Truth and Proof, The Platonism of Mathematics - William W. Tait

W. W.
TAIT
TRUTH AND PROOF: THE PLATONISM OF

MATHEMATICS*
W h a t is the relation in m a t h e m a t i c s between truth and proof?

An arithmetical proposition A, for example, is about a certain
structure, the system of natural numbers. It refers to numbers and
relations a m o n g them. If it is true, it is so in virtue of a certain fact
about this structure. A n d this fact m a y obtain e v e n if we do not or (for
example, because of its relative complexity) cannot know that it does.
This is a typical expression of what has c o m e to be called the Platonist
(or platonist or realist) point of view towards mathematics.
On the other hand, we learn m a t h e m a t i c s by learning how to do
things - for example, to count, compute, solve equations and, m o r e
generally, to prove. M o r e o v e r , we learn that the ultimate warrant for a
m a t h e m a t i c a l proposition is a proof of it. In other words, we are
justified in asserting A - and therefore, in any ordinary sense, the truth
of A - precisely when we have a proof of it.
Thus, we seem to h a v e two criteria for the truth of A: it is true if
(indeed, if and only if) it holds in the system of numbers, and it is true
if we can p r o v e it. But what has what we h a v e learned or agreed to
count as a proof got to do with what obtains in the system of numbers?
I shall call this the Truth/Proof problem. It underlies m a n y c o n t e m -
porary attacks on Platonism.
T h e a r g u m e n t against Platonism begins with the observation that
the first criterion, holding in the system of numbers, is inapplicable
because we h a v e no direct apprehension of this structure. Sometimes
this a r g u m e n t is a u g m e n t e d by the thesis that 'apprehension of' would
involve causal interaction with the elements of the structure and, since
numbers are 'abstract' (i.e., not in spacetime), no such interaction is
possible. In any case, the a r g u m e n t continues: it follows that a proof
cannot be a warrant - or e v e n incomplete evidence - for holding in
the structure. For no kind of evidence is available that the canons of
proof apply to the structure. Thus, if proof is a warrant for A, then A
cannot be about the system of numbers. If, on the other hand, proof is
Synthese 69 (1986) 341-370.

O 1986 by D. Reidel Publishing Company.
342 w.W. TAIT
not a warrant, then we have no mathematical knowledge at all. An

e v e n stronger argument, to the effect that we cannot even meaning-
fully refer to numbers, is based on the thesis that reference also
involves causal interaction.
Because of these considerations, m a n y writers have felt that
mathematics is in need of a foundation in the revisionist sense that we
must so construe the meaning of mathematical propositions as to
eliminate the apparent reference to mathematical objects and struc-
tures. Some of these writers see Platonism itself as a foundation, i.e., as
a theory of what mathematics is about - but one which, no matter how
naively plausible, is refuted by the Truth/Proof problem.
T h e r e are m a n y interesting problems which might reasonably be
called problems in the foundations of mathematics; but I shall argue
here that a m o n g them is not the need for a foundation in this
revisionist sense. T h e T r u t h / P r o o f problem, which seems to d e m a n d
such a revision, will resolve itself once we are clear about what truth
and proof in mathematics m e a n and what is involved in the notion of a
proposition holding in a structure. These notions seem to me to be
surrounded in the literature by a good deal of confusion which gets
attached to Platonism. Free of this confusion, Platonism will appear,
not as a substantive philosophy or foundation of mathematics, but as a
truism.
Many who reject Platonism on the grounds of the Truth/Proof prob-

lem take discourse about sensible objects to be, not only unprob-
lematic, but a paradigm case of the apparent content of a proposition
being its real content. Thus
(1) T h e r e is a prime n u m b e r greater than 10
is not really about the system of numbers, as we might naively read it,
because our warrant for it is a proof - and what has that to do with the
system of numbers? On the other hand,
(2) T h e r e is a chair in the r o o m
really is about the sensible world - about chairs and rooms - because
we verify it by looking about the r o o m and seeing a chair. Thus,
D u m m e t t (1967) begins
THE PLATONISM OF MATHEMATICS 343
Platonism, as a philosophy of mathematics, is founded on a simile: the comparison

between the apprehension of mathematical truth to [sic] the perception of physical
objects, and thus of mathematical reality to the physical universe.
H e then argues that there is no a n a l o g u e to o b s e r v a t i o n in the case of
m a t h e m a t i c s and so the simile is m i s c o n c e i v e d . A n d B e n a c e r r a f
(1973) writes
One of its [i.e., the 'standard' Platonistic account's] primary advantages is that the truth
definitions for individual mathematical theories thus construed will have the same
recursion clauses as those employed for their less lofty empirical cousins. (p. 669)
T h e ' s t a n d a r d ' a c c o u n t is that, for e x a m p l e (1) has the ' l o g i c o -
grammatical form'
T h e r e is an F w h i c h bears the relation G to b
and he takes it as u n p r o b l e m a t i c that (2) has this form. But

For the typical "standard" account (at least in the case of number theory or set theory)
will depict truth conditions in terms of conditions on objects whose nature, as normally
conceived, places them beyond the reach of the better understood means of human
cognition (e.g., sense perception and the like). (p. 667-8)
I shall h a v e m o r e to say bearing on these passages in the course of m y

p a p e r ; but for n o w I intend t h e m only as instances of the view that,
whereas the naive (Platonistic) construal of (1) is problematic, the
naive reading of (2) is acceptable, indeed as a p a r a d i g m c a s e J
W h y does the e x p e r i e n c e that I describe as "seeing a chair in the

r o o m " w a r r a n t the assertion of (2) any m o r e than a p r o o f warrants the
assertion of (1)? I am n o t referring here to the possibility of p e r c e p t u a l
e r r o r or illusion: the nearest a n a l o g u e to that in the case of (1) would
p e r h a p s be e r r o r in p r o o f or ambiguity of symbols. R a t h e r , I a m asking
a traditional sceptical question: w h a t h a v e m y experiences to do with
physical objects and their relationships at all?
For, in the case of (2) also, I am applying the c a n o n s of verification
that I h a v e b e e n trained to apply. A m o n g o t h e r things, this training
i n v o l v e d learning to say and r e a c t to sentences such as " I see a c h a i r " ,
" T h e r e is no chair in the r o o m " , etc., u n d e r suitable circumstances. It
is true that, unlike the case of (1), these c i r c u m s t a n c e s involve sensory
experience. B u t (2) is a b o u t physical objects, n o t m y sensations.
344 w . w . TAIT
One may feel that the crucial difference between (1) and (2) is this:
in the former case, proving is inextricably bound up with what I have
been trained to do; whereas in the latter case, the role of training is
confined to language learning and this consists simply in learning to
put the right (conventional) names to things. And, after that, training
plays no further role: I simply read the true proposition off the f a c t as
I observe it.
This view of (2) is in essentials the so-called 'Augustinean' view of
language which, in my opinion, is thoroughly undermined by Witt-
genstein's Investigations, §§1-32. My learning to put names to things
consists in my learning to use and respond, verbally and otherwise, to
expressions involving these names. For example, how is it that I am
naming the chair as opposed to naming its shape, color, surface,
undetached chair part, temporal slice, etc.? T h e answer is that it is the
way we use the word "chair" that determines this. And the point is not
merely that the act of naming is ambiguous as to which among several
categories it refers. Ambiguity itself presupposes language: to under-
stand words like "shape", "color", etc., is to have a mastery of a
language. My point - or rather, Wittgenstein's point - is that nothing is
established by the act which we call an act of naming, without a
background of language or, at least, without further training in how
the name is to be used. 2 We do not read the grammatical structure of
propositions about sensible objects off the sensible world nor do we
read true propositions about sensibles off prelinguistic 'facts'. Rather,
we master language and, in language, we apprehend the structure of
the sensible world and facts. T o apprehend the fact that A is simply to
apprehend that A. And this apprehension presupposes language mas-
tery.
So, if A is a proposition about the sensible world of rooms and
chairs, then it is true if and only if it holds in that world. But we
sometimes count what we experience as verification for A. And why
should these two things, what holds in the world of rooms and chairs
and what we experience, have anything to do with each other? Note
that it is not sufficient to point out that verification is not conclusive in
the way that the existence of a proof is, since the question is why
verification should have anything to do with what holds in the world of
rooms and chairs.
Thus, I see nothing special to mathematics about the Truth/Proof
problem. We have described a Truth/Verification problem which is its
analogue in the case of the sensible world. Moreover, the latter is not
really a new argument but, in essentials, has been a standard part of
the sceptics armory. It is perhaps this analogy that G6del has in mind
when he wrote (1964, p. 470) that "the question of the objective
existence of the objects of mathematical i n t u i t i o n . . , is the exact
replica of the question of the objective existence of the outer world".
At any rate, I know of no argument against the existence of mathe-
matical objects which does not have a replica in the case of sensible
objects. For example, some writers argue against Platonism that, if
there is a system of numbers, then why shouldn't there be more than
one of them, all indistinguishable - how would we distinguish them?
And why should our theorems refer to one such system rather than
another? Answer: why shouldn't there be more than one physical
world, all indistinguishable from one another and such that my 'seeing
a chair' is a seeing a chair in all of them? Why should (2) be about one
of these worlds rather than another? If you answer that it is about the
world you inhabit, then I shall ask: which you?, etc. Sceptics about
mathematical objects should be sceptics about physical objects too.
Of course, scepticism about either is misplaced, and both the
Truth/Proof and the Truth/Verification problems are consequences of
confusion and are not real problems.
Perhaps this becomes more evident when we note that, in both
cases, the problem purports to challenge our canons of warrant (i.e.,
proof and verification, respectively); but carried to its logical con-
clusion, it also challenges our canons of meaningfulness. Why should
the structure of reality be what is presupposed by the grammatical
structure of our language as we have learned it? For example, the
meaningfulness of a sentence involving " + " presupposes the truth of
the sentence which expresses that " + " is well defined in the numbers.
So scepticism about truth will already imply scepticism about
meaning.
For both Benacerraf and D u m m e t t in the above cited papers, what is

special about discourse about physical objects is the possibility of
sense perception, and the difficulty that they raise for Platonism is
based on the absence in the case of mathematical objects of any such
"better understood means of human cognition."
346 w.w. TAIT
Some writers, for example, G6del (1964), Parsons (1979-80) and

Maddy (1980), attempt to m e e t this difficulty by arguing that there is
perception or something like it in the case of mathematical objects.
But, of course, from the point of view of the T r u t h / P r o o f problem, the
issue is not whether we perceive mathematical objects, but whether
our canons of proof obtain their meaning and validity from such
perceptions. A n d the answer to this seems to me to be clearly no. We
perceive sets, for example, only when we have mastered the concept
'set,' i.e., have learned how to use the word "set." For example, it does
not seem reasonable to suppose of people, before the c o n c e p t of set
was distinguished, that when they perceived a heap of pebbles, they
also perceived a set - a different object - and simply spoke am-
biguously. And, in whatever sense we m a y perceive numbers, it is hard
to see how that can provide a foundation for the use of induction to
define numerical functions, for example. T h e canons of proof are like
canons of g r a m m a r ; they are norms in our language governing the use
of words like "set," " n u m b e r , " etc. W h a t we call 'mathematical
intuition,' it seems to me, is not a criterion for correct usage. Rather,
having mastered that usage, we develop feelings, schematic pictures,
etc., which guide us. Of course, such intuitions m a y play a causal role
in leading us to correct arguments and even to new mathematical
ideas; but that is a different matter. In any case, the appropriate
response to the anti-Platonists is not to argue that there is something
like perception in the case of mathematics. R a t h e r it is to point out
that, e v e n in the supposedly paradigm case of sensible objects, per-
ception does not play the role that they claim for it. T h a t this is so is
manifest from the Truth]Verification problem. 3
Platonism is often identified with a certain " a c c o u n t " of truth in

mathematics, namely Tarski's. T h a t this is so for Benacerraf (1973) is
clear from the first of the a b o v e quotes from that p a p e r and
I take it that we have only one such account [of truth], Tarski's, and that its essential
feature is to define truth in terms of reference (or satisfaction) on the basis of a
particular kind of syntactico-semantical analysis of the language, and thus that any
putative analysis of mathematical truth must be an analysis of a concept which is a truth
concept at least in Tarski's sense. (p. 667)
It is difficult to understand how Tarski's 'account' of truth can have

any significant bearing on any issue in the philosophy of mathematics.
For it consists of a definition in mathematics of the concept of truth
for a model in a formal language L, where both the concept of a
formal language and of its models are mathematical notions. For
example, 3 in L is interpreted in terms of the mathematical 'there
exists.' But Benacerraf is concerned with mathematical truth, not with
truth of a formal sentence in a model. How can Tarski's account apply
here? What is the locus of the definition, i.e., what is the metalan-
guage? Not the language of mathematics, of course, since that is the
language whose meaning they wish to explain.
Benacerraf's remark that truth is defined "in terms of reference (or
satisfaction)" is at first sight puzzling, since truth is a special case of
satisfaction. But by "satisfaction" he undoubtedly means valuation,
i.e., assigning values to variables. But it is misleading to speak here of
reference. T h e model assigns values to the constants of L; but this,
like the notion of valuation, is expressed in terms of the notion of
function, and the concept of reference does not enter in. It is the more
misleading when Benacerraf goes on to advocate a causal theory of
reference.
An enlightening way to look at Tarski's truth definition is in terms
of the notion of an interpretation: with each formula th of L, we define
by induction on $ a formula I(th) (in the same variables) of the
metalanguage, i.e., of some part of the ordinary language of mathe-
matics in which we defined the model. T h e truth definition now is just
the 'material condition for truth': a sentence th of L is true iff I(~b).
Dummett 1973 writes
O n a platonistic interpretation of a mathematical theory, the central notion is that of

truth: a grasp of the m e a n i n g of a sentence belonging to the language of the theory
consists of a knowledge of what it is for that sentence to be true. (p. 223)
But when he speaks of what it is for ~b to be true or "of what the

condition is which has to obtain for [~b] to be true" (p. 224), to what
condition can he be referring here other than the condition that I(~b)?
But now, in what consists a grasp of the meaning of I ( $ ) 7 (Witt-
genstein 1953, §198: " . . . every interpretation, together with what is
being interpreted, hangs in the air . . . . . ") Dummett is aware of this
infinite regress, but he uses it as an argument against classical reason-
ing in mathematics which he identifies with Platonism (pp. 216-17).
348 w . w . TAIT
But of course the infinite regress disappears when we note that

Platonism does not consist in an interpretation of mathematical
theories. We do indeed interpret theories in mathematics, as when we
construct inner models of geometries or set theory or when we
construct examples of groups, etc., with certain properties. But we do
this in the language of mathematics, and our 'grasp' of this consists in
our ability to use it. D u m m e t t agrees with this (p. 217); but because he
takes Platonism to be an interpretation, he believes that this con-
clusion is an argument against Platonism.
Benacerraf and Dummett seem to me to be typical of those who
adopt a particular picture of Platonism. T h e picture seems to be that
mathematical practice takes place in an object language. But this
practice needs to be explained. In other words, the object language
has to be interpreted. T h e Platonist's way to interpret it is by Tarski's
truth definition which interprets it as being about a model - a Model-
in-the-Sky - which somehow exists independently of our mathematical
practice and serves to adjudicate its correctness. So there are two
layers of mathematics: the layer of ordinary mathematical practice in
which we prove propositions such as (1) and the layer of the Model at
which (1) asserts the 'real existence' of a number.
This is the picture that seems to lay behind the distinction in
Chihara (1973, pp. 61-75) between the 'mythological' Platonist and
the 'ontological' Platonist. T h e former simply does mathematics while
refraining from commitment to the interpretation. The latter accepts
the interpretation and so is committed to the 'real' existence of a
prime number greater than 10 and to the 'real' existence of 10.
But one cannot explain what this interpretation is supposed to be.
An interpretation in the ordinary sense is a translation. Into what
language are we supposed to be translating the language of ordinary
mathematics?
T h e Platonist, on this picture, is the Realist that Wittgenstein (1953)
criticizes at §402, along with the Idealist and the Solipsist - and he
might have added the Nominalist in the contemporary sense - , when
he says that the latter "attack the normal form of expression as if they
were attacking a statement" and that the Realist defends it as though
he "were stating facts recognized by every reasonable human being."
Needless to say, it is not this version of Platonism that I am defending
or that I even understand. Thus, I should not be understood to be
taking part in any realism/antirealism dispute, since I do not under-
stand the ground on which such disputes take place. As a m a t h e m a t i -

cal statement, the assertion that numbers exist is a triviality. W h a t
does it m e a n to regard it as a statement outside of mathematics? 4
It is ironic that D u m m e t t should think that Platonism is founded on a

comparison between mathematical reality and the physical universe
and that B e n a c e r r a f should think that it is m o t i v a t e d by the desire to
h a v e the same account of truth for mathematics as for its less lofty
empirical cousins. Plato, who was, as far as we know, the first Pla-
tonist, was entirely m o t i v a t e d by his recognition of the fact that the
exact empirical sciences of his day - geometry, arithmetic, astronomy
and music theory, for example - were not literally true of the sensible
world in the semantical sense and, indeed, did not literally apply to it.
H e did not h a v e our distinction between mathematics and empirical
science nor the idea of mathematical objects. Thus, in no sphere did
he think that scientific truth was truth in the semantical sense. A n d
Tarski's truth definition, while it concerns the semantical notion of
truth, is a piece of mathematics, concerning the mathematical notion
of a model of a formal language.
T h e fact is that w e c a n regard numerical propositions, say, as being
about a well-defined structure - the system of natural numbers. This
would be misleading only if it led us to think that our propositional
knowledge of this structure derives f r o m some sort of non-
propositional cognition of it or of its elements. In the case of sensibles,
on the other hand, there is no such well-defined structure. For e x a m -
ple, if m y desk remains the same object after I scratch it as before -
and clearly we must agree to this for a sufficiently light scratch - then
transitivity of identity fails for sensible objects, since a finite n u m b e r of
such scratches will reduce the desk to a splinter which we would not
identify with it. Nor can we avoid this conclusion by such resources as
speaking of the desk-at-an-instant; for this is no longer a sensible
object. M o r e o v e r , the predicates we apply to sensibles - for example,
of shape, color or size - are inherently vague. Thus, the canons of
exact reasoning, as e m b o d i e d say in some system of deductive reason-
ing, do not apply to the domain of sensible objects.
A n d when we idealize the domain of sensibles so that it takes on the
character of a well-defined structure and logic applies, then the other
350 w . w . TAIT
part of the picture of empirical knowledge painted by Dummett and

Benacerraf becomes manifestly problematic. For example, if reference
to my desk is replaced by reference to a spacetime region and
reference to colors, shapes and sizes by reference to magnitudes, then
the relevance of sense perception becomes less direct. It can no longer
be understood on the model of observation to observation sentence
and, at least judging by the literature on the subject of theory
confirmation, it is not one of the "better understood means of human
cognition." The relevant perceptions are of measurement; and the
measuring devices and measurements perceived are not elements of
the idealized domain, but are sensible objects like my desk. T h e role
of sense perception in confirming or applying mathematical models of
the phenomena is very complex. Yet it is only when we are thinking of
such a model, and not of the sensible world itself, that the picture of
the universe as a well-defined structure applies. Thus, when Benacer-
raf ignores the vagueness of the terms "large" and "older than", he is
not merely setting aside a complication (p. 663). He is raising the 'less
lofty empirical cousins' to an altogether loftier state where they too
may suffer their share of the attacks on Platonism. 5
Benacerraf seems to believe that the "better understood means of

human cognition" all involve causal interaction between the knower
and the known. He writes (p. 671)
I favor a causal account of knowledge in which for X to know that S is true requires
some causal relation to obtain between X and the referent of the names, predicates and
quantifiers of S. I believe in addition in a causal theory of reference, thus making the link
to my saying knowinglythat S doubly causal.
His problem then is that on the Platonist view we would not be able to
refer to mathematical objects, much less know anything about them,
since we do not causally interact with them. (Of course, one may feel
that the same problem arises for the referents of the predicates "large"
and "older than", to take Benacerraf's examples.) His argument for
"some such view" is that we would argue that X does not know that S
by arguing that he lacks the necessary causal interactions with the
grounds of truth of S - for example: he wasn't there. Of course, this
argument is plausible only if S is an empirical proposition (and
Benacerraf's example is empirical) and so it would seem to be a

complete non sequitur in the case of mathematical knowledge. In the
latter case, we might rather argue that X does not know that S by
arguing that he hasn't the c o m p e t e n c e to produce a proof of S.
However, Benacerraf thinks that whatever account we give of
mathematical knowledge, it should be extendable to embrace empiri-
cal knowledge as well (p. 262). And if that is so, then indeed the
correct account of knowledge of sensibles had better be extendable to
mathematics; and so there is no non sequitur. But his argument that
our account of mathematical knowledge should be extendable to
empirical knowledge is that to "think otherwise would be, among
other things, to ignore the interdependence of our knowledge in
different areas." But this seems to me to be a very weak argument.
Consider a case of interdependence: a mathematical prediction of the
motion of a physical object. First, we read the appropriate equations
off the data - i.e., we chose the appropriate idealization of the
phenomenon. Second, we solve the equations. Third, we interpret
the solution empirically. When Benacerraf speaks of mathematical
knowledge in his paper, the relevant kind of knowledge is knowledge
that $, where S is a mathematical proposition. But that kind of
knowledge is involved only at the second step, and it involves nothing
empirical. 6 T h e first and third steps involve only knowing how to
apply mathematics to the phenomena. But I don't see why an account
of this kind of knowing requires that, if empirical knowledge involves
causal interaction, then so does mathematical knowledge. T h e fact is
that we do know how to apply mathematics and we do not causally
interact with mathematical objects. Why doesn't this fact simply refute
a theory of knowing how that implies otherwise?
We may wish to explain why it is that idealization of the phenomena
works. We may also wish to explain why language and inductive
inference work. But these seem to me to be scientific 'why's', to be
answered by an account of how we process information and of how
this means of processing information (and so, creatures like us)
evolved.
Although it is unnecessary for the purpose at hand, let me comment
briefly (and certainly insufficiently) on the double causal interaction
that Benacerraf thinks must be involved in knowledge about objects.
It seems to me that when we speak of mathematical knowledge in the
ordinary way, we are referring to the ability to state definitions and
352 w . w . TAIT
theorems, to compute, to prove propositions, etc.: in general it is a

matter of knowing how, of competence. Anyway, it is this kind of
knowledge that we test students for. T h e ideas of propositional know-
ledge (knowledge that) and knowledge in the sense of acquaintance
with (knowledge of) also seem to me ultimately to reduce to the idea
of knowledge how; and this is so, not only in the (relatively simple)
domain of mathematics, but in general. This is of course very different
from the Cartesian notion of knowledge, since knowledge in this sense
presupposes a communal practice against which c o m p e t e n c e is to be
measured and so cannot serve as an external foundation for a critique
of that practice. Critique must come from within, measuring our
practice against the purpose of that practice. Also, knowledge in this
sense is not a matter of all or nothing: we recognize degrees of
knowing. (For example, when is giving a proof really giving a proof -
with understanding? Compare this with Wittgenstein's discussion of
reading.) We may indeed obtain a causal account of knowledge in this
sense; but it does not seem at all plausible that such an account, even
in the case of knowledge about physical objects, will involve causal
interaction with those objects. (The appearance of plausibility here
arises from the possibility that I might unwarrantedly believe a true
proposition. But one may reasonably doubt that there is a sense in
which my belief is unwarranted that would not show up in what I am
disposed to do.) As for the view that reference involves causal
interaction, the motive for this seems to me to confuse the question of
how we come to use a word in the way we do with the question of how
it is in fact used. (Cf. Wittgenstein 1953, §10.)
Platonism is taken to be an account of mathematics which says, for

example, that number theory is about a certain model. And then it is
challenged to tell us what that model is. One asks: how do we get to
know this model? Or: how do we know when we speak together that
we are speaking about the same model?, etc. It is as if we have a
formal system and are told that there is an intended model for it. But
no one can tell us what this model is and so we do not even know why
the ~ormal system is grammatically correct, much less valid.
Thus, D u m m e t t (1967, p. 210) writes
To say that we cannot c o m m u n i c a t e our intuition of the natural numbers une qui voc a l l y
by m e a n s of a formal system would be tolerable only if we had some other me a ns to
c o m m u n i c a t e it . . . . W e c a n n o t know that other people unde rs t a nd the notion of all
properties (of some set of individuals) as we do, and hence ha ve the same model of the
natural numb ers as we do.
and (pp. 210-11)
• . . we arrive at the d i l e m m a that we are unable to be certain w h e t h e r what s ome one else
refers to as the standard model is really isomorphic to the standard mode l we ha ve in
mind.
W h a t is my intuition of the numbers? I can only be said to have

intuitions about them - and then, only when I have some m i n i m u m
understanding of n u m b e r theory. A n d this understanding is not an
'intuition' (although there m a y be a c c o m p a n y i n g feelings and pic-
tures); it is a c o m p e t e n c e . W h a t does it m e a n to ' h a v e a model ' or to
' h a v e one in mind'? A n d what does it m e a n for us to h a v e the same
one? This can only m e a n that we do the same n u m b e r theory - the one
which is part of our c o m m o n language. 7 A n d I can ask: how do we
know that you h a v e the same physical universe that I have? D u m m e t t
seems to believe that we must explain our ability to c o m m u n i c a t e
mathematics and that Platonism is inadequate because it fails to do
this. But no explanation is necessary, unless one is calling for a general
empirical account of h u m a n communication. Mathematics presupposes
the fact of c o m m u n i c a t i o n - the fact of our c o m m o n disposition to use
and react to symbols in specific ways. If we lacked such c o m m o n
dispositions we could not be said to have mathematics any m o r e than,
if we lacked legs, could we be said to walk.
E v e r y reasonably schooled child understands the language of
arithmetic. It is the schizophrenic parent of the child who, m o t i v a t e d
by an inappropriate picture of meaning and knowledge, develops
'ontological qualms.' T h e picture is read into Platonism and then,
because it is inappropriate, Platonism, i.e., our ordinary conception of
mathematics, is rejected. T h e fact that the picture is generally inap-
propriate is simply ignored. We owe to Chihara a clear illustration of
the schizophrenia, namely in his mythological Platonist.
354 W. W. TAIT
Strangely, D u m m e t t u n d e r s t a n d s that the n o t i o n of a m o d e l is a

m a t h e m a t i c a l n o t i o n a n d that we c o n s t r u c t or d e s c r i b e m o d e l s in
m a t h e m a t i c s (1967, pp. 2 1 3 - 1 4 a n d 1963). H e is a s c r i b i n g to Pla-
t o n i s m an idea that he m u s t find i n c o m p r e h e n s i b l e . W h y ? Part of the
a n s w e r at least m a y b e f o u n d in the t e r m s in w h i c h he argues in the
1967 p a p e r that F r e g e ' s c o n t e x t p r i n c i p l e u n d e r m i n e s realism:
W h e n we scrutinize the doctrines of the arch-Platonist Frege, the substance of the

existential affirmation finally appears to dissolve. For him mathematical objects are as
genuine objects as the sun and moon: but when we ask what these objects are, we are
told that they are the references of mathematical terms, and 'only in the context of a
sentence does a name have a reference.' In other words, if an expression functions as a
singular term in sentences for which we have provided a clear sense, i.e. for which we
have legitimately stipulated determinate truth conditions, then that expression is a term
(proper name) and accordingly has a reference: and to know those truth conditions is to
know what its reference is, since 'we must not ask after the reference of a name in
isolation.' So, then, to assert that there are, e.g., natural numbers turns out to be to
assert no more than that we have correctly supplied the sentences of number theory with
determinate truth conditions; and now the bold thesis that there are abstract objects as
good as concrete ones appears to evaluate to a tame assertion that few would want to
dispute.
I, for o n e , w o u l d d i s p u t e the ' t a m e a s s e r t i o n ' that we h a v e " c o r r e c t l y

s u p p l i e d the s e n t e n c e s of n u m b e r t h e o r y with d e t e r m i n a t e t r u t h c o n -
d i t i o n s , " unless, of course, we are s p e a k i n g a b o u t s o m e f o r m a l system
of n u m b e r t h e o r y a n d we h a v e e x p l a i n e d their m e a n i n g in o r d i n a r y
m a t h e m a t i c a l terms. W e i n t e r p r e t f o r m a l systems; b u t in what lan-
g u a g e do we i n t e r p r e t o r d i n a r y m a t h e m a t i c s to give it ' d e t e r m i n a t e
truth conditions'?
T h e r e are m a n y difficulties a n d c o m p l e x i t i e s in c o n n e c t i o n with
F r e g e ' s c o n t e x t p r i n c i p l e ; he applies it in F r e g e (1884) to justify his
d e f i n i t i o n of the n u m b e r s a n d he applies it in F r e g e (1893) to justify
the i n t r o d u c t i o n of c o u r s e - o f - v a l u e s in t e r m s of w h i c h the n u m b e r s are
defined. O n e c o m p l i c a t i o n is that he is p r o p o s i n g a n e x t e n s i o n of the
o r d i n a r y m a t h e m a t i c a l discourse of his time - a n e w n o r m for m a t h e -
matics - a n d a n o t h e r is that his e x t e n s i o n is i n c o n s i s t e n t . Also, he
f o r m u l a t e s his a r g u m e n t (1884, §60) that n u m b e r s are o b j e c t s a g a i n s t
the b a c k g r o u n d of his o b j e c t / f u n c t i o n o n t o l o g y . Also, b e c a u s e he was
c o n c e r n e d with m a t h e m a t i c s , he did n o t c o n c e r n himself with the
p r o b l e m of t e r m s such as " H o m e r " w h i c h f u n c t i o n g r a m m a t i c a l l y like
terms but may not denote. Moreover, and possibly for the same
reason, he was concerned only with the role of names in the context of
declarative sentences and not in other kinds of linguistic expressions.
Finally, his formulation of the principle leaves open the question of the
meaning of sentences. But one nowhere finds him saying that mathe-
matical objects are the references of mathematical terms in answer to
the question of what they are. Rather, he is giving a criterion for the
meaningfulness of terms and he suggests in §60 that the criterion
extends beyond mathematics. And he then says that there is nothing
more to the question of the self-subsistence of numbers than the role
that number words play in propositions. I take this to mean that to say
that a term refers is to say that it is a meaningful term. T h e r e is
certainly no implication here that every mathematical object is the
reference of a term.
One should note that, anyway, Wittgenstein's reformulation of the
context principle, replacing the context of a proposition by the context
of a language (cf. 1953, §10 and the discussion in footnote 2), must,
for Dummett, also tame the same bold thesis. But, if so, Wittgenstein's
argument also tames the bold thesis that there are physical objects.
T h e issue between the Realist and the Idealist of §402 is a non-issue
too. So if abstract objects are not 'as good as' D u m m e t t conceives
concrete objects to be, then neither are concrete objects. Dummett
(1973) wishes to accept Wittgenstein's critique of language to the
extent of accepting the formula that meaning is determined by use as a
rough guide to the analysis of mathematical language. But I think that
the above passage shows that he does not accept the full consequences
of the critique. However, that is already shown by the fact, noted by
Lear (1982), that he adopts the above formula to argue for a re-
visionist view of mathematics contrary to Wittgenstein (1953, §124).
10.
On the basis of the preceding discussion, I think that we can now

begin to resolve the Truth/Proof problem. This problem arises
because there seem to be two, possibly conflicting, criteria for the
truth of a mathematical proposition: that it hold in the relevant
structure and that we have a proof of it.
T h e first step of the resolution is to see that the-first criterion is not a
criterion at all. T h e appearance that it is arises from the myth of the
356 w.w. TAIT
Model-in-the-Sky, of which we must - but do not seem to - have some

sort of nonpropositional grasp, with reference to which our m a t h e m a -
tical propositions derive their meaning and to which w e appeal to
determine their truth. T h e fact is that there are no such Models; there
are only models, i.e., structures that we construct in mathematics. Our
grasp of such a model presupposes that we understand the relevant
mathematical propositions and can determine the truth of at least
some of t h e m - e.g., those whose truth is presupposed in the very
definition of the model. Thus, rather than saying that holding in the
model is a criterion for truth, we would better put it the other way
around: being true is a criterion for holding in the model.
T h e myth of the Model tends to get attached to Platonism (or at
least to 'epistemological' Platonism in the sense of Steiner [1973])
because the view that mathematics is about things like the system of
numbers is c o m p a r e d with the view that propositions about sensible
things are about the physical world; and here there is a tendency to
believe that there is such a nonpropositional grasp, namely sense
perception, which does endow meaning on what we say and to which
we appeal to determine truth. But I hope that, if not what I have said,
then Wittgenstein's critique of this view of discourse about sensibles
will convince the reader that it is inadequate.
11.
H o w e v e r , the first step of the resolution of the Truth/Proof problem

m a y a p p e a r to have thrown out the b a b y with the bath water so far as
Platonism is concerned; and both Benacerraf and D u m m e t t think that
this is so. For, if we reject the myth of the Model, then how are we to
understand the notion of truth in mathematics? T h e r e might seem to
be no alternative here to identifying it with the notion of provability.
But then the independence of truth f r o m the question of what we
know or can know, which is the essence of Platonism, would be lost.
Benacerraf takes the less dogmatic line, not that this is the only
alternative, but that it is the only one that has been substantially
considered. But he takes the notion of proof here to be that of
deducibility in some formal system, and he argues for the obviously
correct conclusion that this yields an inadequate notion of truth.
D u m m e t t (1973) takes the view that, in giving up the myth of the
model, we are giving up the notion of truth and, with it, classical
mathematics. H e holds that the only viable alternative is to replace the

notion " A is true" by " p is a proof of A " , where the notion of proof
here is the intuitionistic one.
Although I have argued in an earlier paper (Tait, 1983) that
D u m m e t t is wrong here and, indeed, that the intuitionistic conception
is not entirely coherent, I nevertheless think that his response is, in a
sense, in the right direction. Namely, I think that the intuitionists' view
that a mathematical proposition A may be regarded as a type of object
and that proving A amounts to constructing such an object is right. Of
course, to say that we may regard A as a type of object does not mean
that we normally regard theorem proving as a matter of constructing
objects. Indeed, when we are interested in constructing an object, say
a real number, characteristically we are concerned with constructing
one with a particular property. As a proposition, 'Real Number' is
trivial. In the case of propositions, we are generally concerned with
finding s o m e proof and only rarely are we concerned with its proper-
ties. My point is rather that, independently of what we would say we
are doing when we are theorem proving, what we are actually doing
may be faithfully understood as constructing an object. T h e basic
mathematical principles of proof that we use, e.g., the laws of logic,
mathematical induction, etc., are naturally understood as principles of
construction.
12.
However, the intuitionists also hold that the objects that a proposition
A stands for, the objects of type A , are its proofs; and that I think is
wrong. A proof of A is a presentation or construction of such an
object: A is true when there is an object of type A and we prove A by
constructing such an object.
Here then is the answer to one of our questions: why is proof the
ultimate warrant for truth? T h e answer is of course that the only way
to show that there is an object of type A is to present one. (To prove
that there is an object of type A will mean nothing more than to prove
A, and that means to exhibit an object of type A.)
Consider the equation s = t between closed terms of elementary
number theory. What does this equation mean? We may say that it
expresses something about the system of numbers. That is certainly so,
358 w . w . TAIT
but it is also not to the point until we say what that something is,
without simply repeating the equation in the same or other terms.
T h e intuitionists seem to me very convincing when they say that
what the equation expresses is that there is a certain kind of com-
putation, namely, one which reduces s and t to the same term. For not
only do we initially learn the meaning of such terms and equations by
learning how to compute, but we take the existence of such com-
putations as the ultimate warrant for the equation. Thus, it seems
entirely natural to construe the equation as standing for the existence
of such a computation and to take the equation to be true precisely
when there is one.
Dummett (1973) accepts this analysis of such equations, but Dum-
mett (1967) feels that, in accepting it, one is rejecting the Platonist
point of view. His argument is that once we have accepted it there is
no reason to invoke the notion of truth in the sense of 'holding in the
system of numbers' to account for the meaning of the equation. But,
of course, we are not accounting for its meaning in this way and,
indeed, could not do so without circularity. That it holds in the system
of numbers - in other words, the fact about this system which it
expresses - is that there is such a computation. And we prove the
equation by producing one.
At least part of the reason why Dummett believes that the above
analysis of equations amounts to a rejection of Platonism is that he,
along with the intuitionists, identifies the proofs of the equations, i.e.,
the presentations of the computations, with the computations them-
selves; and when we do that we can no longer account for the
possibility of true but unprovable equations (Dummett 1967, p. 203).
One might object that the Platonist need not account for this pos-
sibility providing he can account for there being some true but
unprovable propositions. But the identification of computation with
proof is a special case of the intuitionistic identification of the object
with its construction in general. I do not believe that this identification
is ultimately intelligible; but one sees that, in accepting it, there is in
general no possibility of true but unprovable propositions.
However, it seems to me that, even in the case of the above sort of
equations, the intuitionists are wrong and that one should not identify
computations with proofs. For example, we easily prove 101° -- (105) 2
as an instance of a more general theorem; but in the canonical
notation 0, SO, SS0 . . . . for numbers, I shall be unable to explicitly
compute 101° and, even for terms with much shorter computations,
the chance of my computing accurately is very small. D u m m e t t (1977)
makes the distinction here between 'canonical proofs', which in the
present context are the explicitly presented computations, and the sort
of proof one obtains from proofs of more general propositions, which
are shorthand descriptions of canonical proofs. But when we know
that the computation is longer than human beings, individually or
collectively, are able to preform, we must ask the question: canonical
proof for whom? T o answer this by reference to an 'ideal computer'
seems highly unsatisfactory. In the first place, proof is a human
activity - and this would seem especially important to an intuitionist.
But secondly, I am unable to see a significant difference between
referring to an ideal computer who can compute f ( n ) for each n and
one who can compute it for all n and hence can decide whether
f ( n ) = O for all n or not. I don't mean that there isn't a formal
difference, but rather that it is hard to see why the one idealization is
legitimate and the other not. Yet the intuitionists reject the latter one,
which would lead to the law of excluded middle for arithmetic pro-
positions.
Computations are mathematical objects, forming a mathematical
system like the system of numbers. One may object to the use of the
term " c o m p u t a t i o n " here, because of its association with computing as
a human activity. But the term is also used in my sense, for example in
the mathematical theory of computability. T h e ease with which one
can confuse the two senses may contribute to the apparent plausibility
of the intuitionistic identification of the computation with its presen-
tation.
13.
When we extend the conception of mathematical propositions as types

of objects to propositions other than equations, the distinction be-
tween object of type A and proof of A becomes even more evident.
For example, let q~ be a function which associates with each object a
of type A, expressed by a : A, a type 4)a. T h e n
Vx:A. ckx
is the type of all functions f defined on A such that fx : dpx for all x : A,
360 w . w . TA~T
and
3x: A . 4~x
is the type of all pairs (x, y) such that x : A and y : ~bx. These defini-
tions of the quantifiers are essentially forced on us by the propositions
as types conception. 8 T h e remaining logical constants are definable
from the quantifiers, the null type 0 and the two-element type 2, whose
objects we denote by T and _1_.Thus, if we identify the t y p e / 3 with the
constant function ~b - B, then implication and negation are defined by
A---~B=Vx:A.B ~A=A--~O
and, if tot = A and tO& = B, then
AAB=Vx:2.tox AvB=3x:2.tox
Again, these definitions are essentially forced on us. 9
In this way, the logical operations appear as operations for con-
structing types and the laws of logic as principles for constructing
objects of given types. In this respect, there is no essential difference
between constructing a number or set of numbers and proving a
proposition. As Brouwer insisted should be the case, the logic of
mathematics becomes part of mathematics and not a postulate about
some transcendent model. However, Brouwer's view that the objects
of mathematics be mental objects does not seem to me coherent. And
the intuitionists' view that, for example, when we construct a number,
we should be able to determine its place in the sequence 0, 1, 2 , . . .
ignores the difficulty that we cannot in any case do this for sufficiently
complex constructions. Anyway, it is a restriction on ordinary
mathematical practice that is inessential to the conception of pro-
positions as types. T h e law of excluded middle amounts to admitting
objects of types A v ~ A which we may not otherwise be able to
construct; and this does indeed lead to the construction of numbers
whose positions in the above sequence are not computable. But it is
not essential to our conception that they should be.
An object of an V-type is a function and I have argued elsewhere
(cf. Tait, 1983) that, even in the case of constructive mathematics, one
must distinguish between a function and a presentation of it, by a rule
of computation or otherwise. I shall assume that, in the case of
nonconstructive mathematics, no argument is needed for this and,
therefore, that the distinction between objects of type A and proofs of

A is clear.
14.
Of course, we have not really specified the types 0 and 2 nor the
operations V and 3 until we have specified the principles of con-
struction or proof associated with them. A brief discussion of this
occurs in Tait (1983) (though the treatment of equations there is
inadequate) and a fuller treatment is in preparation. These principles
underlie mathematical practice in the sense that arguments that can-
not be reduced to them are as a matter of fact regarded as invalid.
Questions about the legitimacy of principles of construction or
proof are not, in my opinion, questions of fact. For mathematics
presupposes a c o m m o n mathematical practice and it is this that such
principles codify. Without agreement about these principles and their
application, there are no mathematical 'facts' (cf. note 8). Of course,
many factors, including the requirement of logical consistency, would
be involved in explaining why our mathematics takes the form it does;
but the view that there is some underlying reality which is independent
of our practice and which adjudicates its correctness seems to me
ultimately unintelligible. 1°
In this respect, the controversy between constructivists and non-
constructivists is similar to controversies about what is good or just
between people of different moral or political outlook. In the latter
case, one may ask what precisely is the issue. Why not simply use the
terms 'just1' and 'just2'? It seems to me that the answer is that there is
agreement about what I shall call the normative content of the term
'just' (or 'good'). Namely, to hold an action X to be just is to be
disposed to act in certain ways. A n d I am not referring here entirely to
linguistic acts such as affirming that one ought to do X. Rather, I have
in mind Aristotle's practical syllogism: to hold that X is just is to be
disposed to do X . If there were no agreement about this normative
content of the term 'just', then there would be no point in disputing its
material content, i.e., the question of what acts are to count as just.
But the latter sort of dispute seems to me not necessarily to involve
matters of fact, in as much as there may not be a sufficient basis of
ethical agreement to decide the issue.
In the same way, there is a normative content of the term 'valid'. T o
362 w . w . TAIT
hold an inference to be valid is to be disposed to make the inference.

B e c a u s e we agree about this normative content, it is significant to
argue about its material content, about what inferences are to count as
valid. But, here too, there may be no matter of fact, only a matter of
persuasion and adjustment of mathematical 'intuitions'. It is no ac-
cident that the dispute over the law of excluded middle often takes a
moralistic tone. T h e r e are no noncircular arguments for this law and,
in spite of all efforts to show otherwise, there are no arguments against
it which are not essentially to the effect that it leads to noncomputable
objects.
Constructivists do not deny any instance of the law of excluded
middle, of course: that would lead to inconsistency. Rather, they
refrain from its application. Thus, in principle, constructive mathema-
tics may be viewed as a restriction within ordinary mathematics on the
methods of proof or construction. 11 Aside from this, it is a striking fact
that there simply is no disagreement concerning the valid principles of
mathematical reasoning. Of course, I have not mentioned all of the
type-forming operations involved in mathematics; nor is it clear that
one could do so. For example, set theory involves the types obtained
by 'iterating' the operation of passing from a type A to P A = A - - ~ 2
into the transfinite. This involves the idea of creating new types by
inductive definitions. However, although there might be disagreement
about what inductive definitions one ought to admit, there is none
about the principles of proof to be associated with such a definition
when it is admitted. 12
15.
T h e answer to the initial question of this paper, concerning the

relation between truth and proof in mathematics, is that a proposition
A is true when there is an object of type A and that a proof of it is the
construction of such an object. T h a t there is an object of type A is the
'fact' about, say, the system of numbers that A expresses. It is clear
from this why proof is the ultimate warrant for truth.
T h e Platonist view that truth is independent of what we know or can
know is entirely correct on this view. In the first place, there may be
propositions which we can in principle prove on the basis of existing
mathematics, but whose proofs are too complex for us to process.
Secondly, there may be propositions which are not provable on the
basis of what we now accept, but are provable by means that we would
accept. When I Speak here of new means of proof, I do not of course
mean the acceptence of new logical principles concerning 0, 2, V, 3,
inductive definitions, etc., but rather the introduction of further types
to which we can apply these principles. For example, by the intro-
duction of new types we may construct numerical functions, i.e.
'proofs' of N---~ N, which we cannot otherwise construct.
It is, incidently, this open-endedness of mathematics with respect to
the introduction of new types of objects that refutes the formalistic
conception of mathematics, even if we leave aside the fact that
mathematical concepts such as the number concept have a wider
meaning than that given by their role in mathematics itself. T h e
formalists seem to me right - in any case, we have not one example to
refute them - that the above type-forming operations are completely
determined in mathematics by the principles of inference we as a
matter of fact associate with them. T h e incompleteness of formal
systems such as elementary number theory is best seen as an in-
completeness with respect to what can be expressed in the system
rather than with respect to the rules of inference. For example,
G6del's undecidable proposition for elementary arithmetic can indeed
be proved by induction; but the induction must be applied to a
property not expressed in the system itself.
NOTES
* Earlier versions of this paper were presented to the Philosophy Department of the
University of Wisconsin at Madison in the Winter of 1984, at the Tarski Memorial
Conference at Ohio State in the spring of 1984 and at the Pacific Division meeting of
the A P A in the spring of 1985. I received many valuable comments on all of these
occasions and, in particular, from Paul Benacerraf and Clifton Mclntosh, who com-
mented on my paper at the A P A meeting. I should also like to thank Michael Friedman
for his comments on an earlier version and for our many discussions of its subject
matter.
Many other contemporary authors could of course have been cited for essentially the
same point. I shall focus primarily on Benacerraf (1973) and D u m m e t t (1967, 1973) in
citing the literature because these seem to me to represent most clearly and fully the two
most important formulations of difficulties with Platonism. Benacerraf's paper is
frequently cited as grounds for revisionist foundations of mathematics - e.g., in Field
(1980, 1981), Kitcher (1978, 1983) and Steiner (1975). It consists in arguing that, in the
context of mathematics, there is an apparent conflict between our best theory of truth,
which is Tarski's, and our best theory of knowledge, which is causal, because we do not
364 W.W. TAIT
causally interact with mathematical objects. As I understand him, Benacerraf himself,

unlike many who cite him, is not calling for a revision of our conception of mathematics
but only for a resolution of the apparent conflict. Dummett's critique of Platonism rests
on a conception of meaning which he argues is incompatible with Platonism and,
indeed, leads to the intuitionistic conception of mathematics; and so it is revisionist. My
purpose, however, is not to review these papers. I cite them only because I wish to
undermine conceptions which I myself cannot coherently formulate. On the other hand,
I shall, I believe, resolve the difficulties that they find with Platonism in the course of
this paper.
2 The issue here is not 'inscrutability of reference.' That idea makes sense in connection
with translating one language into another. But in what sense is our reference to the
chair or to the number two inscrutable? When Wittgenstein (1953) writes "What is
supposed to show what [the words] signify, if not the kind of use they have" (§10), his
point is not that there is a well defined universe of things (perhaps described in the
language of God) and that a word succeeds in refering to one of these things rather than
another because of the kind of use it has. Rather, it is that we call a word 'referring'
because of the kind of use that it has. And we ask the question "To what does the word
' X ' refer?" in language, and it can only be answered there, by pointing perhaps or by
saying " X refers to Y", where " Y " is " X " or some other tenn.
3 Gtdel 1964 wrote:
But, despite their remoteness from sense experience, we do have something like a
perception also of the objects of set theory, as is seen from the fact that the axioms
force themselves Ul~Onus as being true. (pp. 483-84)
Many authors regard G6del as an archetypal Platonist and this passage as a bold
statement of what every Platonist must hold if he is to account for mathematical
knowledge. In the words of Benacerraf (1973) (who, incidently, inadvertently left out
the words "something like" in quoting the above passage):
[Gtdel] sees, I think, that something must be said to bridge the chasm, created by his
realistic and platonistic interpretation of mathematical propositions, between the
entities that form the subject matter of mathematics and the human knower . . . . he
postulates a special faculty through which we "interact with these objects. (p. 675)
But I don't think that this is a fair reading of G-tdel's remark. To understand what he
means by "something like perception", one should look at his argument for it: "the
axioms force themselves upon us as being true." One should also look at the paragraph
immediately following the quoted one:
It should be noted that mathematical intuition need not be conceived of as a faculty
giving an immediate knowledge of the objects concerned. Rather it seems that, as in
the case of physical experience, we .form our ideas also of those objects on the basis of
something else which is immediately given. Only this something else here is not, or
not primarily, the sensations. That something besides the sensations actually is
immediately given follows (independently of mathematics) from the fact that even our
ideas referring to physical objects contain constituents qualitatively different from
sensations or mere combinations of sensations, e.g., the idea of object itself, whereas,
on the other hand, by our thinking we cannot create any qualitatively new elements,
but only reproduce and combine those that are given. Evidently the "given" underly-
ing mathematics is closely related to the abstract elements contained in our empirical
ideas. It by no means follows, however, that the data of the second kind, because they
cannot be associated with actions of certain things upon our sense organs, are
something purely subjective, as Kant asserted. Rather they, too, may represent an
aspect of objective reality, but, as opposed to the sensations, their presence in us may
be due to another kind of relationship between ourselves and reality.
If anything is being 'postulated' here it is this other kind of relationship, not a faculty.
This relationship is to account for the objective validity, not only of the 'something like
a perception' of mathematical objects, but also of our ideas referring to physical objects.
For it concerns the 'given' underlying mathematics, which are closely related to the
abstract elements contained in our empirical ideas - e.g., the elements giving rise to our
idea of an object (cf. Theaetetus 184d-186). That G6del intends this relationship to be
necessary for the objective validity of empirical as well as mathematical knowledge is
indicated by the first sentence of the next paragraph, which I have already partially
quoted, indicating that the question of the objective existence of mathematical objects is
the exact replica of that concerning the objective existence of the outer world.
But he writes that the former question "is not decisive for the problem under
discussion here. The mere psychological fact of the existence of an intuition which is
sufficiently clear to produce the axioms of set theory and an open series of extensions of
them suffices to give meaning to the truth or falsity of propositions like Cantor's
continuum hypothesis." The point seems clear: the 'something like a perception',
namely, mathematical intuition, is not what bestows objective validity on our theorems,
any more than the perceptions of the Brain-in-the-Vat bestow objective validity on its
assertions about the physical world. Yet, the Brain-in-the-Vat will have grounds for
asserting (2); and, in the same way, mathematical intuition yields grounds for asserting
(1). Thus, the 'something like a perception' is not the 'another kind of relationship
between ourselves and reality' to which G6del refers.
I do not entirely agree with G6del here. What is objective about the existence of
mathematical or empirical objects is that we speak in a common language about them -
and this includes our agreement about what counts as warrant for what we say. And this
view guides my estimation, stated above, of the role of mathematical intuition vis-a-vis
grounds for asserting mathematical propositions. I cannot make the distinction G6del
seems to want to make between subjective validity, founded on our intuition, and
objective validity. But it is worthwhile to point out that G6del's 'something like a
perception' is not a 'special faculty through which we interact with [mathematical]
objects.' Indeed, he was far less naive about the role of ordinary sense perception in
empirical knowledge than the many writers who have focused on the passage in question
as the Achilles heel of Platonism.
4 The 'external question' of the existence of numbers would seem to presuppose a
univocal and nonquestion begging notion of existence against which to measure
mathematical existence. But what is it? Quine (1953, fn. 1), indeed attempts an
argument to the effect that the desire to distinguish mathematical from spacetime
existence on the grounds that the latter, but not the former, involves empirical
investigation is unfounded. His argument is that showing that there is no ratio between
the number of centaurs and the number of unicorns involves empirical investigation.
366 w.w. WAIT
But the mathematical fact here is that 0 has no reciprocal; and that needs no empirical
investigation.
I think that Carnap (1956) is right that 'external questions' of existence have no prima
facie sense. But his attempt to make an absolute distinction between theoretically
meaningful questions and those without theoretical meaning on the basis of his notion of
a linguistic framework fails. For example, his framework for number theory is a formal
system. But correct and sufficiently expressive formal systems for number theory are
incomplete and, moreover, do not express all the properties of numbers. In later
writings, Carnap attempted to solve the problem of incompleteness by allowing the
system to contain the infinitary oJ-rule. But now the internal question " D o e s there exist
a number n such that ~b(n)" can only mean " D o e s there exist an infinitary deduction of
3x~b(x)?". But this is an external question and may be mathematically nontrivial. But,
anyway, linguistic frameworks are constructed in our everyday language; and it is hard
to see how, lacking a precise notion of theoretical meaningfulness for it, we can
convincingly determine when we have a 'good' framework and when we do not.
5 In the nominalism of Field (1980, 1981), the mathematical model is identified with the
physical world. Thus, spacetime regions become nominalistically acceptable objects and
mathematics is involved only insofar as such objects as numbers, sets and function are.
Regions are real because we causally interact with them or at least can do so with some
of them. This idea is developed in the 1980 book to show how to free Newton's theory
of gravitation of mathematics, to make it a nominalistic theory. Of course, there is a
difficulty in that, for a wide range of phenomena, Newton's theory is inadequate and, if
we replace it by Einstein's theory, for example, the 'nominalization' has yet to be
demonstrated. Moreover, Einstein's theory does not account for other ranges of
phenomena and it is open whether it is compatible with an account of them. Finally,
even if we had a reasonable universal physics, i.e., an account of all known forces, we
should still have to ask (at least if we took Fiel~l's position) whether it was true. Well, let
us suppose that we have such a 'true' universal physics, which is a spacetime physics.
W o n ' t causation be a relation between spacefime points or regions? But, unless some
Supreme Court decisions - made with greater precision than, not only is it accustomed
to, but than it is in principle capable of - are begged, I am not a spacetime region and so
do not causally interact with such things. The world of chairs and rooms and us is
different from the world of mathematical physics. The latter is called an idealization of
the former; and this only means that we can use the mathematical theory in a certain
way.
6 Putnam (1979) also seems confused on this point. He writes that Wittgenstein may
have had in mind the following 'move':
One might hold that it is a presupposition of, say, "2 + 2 = 4," that we shall never
meet a situation that we would count as a counterexample (this is an empirical fact):
and one might claim that the appearance of a "factual" element in the statement
"2 + 2 = 4" arises from confusing the mathematical assertion (which has no factual
content, it is claimed) with the empirical assertion first mentioned,
The 'empirical fact' and 'empirical assertion first mentioned', I assume, is that we shall
never meet a situation that we would count as a counterexample. But this is, for
Wittgenstein and in fact, no more an empirical fact than that we shall never meet, in a
game of chess, a situation which we would count as one in which the king is captured.
Of course, neither of these assertions is a prediction about our future behavior or an
assertion about our past behavior; they are each part of a description of a certain game.
It is indeed an empirical fact that we play the game - that we do mathematics and play
chess - but that is another matter. Putnam goes on:
This move, however, depends heavily on overlooking or denying the circumstance

that an empirical fact can have a partly mathematical explanation. Thus, let T be an
actual (physically instantiated) Turing machine so programmed that if it is started
scanning the input "111," it never halts. Suppose that we start T scanning the input
"111," let T run for two weeks, and then turn it off. In the course of the two-week
run, T did not halt. Is it not the case that the explanation of the fact that T did not
halt is simply the mathematical fact that a Turing machine with that program never
halts on the input, together with the empirical fact that T instantiates that program
(and continued to do so throughout the two weeks)?
The answer is simply: yes. But what has this to do with the fact that the mathematical
proposition "2 + 2 = 4" or "Turing machine t with input '111' never halts" is not the
sort of proposition for which the idea of empirical counterexamples makes sense? This
example is no different from our explanation of the motion of a physical object. We
model the behavior of T with t. If it is a good model (and this idea defies precise
analysis), then the fact that t doesn't halt (in the mathematical sense) should lead us to
believe that T doesn't (in the physical sense) halt. But what has this to do with the
conceivability of an empirical counterexample to the statement that t doesn't (in the
mathematical sense) halt? The sense in which it is claimed that "2 + 2 = 4" has no
'factual content' is not intended to imply that it has no empirical applications.
7 Consider systems ~ = (A, a, f), where A is a type of object, a is an object of type
A(a : A) and f is a function from A to A (f: A--~ A). Dedekind (1887) characterized the
system N = (N, 0, S) of numbers as such a system in which 0 ~ Sn for all n : N ,
S m = Sn---~ m = n and, if X is any set of numbers containing 0 and closed under S, then
it is the set of all numbers. There is no question of identifying the system of numbers: it
is, as Dedekind puts it (§73), a 'free creation of the human mind.' We have created it in
the sense that we have specified once and for all its grammar and logic. Moreover, given
any other system ~/satisfying this characterization, the proof that ~/is isomorphic to N
is a triviality and we shall not disagree about that. We might indeed disagree about the
principles used to construct some set P of numbers or some system ~1; but that is a
different matter and, anyway, if we leave aside those who wish to use only constructive
principles, then as a matter of fact, there is no such disagreement (cf. §15). Moreover,
the possibility of this kind of disagreement exists even in constructive mathematics,
which D u m m e t t (1973) is advocating. In that case, Dedekind's characterization should
be replaced by the classically equivalent one essentially given by Lawvere (1964),
namely that X has the property of unique iteration: given any system ~t, the equations
gO = a and gS = fg define a unique function g: N---~ A. But we may still disagree about
when a system ~ / h a s been legitimately introduced.
8 Suppose that we already have that, for any x : A, ~b(x) is already identified with a type
of object. T h e n 3 x : Ack(x) means that, for some x : A, ~b(x), and so that, for some x : A,
368 w.w. TAIT
there is a y:th(x), and so that there is a pair (x, y) of the required type. V x : A t h ( x )
means that ~b(x) for all x : A, and so that, for each x : A, there is a y : ~b(x). So we have
'reduced' the meaning of V to 'for all x, there exists a y.' We avoid an infinite regress
here only by taking the latter to mean that we have a function f which gives us y = fx
for each x. This is, as a matter of fact, the way in which we do reason. The appearance
that it isn't arises from the fact that we often are thinking of the reasoning as taking
place in a model in which no such [ occurs. So Vx : Aep(x) may be true in the model
without there being the required f in the model. But that of course is different from
saying that there is no s u c h / .
Our analysis of the quantifiers yields the Axiom of Choice in the form
V x : A 3 y : B ~ ( x , y) ~ 3 z : A ~ B V x : Ad/(x, zx).
For let f be of the antecedent type. Then for each x : A , fx is of the form (y, u), where u
is of type ff(x,y). Let z:A---->B be defined by z x = y and let v : V x : A ~ ( x , zx) be
defined by vx = u. Then (z, v) is of the type of the conclusion. The argument above for
our analysis of the universal quantifier looks itself like an application of the Axiom of
Choice:
Vx : A 3 y ( y : ~b(x))--~ 3 f V x : A ( f x : d,b(x)).
But there are two respects in which it is different. First, it contains two variables, y and
jr, whose type is unspecified and, secondly, it involves 'propositions' of the form ' u : C'.
Concerning the first point, the notion of a mathematical object in general seems a
problematic notion and certainly is no part of the Platonistic conception that I am
discussing. Concerning the second, ' u : C' is not a mathematical proposition in the sense
that I am discussing. Otherwise, we would have to know what are the objects v of type
u : C, the objects of type v : (u : C), etc., leading to an infinite regress. The fact is that we
have a type C only when we have agreement as to what counts as an object of type C.
Thus, statements such as ' u : C ' are grammatical statements. It is the wrong picture to
think that there is a universe of 'mathematical objects' and then we must determine for
one of them, u, what type it has. (This seems to me to be the view behind Jubien (1977),
where its absurdity is well illustrated.) 0 is a mathematical object because it is a number.
If I have introduced an object u of type C, then either u : C or else I have been
indulging in nonsense.
9 Cf. Tait (1983). In the case of negation, note that A A B--> 0 should he a requirement
for a negation B of A. But when this holds, B - - ~ - I A ; and so -hA is the weakest
candidate for negation. One may feel, nonetheless, that negation presents a counterex-
ample to the view of propositions as types; since, if ' A is true' is to mean that there is an
object of type A, then ' A is false' ought to mean that there is no such object, and this is
not an existence statement. But if there is no object of type A , then there is an object of
type A---~ 0, namely the null function. But, in any case, there is something deceptive
about the discussion. What does 'not' mean when we say that it is not the case that there
is an object of type A? For this too is a mathematical proposition and, indeed, simply
means 7 A . We should not think that there is a meaning of 'not' that somehow
transcends mathematical practice.
1o The question of the truth of mathematics, as opposed to truth in mathematics has
historically been the concern of many philosophers. In some cases, e.g., Plato and
Leibniz, this question has been distinguished from that of why mathematics applies to
the phenomena and in others, such as Aristotle and Kant, it has not. This latter
question, of why mathematization of the phenomena works, has itself been a source of
anti-Platonism. But, as I have indicated in §7, the only kind of answer to that question
would be in terms of cognitive science and an account of why it is that we have evolved.
E1 Of course, if one is interested only in constructive mathematics, one may diverge
from the classical development of, say, analysis, by choosing concepts more amenable to
constructive treatment than the classical analogues. My point is only that the principles
of construction and reasoning used in the development remain classically valid. Ap-
parent counterexamples such as Brouwer's proof that every real-valued function on the
continuum is continuous are a result of ambiguity, not of using classically invalid
principles.
12 There is another method of obtaining new types which derives from Dedekind (1888)
and which we may refer to as 'Dedekind abstraction.' For example, in set theory we
construct the system (to, th, tr) of finite von Neuman ordinals, where trx = xU{x}. We
may now abstract from the particular nature of these ordinals to obtain the system ~/" of
natural numbers. In other words, we introduce N together with an isomorphism between
the two systems. In the same way we can introduce the continuum, for example, by
Dedekind abstraction from the system of Dedekind cuts. In this way, the arbitrariness of
this or that particular 'construction' of the numbers or the continuum, noted in
connection with the numbers in Benacerraf (1965), is eliminated. It is incidently
remarkable that some authors such as Kitcher (1983) have taken Benacerraf's obser-
vation to be an argument against identifying the natural numbers with sets, but have
been content to identify the real numbers with sets, although there are again various
ways to do that. Kitcher (1978) contains an amazing argument based on Benacerraf's
observation, to the effect that Platonism is false: on grounds of economy, all 'abstract'
objects should be sets. Numbers are abstract. But there is no canonical representation of
the numbers as sets. Therefore, the view that there are such things as numbers is false.
(A person makes up a budget and, on grounds of economy, fails to budget in for food.
But we need to eat. So the notion of a budget is incoherent.)
REFERENCES
Benacerraf, P.: 1965, 'What Numbers Could Not Be', Philosophical Review 74, 47-73.
Benacerraf, P.: 1973, 'Mathematical Truth', The Journal of Philosophy 70, 661-79.
Benacerraf, P. and H. Putnam (eds.): 1984, Philosophy of Mathematics: Selected
Readings, 2nd. edn., Cambridge University Press, Cambridge, Mass.
Carnap, R.: 1956, 'Empiricism, Semantics, and Ontology', Meaning and Necessity, 2nd.
edn., University of Chicago Press, Chicago, reprinted in 3.
Chihara, C.: 1973, Ontology and the Vicious Circle Principle, Cornell University Press,
Ithaca.
Dedekind, R.: 1888, Was sind und was sollen die Zahlend, Brunswick, Vieweg.
Dummett, M.: 1963, 'The Philosophical Significance of GSdel's Theorem', Ratio 5,
140-55, reprinted in 11, page references are to 11.
Dummett, M.: 1963, 'Platonism', first published in 11.
Dummett, M.: 1975, 'The Philosophical Basis of Intuitionistic Logic', H. E. Rose and J.
370 w.w. TAIT
C. Shepherson (eds.), Logic Colloquium '73, North-Holland, pp. 5-40, reprinted in

11, page references are to 11.
Dummett, M.: 1977, Elements of Intuitionism, Clarendon Press, Oxford.
Dummett, M.: 1978, Truth and Other Enigmas, Harvard University Press, Cambridge,
Mass.
Field, H.: 1980, Science Without Numbers, Princeton University Press, Princeton, N.J.
Field, H.: 1981, 'Realism and Anti-Realism About Mathematics', Rice University
Conference on Realism and Anti-Realism, unpublished.
Frege, G.: 1884, Die Grundlagen der Arithmetik, Verlag von Wilhelm Koebner,
Breslau.
G6del, K.: 1947, 'What Is Cantor's Continuum Problem?', American Mathematical
Monthly 54, 515-25. A revised and supplemented version appears in 3. Page
references are to 3. 'G6del 1964' refers to the supplement in the later version (which
first appeared in the 1st. edn. of 3 in 1964).
Jubien, M.: 1977, 'Ontology and Mathematical Truth', Nous 11.
Kitcher, P.: 1978, 'The Plight of the Platonist', Nous 12, 119-36.
Kitcher, P.: 1983, The Nature of Mathematical Knowledge, Oxford University Press.
Lawvere, W.: 1964, 'An Elementary Theory of the Category of Sets', Proceedings of the
National Academy of Science 52, 1506-11.
Lear, T.: 1982, 'Leaving the World Alone', The Journal of Philosophy.
Maddy, P.: 1980, 'Perception and Mathematical Intuition', Philosophical Review 89,
163-96.
Parsons, C.: 1979-80, 'Mathematical Intuition', Proceedings of the Aristotelian Society
pp. 145-68.
Putnam, H.: 1979, 'Analyticity and Aprioricity: Beyond Wittgenstein and Quine',
Midwest Studies in Philosophy IV (Studies in Metaphysics).
Quine, W. V.: 1953, 'On What There Is,' From A Logical Point of View: Harvard
University Press, Harvard.
Steiner, M.: 1975, Mathematical Knowledge, Cornell University Press, Ithaca.
Tait, W.: 1983, 'Against Intuitionism: Constructive Mathematics Is a Part of Classical
Mathematics', The Journal of Philosophy 12, 173-95.
Wittgenstein, L.: 1953, Philosophical Investigations, Macmillan.
Department of Philosophy
University of Chicago
Chicago, IL 60637
U.S.A.

Truth and Proof, The Platonism of Mathematics - William W. Tait

Uploaded by

Copyright:

Available Formats

Truth and Proof, The Platonism of Mathematics - William W. Tait

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Truth and Proof, The Platonism of Mathematics - William W. Tait

Uploaded by

Copyright:

Available Formats

W. W.

TRUTH AND PROOF: THE PLATONISM OF

W h a t is the relation in m a t h e m a t i c s between truth and proof?

Synthese 69 (1986) 341-370.

not a warrant, then we have no mathematical knowledge at all. An

Many who reject Platonism on the grounds of the Truth/Proof prob-

Platonism, as a philosophy of mathematics, is founded on a simile: the comparison

and he takes it as u n p r o b l e m a t i c that (2) has this form. But

I shall h a v e m o r e to say bearing on these passages in the course of m y

W h y does the e x p e r i e n c e that I describe as "seeing a chair in the

For both Benacerraf and D u m m e t t in the above cited papers, what is

Some writers, for example, G6del (1964), Parsons (1979-80) and

Platonism is often identified with a certain " a c c o u n t " of truth in

It is difficult to understand how Tarski's 'account' of truth can have

O n a platonistic interpretation of a mathematical theory, the central notion is that of

But when he speaks of what it is for ~b to be true or "of what the

But of course the infinite regress disappears when we note that

stand the ground on which such disputes take place. As a m a t h e m a t i -

It is ironic that D u m m e t t should think that Platonism is founded on a

part of the picture of empirical knowledge painted by Dummett and

Benacerraf seems to believe that the "better understood means of

Benacerraf's example is empirical) and so it would seem to be a

theorems, to compute, to prove propositions, etc.: in general it is a

Platonism is taken to be an account of mathematics which says, for

Thus, D u m m e t t (1967, p. 210) writes

and (pp. 210-11)

W h a t is my intuition of the numbers? I can only be said to have

Strangely, D u m m e t t u n d e r s t a n d s that the n o t i o n of a m o d e l is a

W h e n we scrutinize the doctrines of the arch-Platonist Frege, the substance of the

I, for o n e , w o u l d d i s p u t e the ' t a m e a s s e r t i o n ' that we h a v e " c o r r e c t l y

On the basis of the preceding discussion, I think that we can now

Model-in-the-Sky, of which we must - but do not seem to - have some

H o w e v e r , the first step of the resolution of the Truth/Proof problem

mathematics. H e holds that the only viable alternative is to replace the

When we extend the conception of mathematical propositions as types

therefore, that the distinction between objects of type A and proofs of

hold an inference to be valid is to be disposed to make the inference.

T h e answer to the initial question of this paper, concerning the

causally interact with mathematical objects. As I understand him, Benacerraf himself,

This move, however, depends heavily on overlooking or denying the circumstance

C. Shepherson (eds.), Logic Colloquium '73, North-Holland, pp. 5-40, reprinted in

You might also like