know which of the two disjuncts
we can prove. It is very difficult for us to resist the temptation to suppose
that there is already, unknown to us, a determinate answer to the question
which of the two disjuncts we should obtain a proof of, were we to apply
the decision procedure; that, for example, that it is already the case either
that, if we were to test it out, we should find that + 1 is prime, or that,
if we were to test it out, we should find that it was composite. What is
involved here is the passage from a subjunctive conditional of the form:
A +(B v C)
 THE PHILOSOPHICAL BASIS OF INTUITIONISTIC LOGIC 37

to a disjunction of subjunctive conditionals of the form

(A -+ B) v (A -+ C).
Where the conditional is interpreted intuitionistically, this transition is, of
course, invalid: but the subjunctive conditional of natural language does
not coincide with the conditional of intuitionistic mathematics. It is,
indeed, the case that the transition is not in general valid for the
subjunctive conditional of natural language either: but, when we reflect
on the cases in which the inference fails, it is difficult to avoid thinking
that the present case is not one of them.
There are two obvious kinds of counter-example to this form of
inference for ordinary subjunctive conditionals: perhaps they are really
two sub-varieties of a single type. One is the case in which the antecedent
A requires supplementation before it will yield a determinate one of the
disjuncts B and C. For instance, we may safely agree that, if Fidel Castro
were to meet President Nixon, he would either insult him or speak
politely to him; but it might not be determinately true, of either of those
things, that he would do it, since it might depend upon some so far
unspecified further condition, such as whether the meeting took place in
Cuba or outside. Schematically, this kind of case is one in which we can
assert:
A (B v C ) ,
-+

(A & Q > B,
-+

(A 1Q) + C,
but in which the subjunctive antecedent A neither implies nor presup-
poses either Q or its negation; in such a case, we cannot assert either
A + B or A -+ C. The other kind of counter-example is that in which we
do not consider the disjuncts to be determined by anything at all: no
supplementation of the antecedent would be sufficient to decide between
them in advance. If an electron were to pass through that screen, it would
have passed through one aperture or the other: but there is nothing at all
which will determine in advance through which it would pass. Similar
cases will arise, for those who believe in free will in the traditional sense,
in respect of human actions.
If we were to carry out the decision procedure for determining the
primality or otherwise of some specific large number N,we should either
obtain the result that N is prime or obtain the result that N is composite.
38 M I c ' H A E L D U MMET7

Is this, or is it not, a case in which we may conclude that it either holds

good that, if we were to carry out the procedure, we should find that N is
prime, or that. if we were to carry out the procedure, we should find that
N is composite? The difficulty of resisting the conclusion that it is such a
case stems from the fact that it does not display either of the characteris-
tics found in the two readily admitted types of counter-example to the
form of inference we are considering. No further circumstance could be
relevant to the result of the procedure-this is part of what is meant by
calling it a computation; and, since at each step the outcome of the
procedure is determined, how can we deny that the overall outcome is
determinate also?
If we yield to this line of thought, then we must hold that every
statement formed by applying a decidable predicate to a specific natural
number already has a definite truth-value, true or false, although we may
not know it. And, if we hold this, it makes no difference whether we chose
at the outset to say that natural numbers are creations of the human mind
or that they are eternally existing abstract objects. Whichever we say, our
decision how to interpret undecidable statements of number theory, and,
in the first place, statements of the forms V x A (x) and 3 x A (x), where
A ( x ) is decidable, will be independent of our view about the ontological
status of natural numbers. For on this view of the truth of mathematical
statements, each decidable number-theoretic statement will already be
deterrninately true or false, independently of our knowledge, just as it is
on a platonistic view: any thesis about the ontological character of natural
numbers will then be quite irrelevant to the interpretation of the quantifi-
ers. As we noted, it would be possible for someone to be prepared to
regard natural numbers as timeless abstract objects, and to regard
decidable predicates as being determinately true or false of them, and yet
to be convinced by an argument of the first type, based on quite general
considerations concerning meaning, that unbounded quantification over
natural numbers was not an operation which in all cases preserved the
property of possessing a determinate truth-value, and therefore to fall
back upon a constructivist interpretation of it. Conversely, if someone
who thought of the natural numbers as creations of human thought also
believed, for the reasons just indicated, that each decidable predicate was
determinately true or false of each of them, he might accept a classical
interpretation of the quantifiers. H e would do so if he was unconvinced by
the general considerations about meaning which we reviewed, i.e., by the
 T H E PHl1.OSOPHICAL BASIS OF INTUITI0NI';TIC I O(ilC 39

first type of argument for the adoption of an intuitionistic logic for

mathematics: the fact that he was prepared to concede that the natural
numbers come into existence only in virtue of our thinking about them
would play no part in his reflections on the meanings of the quantifiers.
Dedekind, who declared that mathematical structures are free Creations
of the human mind, but nevertheless appears to have construed state-
ments about them in a wholly platonistic manner, may perhaps be an
instance of just such a combination of ideas.
One who rejects the idea that there is already a determinate outcome
for the application, to any specific case, of an effective procedure is,
however, in a completely different position. If someone holds that the
only acceptable sense in which a mathematical statement, even one that is
effectively decidable, can be said to be true is that in which this means
that we presently possess an actual proof or demonstration of it, then a
classical interpretation of unbounded quantification over the natural
numbers is simply unavailable to him. As is frequently remarked, the
classical or platonistic conception is that such quantification represents an
infinite conjunction or disjunction: the truth-value of the quantified
statement is determined as the infinite sum or product of the truth-values
of the denumerably many instances. Whether or not this be regarded as an
acceptable means of determining the meaning of these operators, the
explanation presupposes that all the instances of the quantified statement
themselves already possess determinate truth-values: if they do not, it is
impossible to take the infinite sum or product of these. But if, for
example, we do not hold that such a predicate as 'x is odd 4 x is not
perfect' already has a determinate application to each natural number,
though we do not know it, then it is just not open to him to think that, by
attaching a quantifier to this predicate, we obtain a statement that is
determinately true or false.
One question which we asked earlier was this: Can the thesis that
natural numbers are creations of human thought be taken as a premiss for
the adoption of an intuitionistic logic for number-theoretic statements?
And another question was: What content can be given to the thesis that
natural numbers are creations of human thought that does not prejudge
the question what is the correct notion of truth for number-theoretic
statements in general? The tentative answer which we gave to this latter
question was that the thesis might be taken as relating to the appropriate
notion of truth for a restricted class of number-theoretic statements, say
numerical equations, or, more generally, decidable statements. From
what we have said about the intuitionistic notion of truth for mathemati-
cal statements, it has now become apparent that there is one way in which
the thesis that natural numbers are creations of the human mind might be
taken, namely as relating precisely to the appropriate notion of truth for
decidable statements of arithmetic, which would provide a ground for
rejecting a platonistic interpretation of number-theoretic statements
generally, without appeal to any general thesis concerning the notion of
meaning. This way of taking the thesis would amount to holding that there
is no notion of truth applicable even to numerical equations save that in
which a statement is true when we have actually performed a computa-
tion (or effected a proof) which justifies that statement. Such a claim must
rest, as we have seen, on the most resolute scepticism concerning
subjunctive conditionals: it must deny that there exists any proposition
which is now true about what the result of a computation which has not
yet been performed would be if it were to be performed. Anyone who can
hang on to a view as hard-headed as this has no temptation at all to accept
a platonistic view of number-theoretic statements involving unbounded
quantification: he has a rationale for an intuitionistic interpretation of
them which rests upon considerations relating solely to mathematics, and
demanding no extension to other realms of discourse (save in so far as the
subjunctive conditional is involved in explanations of the meanings of
statements in these other realms). But, for anyone who is not prepared to
be quite as hard-headed as that, the route to a defence of an intuitionistic
interpretation of mathematical statements which begins from the ontolog-
ical status of mathematical objects is closed: the only path that he can
take to this goal is that which I sketched at the outset: one turning on the
answers given to general questions in the theory of meaning.