Write paper 1700-2000words
dream86
KANT-APrioriTruths.pdf
1
A Priori Truths
Greg Restall*
The University of Melbourne
Philosophers love a priori knowledge: we delight in truths that can be known from the
comfort of our armchairs, without the need to venture out in the world for
confirmation. This is due not to laziness, but to two different considerations. First, it
seems that many philosophical issues aren’t settled by our experience of the world —
the nature of morality; the way concepts pick out objects; the structure of our
experience of the world in which we find ourselves — these issues seem to be decided
not on the basis of our experience, but in some manner by things prior to (or
independently of) that experience. Second, even when we are deeply interested in
how our experience lends credence to our claims about the world, the matter remains
of the remainder: we learn more about how experience contributes to knowledge when
we see what knowledge is available independent of that experience.
In this essay we will look at the topic of what can be known a priori. We will start with
some examples of truths which we have thought to be uncontentiously a priori known.
Examples
From Logic: I can know that if every student has either passed an exam or completed
an assignment, then either every student has passed an exam, or some student has
completed an assignment. I can know this without concerning myself with the details
of which students have passed an exam or have completed an assignment, for I can
reason as follows: Let’s suppose every student has either passed an exam or completed
an assignment. We want to show that either every student has passed an exam or
some student has completed an assignment. Suppose I choose a student. By
hypothesis, he or she has either passed an exam or completed an assignment. If the
assignment is completed, then I can conclude that some student has completed an
assignment, which gives us what I wanted to show. If this piece of reasoning fails for
every student, then we see that every student has completed an exam, which also gives
us our desired conclusion.
This piece of argumentation suffices to prove what we wished to show. It used no
details of students, assignments or exams. The form of the reasoning would work for
any claim of the structure: if every F is either G or H, then either every F is G or some F
is H. Valid deductive reasoning seems to give us a priori knowledge on the basis of
logical structure.
* Thanks to John Shand for the invitation to write this essay, for supererogatory patience in waiting for
its arrival, and for very useful comments on the initial draft, to Allen Hazen for very many discussions
on the issues contained herein, and to my graduate students at the University of Melbourne, who have
put up with many more discussions about a priori truth than is healthy.
2
From Semantics: It seems that I can know that all bachelors are unmarried, without
going to the trouble of interviewing bachelors and checking their marital status. I
know this because if I find out that someone is married, then this counts decisively
against their being a bachelor. I know this because this is how I understand the terms
‘married’ and ‘bachelor.’ It seems that the meanings of the expressions I use governs
how I treat the evidence I may find. I don’t wait to learn of the marital status of a
number of bachelors to conclude a generalization on the basis of this evidence. I use
this generalization to govern the evidence I encounter.
From Mathematics. Take a case of mathematical
reasoning. A salient example in the development of our
understanding of the a priori and is relationship with
logic and the analytic is the intermediate value theorem,
which states that every continuous function f of real
numbers, where the value f(–1) that it takes on the input
–1 is smaller than 0, and whose value f(1) that it takes on
the input 1 is greater than 0, has some input value x
between –1 and 1 where the output f(x) is 0. A
continuous function which is below the origin line at –1
and above the origin line at 1, must have crossed the line at
some point between –1 and 1. In other words, any continuous path (a path without
jumps or breaks) starting on one side of a line and ending up on another side of that
line must have at least one point at which it crosses the line.
This is obvious. We can see that it is true in many cases, and furthermore, we may find
it very difficult to know what it would be for it to be false in any case at all. For many
years, mathematicians claimed that they could see that the theorem is true in full
generality, without being able to offer anything resembling a proof. The key is the
notion of continuity. A function is continuous if it has no breaks or jumps — and
making this notion precise, in the development of the Calculus in the work of Bolzano
and Weierstrass (see Coffa (1993), Chapters 2 and 3) was required for this piece of a
priori knowledge to make the transition from something which seems obvious but is
hard to justify, to something which could be proved on the basis of an analysis of the
concepts involved.
From Indexicality: I may have amnesia and not remember where I am. However,
wherever I am, if I say “I am here now” this seems to be something which is true, and
which I can know. You use external evidence to find where I am, but I don’t need to
take in any evidence to convince myself that I am here. Evidence enters the picture
when I want to know where “here” is, but as it stands, this is an item of a priori
knowledge.
This case has interesting features, for while it seems to be a tautology for me to say that
I am here now, that is not something that you can know a priori. You can know that
whenever I say “I am here now” that I am speaking truly, but you cannot always know
what I am saying. To see this consider me speaking to my spouse at the end-of-school-
year concert, which I have unexpectedly been able to attend: “Zachary doesn’t know that
f(x)
–1 1
3
I’m here now, he thinks that I’m still at work.” I can know a priori that I am here, but
Zachary, to learn that fact, must see me in the audience.
From Ethics: consider the judgment that making other people happy is, all things
considered, a good thing to do. Perhaps when you consider this judgment you
remember particular acts in which you have made others happy. But upon reflection,
it seems that there’s no reason to appeal to this or that experience, or this or that
evidence that happiness is a good thing. Perhaps it is bound up in the very notions of
happiness, others, and goodness, and we reasonably could come to that conclusion on
the basis of reflection on those notions. Many believe that our fundamental ethical
principles are known prior to the evidence. Such statements certainly don’t look like
they can be refuted (or proved) on purely empirically, and the fact that they concern
what should happen seems to mean that mere description of how the world is cannot
count decisively against the kind of claim about how it should be,
Are these good examples of a priori truths? How do they fare when we attend to the
connections, if any, between a priority, necessity, analyticity, and infallibility? What
have philosophers said about the notion of the a priori, and does the notion survive
close scrutiny? These are the questions we will examine in the rest of this essay.
Definitions
As we have seen in our initial meeting with examples, an a priori truth is something
that can be known independently of any particular evidence or experience. This rough
and ready idea has been the basis of the claim to a priority for each of our examples.
You do not need to know anything about the world in order to verify that if all Fs are G
or H, then either all F are G or some F are H, or that all bachelors are unmarried, and I
need to take in no evidence to know that I am here. Or so the reasoning has gone.
This does not mean that the notion of a priori knowledge is unproblematic. We have
not given a characterization what is necessary for knowledge to be independent of the
external evidence, and neither have we had anything to say about what is to count as
evidence, and of the evidence we have, what it means for some of it to be external.
Similarly, there is confusion on another aspect of the notion of independence. It is one
thing to say that knowledge can be acquired without appeal to some antecedent
experience. It is another to say that that item of knowledge cannot have its status as
knowledge undermined by further experience of the outside world. This is a much
stronger requirement. I may be convinced of some reasonably complicated
mathematical result (such as the intermediate value theorem) by way of stepping
carefully through some proof, with the help of a mathematically sophisticated friend.
Although aided by props and outside support, this can count as a priori reasoning, for
none of the props are essential to the content of what is conveyed, if I come to
understand the proof. However, my confidence in my understanding of the proof
could be shaky — although I think I have understood it, I may not be confident. In
such a circumstance, my belief in the theorem may be undermined if my
mathematically sophisticated friend then tells me that actually, that proof contained a
4
subtle mistake and that mathematicians now believe that the theorem is false. My trust
of my friend and in her authority as an expert may override my a priori knowledge, if I
am not confident in that knowledge. This further information — that my expert friend
tells me that the so-called theorem is false — is anything but a priori. This means that
my knowledge of the intermediate value theorem, if it could be undermined by the
claims of an expert, does not count as a priori in a stronger sense that requires
unrevisability.
Whether the independence needed for a priority requires infallibility and
unrevisability on the basis of any empirical evidence is a controversial issue (see, for
example Casullo 2006, Kitcher 2000, Field 2000).
There are many other current debates concerning the nature of a priority. To take them
in turn, we would do well to attend how thinking about the a priori has developed
from Kant to the present day. While there is no doubt that the a priori played an
important role in Ancient, medieval and early modern philosophy, the central
importance of the notion, and its currently use is indelibly shaped by the work of
Immanuel Kant.
The Synthetic A Priori
Before Kant’s Critique of Pure Reason, the three allied notions of necessity, analyticity
and a priority were not clearly distinguished. In Kant’s defense of the synthetic a
priori, the notions of the analytic and the a priori come apart. For Kant, an important
class of truths can be known a priori but not through analysis. They are synthetic a priori.
In our list of examples of a priori truths, the boundary between the analytic and the
synthetic occurs in the split between logical and semantic/conceptual truths (which are
analytic) and truths of arithmetic and geometry, which for Kant can be shown a priori
only by the operations of the intuitions of time and space and not by analysis. While a
considered discussion of Kant’s account of the a priori is beyond us, we should venture
into this territory just a little, for Kant’s approach to the a priori set the scene for crucial
developments into the 20th Century and beyond.
The distinction between the analytic a priori and the synthetic a priori is sharply drawn:
for Kant, the boundary is found at the limits of what is possible through the analysis of
concepts. Purely formal logic shows structural relationships between concepts (so
Aristotle’s syllogisms show formal and analytic conceptual relations between
judgments), and the analysis of concepts into constituent parts grounds another kind
of deduction between judgments. To say that Gilles is a bachelor is in part to say that
Gilles is unmarried, and so, the conditional judgment if Gilles is a bachelor, then Gilles is
unmarried, and the corresponding generalization all bachelors are unmarried, can be
shown to be true, a priori, by means of the analysis of concepts.
Nothing, according to Kant, can do the same thing with judgments such as simple
arithmetic claims like: 8 + 5 = 13, or geometric claims, such as the claim that the interior
angles of triangles add up to 180 degrees; let alone claims such as the Intermediate Value
Theorem. In each case here, the reasoner must engage in some rational deduction in
order to demonstrate the claims (and since the rational deduction is pure and not
5
empirical, it is a priori, well enough), but his kind of reasoning goes beyond what is
analytic. In the case of arithmetic reasoning, try as you like, you will not find the
concept of 8 in the concept of 13, or vice versa. You will pass through 8 on the way to
counting to 13, and you will hit 13 exactly after taking as many steps counting from 8
that one takes when counting to 5 — this is one way to show that 8 + 5 = 13, a priori.
But, for Kant, this is not an analytic deliverance of some formal meaning: this is the
kind of demonstration possible for one who has the concept of time — the concept of
one thing coming after another, which is crucial to our practice of counting.
The same goes for geometric reasoning, but here, the requirement is not the concept of
time, but the concept of space. We must use our spatial intuition in order to engage in a
priori spatial reasoning. It is not for nothing that presentations Euclid’s Elements are
filled with diagrams, and geometrical reasoning is filled with instructions to “extend a
line from point A through the intersection of lines l and m until it intersects line k” and
the like — they are instructions to engage in our spatial reasoning by way of our
spatial intuition. (Note here Kant does do not mean our hunches by the notion of
“intuition” but rather, our grasp of concepts and the way we structure them together,
pre-conceptually.) The findings of this synthetic a priori reasoning are as firm and as
necessary as any analytic truth, despite the fact that it may be (in some sense) a
contingent matter that we have the intuitions that we do. These intuitions of space and
time are, for Kant, pure, because we have them antecedently to the acquisition of
empirical concepts: we do not learn the pure intuitions of time and of space by
example or by experiment: rather, we use these intuitions to structure the empirical
intuitions we have in time and in space.
The ‘contingency,’ in some sense, of our having these temporal and spatial intuitions
instead of others does not lead to the contingency of the judgments involving them:
any more than the contingency of your having the concept of a leg before wicket and a
dismissal means that it is not necessary that any batsman out by way of a leg before
wicket has been dismissed by the bowler: what is contingent is your having the concept,
and your ability to express (or understand) that necessary truth. It would not be any
less necessary if we did not have the concepts needed to express it. Though it must be
noted that in the case of the socially mediated practice of cricket, if no-one has the
concept of lbw or dismissal or bowler, then it is plausible no-one can play cricket, either.
Were we to not possess the spatial and temporal intuitions that we have, then we could
not form these numerical or spatial judgments. Despite the internal and contingent
nature of the forms of pure intuition, for Kant the link between the necessary and the a
priori is fixed.
As we will see in the next section, philosophical orthodoxy has not remained Kantian
on the matter of the a priori. Nonetheless, Kant has able defenders on the identification
of necessity with a priority and the restriction of analyticity to a strict subset of the a
priori.*
*Hanna (2001, 2006) is a clear and lucid contemporary exponent of this view.
6
The A Priori and the Analytic
It is perhaps not surprising that with Kant’s key examples of synthetic a priori
knowledge coming from mathematics; it was from developments in mathematics that
revolutionary new ideas took their root. In the first instance, these ideas came from the
nascent discipline of the calculus and the theory of real numbers and functions. We
don’t have space to recount the intellectual trajectory in any detail, suffice to say that
by the 18th and 19th Centuries, since the pioneering work of Newton and Leibniz, the
mathematical practice of differentiating and integrating functions was well established
and largely well behaved, but — for both the pure mathematician and the philosopher
— it was anything but well understood. What is it for a function to be continuous?
Can we make the notion of “no jumps or gaps” precise, in a way that is amenable to
reasoning? In the work of Bolzano and Weierstrass, Cauchy and Dedekind, through
the 19th Century, advances were made on these fronts. Results such as the
intermediate value theorem were provided with proofs in the 19th Century. The
necessary ingredient wasn’t the ingenuity to fill in a missing step in a calculation or a
new technique for solving puzzles, but something much more fundamental, a new
definition. Bolzano and Weierstrass’s definition of what it was for a function to be
continuous * allows one to prove the intermediate value theorem using logic alone. As a
matter of logic, any continuous function in that sense of continuity we have defined, such
that f(–1) < 0 and f(1) > 0 has some point t between –1 and 1 where it crosses the line —
where f(t) = 0. The derivation is purely formal, on the basis of definition, without the
need for a diagram, picture or any requirement to see the conclusion. Advances in
mathematics gave the analytic the means to recover lost ground — and to go much
further.
It did not go unnoticed that to make sense of the derivation of results such as the
intermediate value theorem, we needed an expanded notion of logic in order to
accommodate reasoning with definitions such as these. The construction “for every …
there is a … such that whenever…” in the definition of continuity is a complex nesting of
what we now understand as quantifiers, and it took the development of logic in the 19th
and 20th Century in the work of Peano, Frege, Russell and Whitehead to give an
account of formal deductive consequence in a vocabulary so expressive. With the
expansion of the notion of a definition (to include the Bolzano–Weierstrass definition of
continuity) and the expansion of the notion of logic (to include what we now recognise
as classical first-order logic), the landscape of the territory between the a priori and the
analytic changed shape.
The development of the tools of logic beyond their Aristotelian bounds gave shape to
an important question: where exactly are the bounds of logic? What makes an item of
vocabulary a logical constant? This became a live issue with the work of Russell and
Whitehead, who appealed to an axiom of infinity in the type theory of Principia
Mathematica. They acknowledged that it was not satisfactory to conceive of this as a
logical notion in exactly the same way as other logical axioms. But why? What is the
* (f is continuous at x if and only if for every δ > 0 there is an ε > 0 such that whenever |x’ – x| < δ, |f(x)
– f(x’)| < ε)
7
ground for calling something a logical notion? Clearly a claim that there are infinitely
many things is not empirically verifiable in any straightforward fashion, so in some
sense it may be thought to be a priori if it is, in fact, true. But could a logical axiom be
the kind of thing we could debate? Contemporary discussions of the bounds of logical
concepts have not settled on a sharp delineation beyond broad agreement that what is
known as ‘classical first-order logic’ is a uncontroversially within the scope of logic
properly so called (see Etchemendy 1990, Read 1994, Shapiro 1991, Sher 1991).
Now, not only did the new mathematics and logic make possible the thought that the
truths of calculus could be true ‘by definition’ without the aid of pure intuition, but
advances in geometry and set theory paved the way to do the same thing for space and
time, Kant’s core notions of geometry and numbers.
Advances in thinking about Euclid’s fifth postulate, the parallel postulate, led to the
construction of formal models of geometric theories in which space behaved radically
differently (in the work of Lobachevsky, Reimann and others, and the question
naturally arose as to whether these non-Euclidean geometries did any better than
Euclid at characterising our pre-theoretic concept of space. Advances in theories of
sets, classes and types, in the work of Frege, Russell and Whitehead, and others, meant
that such fundamental notions of number were having their internal structure plumbed
and various analyses of the notions were proposed in order to more clearly articulate
the commitments in a theory of number, and to propose various definitions of that
notion.
At the height of logical positivism, analyses of notions such as space, time and number
had been proposed, refined and developed. Carnap’s Aufbau proposed a logical
framework for the definition of concepts, for relating them to sense experience, and for
analysing necessity and a priority as due to analyticity, which in turn was a purely
conventional matter. It is up to us which language we use, the adoption of one language
over another is an arbitrary or pragmatic matter, and relative to that choice of language,
a space of necessities and possibilities is defined, which can be analysed a priori using
the tools of logic. It is up to us that we define logical, numerical, geometrical concepts
in the way that we do, in just the same way that we define other terms of our language.
Relative to that definition, some truths are necessary and others are contingent, and
this matter is purely linguistic, relative to the choice of language employed. Necessity
and possibility; a priority and a posteriority as internal questions are to be answered
relative to a linguistic framework. The external question, of whether to employ this
language or that one is a pragmatic affair, to be answered not by asking which
language better reflects truth (for any question of truth is relative to the choice of a
language in which that truth is expressed), but rather, by other concerns such as ease of
use, better fit with practical or theoretical virtues or other aims of inquiry. The choice,
say, between the employment of a non-Euclidean or Euclidean geometry is not the
empirical question of which one is correct, in the absence of some prior choice of how
we are to identify what counts as a point and what counts as a line. Once we have
made that choice — and choice indeed it is, are points abstract, located in experience,
are lines the paths of uninterrupted light beams, or to be identified in some other
empirical fashion? — then the properties expressed in that language become a matter
8
of empirical or logical investigation, depending on the language chosen at the outset.
On this view, the only necessity is verbal necessity — necessity grounded in the
analytic, the choice of language. What is a priori and necessary is true by convention,
the conventional choice of a linguistic framework.
In this way, by the end of the first third of the 20th Century, the theoretical landscape
had inalterably changed. In the analytic tradition at least, Kant’s intuition was largely
banished in favour of logic, language and convention.
Gödel and Quine
Such an consensus did not survive. The decline of the logical positivists’ identification
of necessity and a priority with analyticity, and the split between necessary truth-by-
convention and contingent truth-on-the-basis-of-reality came in two fronts: one
another new mathematical result, and the other, a powerful philosophical metaphor.
The mathematical result is Gödel’s justly celebrated incompleteness theorem. Gödel’s
result dealt a deathblow to the naïve identification of necessary and a priori truth (in
some language) with what is analytically true (in that language). Gödel showed that in
the case of very many mathematical theories — in particular, any theory strong enough
to include a small proportion of modern mathematics — there are statements which are
true of that theory but are not provable within that theory, and this can be
straightforwardly proved a priori. This would not be a problem if it applied to some
theories, for we could say that even though theory T is incomplete, the theory we are
using to reason about T is the real theory, whose notion of necessity is to be identified
with analytic truth relative to it. But Gödel’s theorem will apply to this theory too, if it
is consistent. For any theory (if true), we can construct a stronger theory including
truths the first theory missed out. The only mathematically realistic theories to escape
Gödel’s incompleteness are theories that are inaccessible — theories whose axiomatic
basis is so complicated as to not be in principle enumerated. Such theories are not
candidates to be languages in Carnap’s sense, for they are not the sort of thing that can
be used as frameworks governing our use of mathematical vocabulary. Carnap’s
program is dealt a deathblow on mathematical grounds.
Not only was Carnap’s program dealt a blow by Gödel — these results provide grist
for the mill for anyone concerned with the epistemology of mathematics. Which
mathematical claims are knowable a priori? How are we to account for mathematical
truth and our access to that truth? Gödel’s results show that this branch of a priori
truth is a very delicate matter (Franzén 2004, Potter 2000, Smith 2008). One
contemporary proposal is to see that numerical vocabulary as introduced by abstraction
over a relation of equinumerosity. If the Fs are equinumerous with the Gs we say that
the number of Fs is the same as the number of Gs (Wright 1983). In this way it could be
(in some sense) a priori, without reducing to logic, and we do not need to identify the
truths of arithmetic with any formal theory, because what we take to be true
concerning numbers may well depend on our conceptual apparatus in other matters.
(What numbers we are able to countenance depends on what predicates we can
9
construct.) Matters here are, of course, subtle, both philosophically and
mathematically (Burgess 2005, Fine 2002).
Gödel’s results showed that the a priori is not to be identified with the analytic in some
language. Quine’s arguments against analyticity not only led analytic truth into
disrepute, but brought down the a priori with it. Quine’s compelling metaphor of the
web of belief, and the attack on the analytic/synthetic distinction apply not only to
analyticity but also to the a priori.
In a series of papers, including “Truth by Convention” (1936), “Two Dogmas of
Empiricism” (1951), culminating in a book Word and Object (1960), Quine led a
sustained attack on the logical positivism of his mentor, Carnap. In “Truth by
Convention” he argued that the Carnap’s distinction between conventional and
empirical truth is unsustainable, and along with it, there is no place for the distinction
between internal and external questions. In “Two Dogmas,” Quine argues that the
notion of synonymy required in any account of analytic truth is anything but
unrevisable or knowable a priori, and so cannot play the role required of it in the
logical positivist programme. Instead of privileging a class of statements as a priori and
immune from revision on internal grounds (to be changed only on external grounds on
the basis of a pragmatic choice for one vocabulary over another), Quine argues that the
network of beliefs is so interconnected that a difficulty in one area may be fixed by a
revision in another, whether that area is close to the periphery of perception
statements, some way in at the level of generalizations and lawlike statements, or deep
in the centre, at the level of mathematics and logic. All is of a piece, and the entire
edifice of commitments stands under judgment from the tribunal of experience
together. Only as a whole does a theory (an entire epistemic standpoint) serve to be
confirmed or disconfirmed by evidence, and any part of it can be revised to better fit
that evidence. The epistemology is explicitly holist, and a posteriori. Putatively a priori
claims seem independent from experience because they are general, applying
regardless of the experience we receive from the world and therefore telling us nothing
about the world of experience, but nonetheless, perhaps they are revisable in the light
of other statements in just the same way as any other claim.
Such a wide-ranging attack on the a priori was unprecedented: in Quine’s vision of
philosophy, none of the notions of a priority, analyticity nor necessity play a central
role. Philosophy is continuous with the empirical sciences: it differs only in generality.
Kripke and Kaplan
The a priori, analytic and the necessary were not without friends in the second half of
the 20th Century, despite Quine’s attack. Among defenders of the of the notion, closer
attention was paid to the relationship between the analytic and the necessary, and this
shed new light on possibilities for theories of the a priori. Two great insights came from
Kripke and from Kaplan.
From Kripke (1972) we learned an example of what may be necessary but not a priori.
The famous examples are all identity statements. Take the claim that Hesperus is
Phosphorous, where Hesperus is a name for the morning star, and Phosphorous is a
10
name for the evening star. It is plausible that Hesperus is Phosphorous, and that this
truth is not merely contingent. That planet (said, pointing to Hesperus) is the very
same thing as that planet (pointing to Phosphorous). There is no way that they could
differ, for there is no ‘they,’ it is merely an ‘it’ pointed to twice. So, says Kripke, we
should conclude that the claim that Hesperus is Phosphorous is necessary. But can we
know that Hesperus is Phosphorous a priori? This seems to not be the case: As a matter
of contingent fact, we learned our names in a context in which the Morning and
Evening Stars were the same planet. Had things been different, the last body seen in
the morning and the first seen in the evening, could have well been different celestial
bodies. In that case, the sentence “Hesperus is Phosphorous” would not only fail to be
a priori, it would fail to be true. The only way we can reassure ourselves that we are in
our circumstance rather than that circumstance is to to engage in astronomical
observations and theorising. We learn that Hesperus is Phosphorous only a posteriori.
From Kaplan (1989) we learned the opposite lesson. Not only are some necessities not
a priori, but some contingent things can be known a priori in reverse, not all a priori
knowable truths are necessary. I can know, it seems, that I am here now. This claim is
true, and in some sense, analytically true. However, it is contingent. Had things been
different, I wouldn’t have been here, I would have been elsewhere. The logic of
indexicals, (and demonstratives, such as this or that) has subtle connections with
necessity. The presence of contextual features to be filled in at the circumstances of
utterance mean that we can exploit our understanding of these features (or the lack
thereof) to generate interesting cases of knowledge, independent of our external
evidence.
There are many current issues in these areas: the logic of necessity and names is
controversial, and so is the issue of how to extend Kripke’s ideas from names to
natural kind terms such as water and H2O, as we shall see in the next section. Similarly,
when it comes to indexicals and demonstratives, How far are we to go in contextually
settled parameters such as that of location, speaker and time of utterance? Does this
example mean that we can know statements to be true a priori, without knowing what
propositions those statements express? What exactly is the item of knowledge claims
in these circumstances? What notions of analyticity are in play in these examples? Is
the discussion of analyticity merely ignoring Quinean objections, or do advances in
linguistics and semantics such as those found in Kaplan, Montague and others give us
the tools to defuse Quine’s objections? (See Russell 2008).
Externalism and Skepticism
Kripke’s example has shown us that the behaviour of concepts we have may depend
on external factors beyond our immediate grasp, and the ability for us to acquire a
concept may depend, in a straightforward manner on external factors. If this is the
case, then we seem to have the following puzzle. Consider this argument (Boghossian
1997):
11
1. If I have the concept water, then water exists.
2. I have the concept water.
Therefore,
3. Water exists.
We seem to know the first premise a priori, on general semantic grounds. If I have the
concept water, this could only have been acquired in an environment in which
someone has been in the presence of water. On the other hand, it is true that we do
have the concept water. (How could we have formulated this argument if we didn’t
possess the concept?) It seems to inescapably follow that water exists. But if the
premises are known a priori, then it seems that the conclusion is known a priori too, for
it is the conclusion of a valid argument. How can this be? Can we truly know that
there is water a priori?
This is a stripped down version of an argument raised by Putnam’s famous Brain in a
Vat thought experiment. It seems that we have proved that skepticism is impossible,
given the externalist nature of or concepts.
This seems like too much of a success for a priori knowledge: some have thought to
resist the argument at the conclusion: to say that warrant is not always transmitted
from premises to conclusion of an argument, even when those premises are a priori
known (see, e.g., Wright 1991). Others have sought to clarify the premises: while
premise 1. is true, if read in the form it appears to take, it cannot be known a priori.
What can be known a priori is that
1*. If I have the concept water, then whatever plays the water role exists.
That may not be water, were we to be systematically deceived by a demon, or to be
running in a computer simulation. From this premise, all we are entitled to derive is
the weaker conclusion 3*, that whatever plays the water role exists, and this is much
less surprising as an example of a priori knowledge.
A Priori Truth from above & from below
I’ll end with a short sketch of two constructive pictures of the nature of a priori truth.
The first is the framework in which the above disambiguation can be made out: it is the
Two-Dimensionalist account of necessity, analyticity and the a priori. Originating in
work of Humberstone and Davies (1980) analysing the distinction between necessity
and analyticity in the presence of operators for actuality, the two-dimensional
framework has been adopted for a wide-ranging perspective on the connection
between necessity — conceived of as truth in all possible worlds (where a possible
world is how things could have been had things gone differently) and a priori knowability
as true in all epistemic scenarios (where an epistemic scenario is a way that things as
they are could be taken to be). In this case, there are epistemic scenarios in which
Hesperus is not Phosphorous, and epistemic scenarios in which water is not H2O, but
some other material plays the water role. However, water (the substance that plays the
watery role) is of necessity H2O, so water is H2O in every possible world.
12
Understanding what might play the role of an epistemic scenario is not
straightforward (Chalmers 2004), but this broad approach to analyticity, necessity and
the a priori has motivated a great deal of contemporary philosophy (e.g. Jackson 2000).
This approach to grounding the a priori in truth-in-all-epistemic-alternatives is a ‘top-
down’ all at once account of the a priori. It is hostage to giving an account of what is
true in an epistemic alternative, and it gives little insight into how that might be found
in any concrete case. The main rival to this sort of account of a priority is to be found in
the inferentialist traditions, exemplified currently in the work of Brandom (1994) and
Peacocke (2004). In this tradition, we return to the connection between the a priori and
reasoning: if fundamental inferences involving concepts are to be a priori, then perhaps
we can use these inferential proprieties to give an account of what it is to possess that
concept — indeed, what it is to be that concept. To possess the concept of conjunction
is, in part, to be disposed to use the inference rules of conjunction elimination (from (p
and q) to infer p) and conjunction introduction (from both p and q to infer their
conjunction (p and q)). To possess a concept such as the colour concept green is to place
it in a network of inferences with other colour concepts, as well, perhaps to place it
inferentially in a network of input and output rules governing circumstances where it is
permissible to introduce the concept or to exploit it.
Such accounts of the a priori have the great advantage of paying attention to the fine
detail of each concept under discussion, and to play an important role in grounding the
propriety of inferences employing these concepts. However, they have he great
disadvantage of requiring a great deal of attention to every single case. Furthermore, it
is an open question of whether a non-circular reason can be given for why some
concepts may be introduced by definition or stipulation and others may rejected as
incoherent or defective. If concepts are individuated by their inferential roles, then
why can we not justify any questionable inference by the adoption of a concept which
just happens to take that inference as one of its defining conditions? (Williamson 2003).
Perhaps some combination of a top-down two-dimensional approach and bottom-up
inferentialist one will provide resources to respond to these concerns.
As we have seen, since Kant the notion of the a priori has waxed and waned in its
philosophical fortunes: at some times it has been the centrepiece of philosophical
concerns, and at others, it has been peripheral. Just as the relationships between
metaphysics, semantics and epistemology vary between philosophical positions and
fashions, we may also draw the connections between the notions of necessity,
analyticity and the a priori (or truth in all possible worlds, truth independent of
experience, and truth determined only by meaning) in correspondingly different ways.
The different ways we treat the a priori reflect broader concerns and larger themes in
philosophy, and bring to light our views of the aims and methods of philosophical
inquiry.
Reading
There is much to read on current work on the a priori. A useful complement to this
article is Carrie Jenkins’ survey article (2008). It is especially good on the twists and
13
turns of contemporary efforts to define the notion of the a priori, and the debate over
whether what is a priori is unrevisable.
For another, longer general treatment of the current debate, Albert Casullo’s article in
Moser’s Oxford Handbook of Epistemology is very clear, as is his 2006 book. For an
excellent collection of core readings, Moser’s A Priori Knowledge is invaluable.
For an entertaining and illuminating account of the a priori in Kant, its attempted
capture in the work of the logical positivists, and the eventual disintegration of that
program, you must read Alberto Coffa’s The Semantic Tradition from Kant to Carnap.
The best account of Quine’s criticisms of analyticity and (implicitly) a priority are still
Quine himself: both “Two Dogmas” and Word and Object are models of clarity. For
Gödel’s theorems, Smith’s new Introduction is second-to-none.
The best overview of the mechanics of the two-dimensional account of the a priori is to
be found in Chalmers’ long paper “Epistemic Two-Dimensional Semantics,” but a little
book that shows how the general approach can be applied to a vast range of
philosophical issues is Jacksons From Metaphysics to Ethics.
For an introduction to inferentialism, the first port of call must be Brandom’s
Articulating Reasons, but the interested reader must move beyond this short
introduction, either to the big book Making it Explicit, or to Peacocke’s very different
The Realm of Reason.
Paul Boghossian (1997) “What the Externalist Can Know A Priori”. Proceedings of the Aristotelian Society,
97(2):161–175.
Robert B. Brandom (1994) Making It Explicit. Harvard University Press.
Robert B. Brandom (2000) Articulating Reasons: an introduction to inferentialism. Harvard University Press.
John P. Burgess (2005), Fixing Frege, Princeton University Press, Princeton.
Albert Casullo (2006) A Priori Justification, Oxford University Press, Oxford.
David Chalmers (2004) “Epistemic Two-Dimensional Semantics”. Philosophical Studies, 118(1):153–226.
J. Alberto Coffa (1993) The Semantic Tradition from Kant to Carnap. Cambridge University Press. Edited by
Linda Wessels.
Martin Davies and Lloyd Humberstone (1980) “Two notions of necessity”. Philosophical Studies, 38(1):1–
30, 1980.
John Etchemendy (1990) The Concept of Logical Consequence, Harvard University Press, Cambridge,
Massachussetts.
Hartry Field (2000), “A Priority as an Evaluative Notion,” pp. 117–149 in New Essays on the A Priori,
edited by Paul Boghossian and Christopher Peacocke, Oxford University Press, Oxford.
Kit Fine (2002) The Limits of Abstraction, Oxford University Press, Oxford.
Torkel Franzén (2004) Inexhaustibility: a non-exhaustive treatment, Volume 16 of Lecture Notes in Logic,
Association for Symbolic Logic.
Robert Hanna (2001) Kant and the Foundations of Analytic Philosophy, Oxford University Press, Oxford.
Robert Hanna (2006) Rationality and Logic, MIT Press, Cambridge, Massachusetts.
Frank Jackson (2000) From Metaphysics to Ethics: A Defense of Conceptual Analysis, Oxford University
Press, Oxford.
Carrie Jenkins (2008) “A Priori Knowledge: Debates and Developments” Philosophy Compass 3(3), 436–
450.
14
David Kaplan (1989) “Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and
Epistemology of Demonstratives and Other Indexicals” pp. 481–614 in Themes from Kaplan, edited
by Joseph Almog, John Perry and Howard Wettstein, Oxford University Press, New York.
Philip Kitcher (2000) “A Priori Knowledge Revisited,” pp. 65–91 in New Essays on the A Priori, edited by
Paul Boghossian and Christopher Peacocke, Oxford University Press, Oxford.
Saul A. Kripke (1972) Naming and Necessity. Harvard University Press, Cambridge, Massachussetts.
Paul K. Moser, editor. (1987) A Priori Knowledge, Oxford Readings in Philosophy, Oxford University
Press, Oxford.
Paul K. Moser, editor. (2002) Oxford Handbook of Epistemology, Oxford University Press, Oxford.
Christopher Peacocke (2004) The Realm of Reason, Oxford University Press, Oxford.
Michael D. Potter (2000) Reasons Nearest Kin: philosophies of arithmetic from Kant to Carnap, Oxford
University Press, Oxford.
W. V. O. Quine (1936) “Truth by Convention” pages 90–124 in Philosophical Essays for Alfred North
Whitehead, Longmans, Green and Co., New York. Reprinted in The Ways of Paradox, and Other
Essays.
W. V. O. Quine (1951) “Two Dogmas of Empiricism”. The Philosophical Review, 60(1):20–43.
W. V. O. Quine (1960) Word and Object. MIT Press.
Georges Rey. “A Naturalistic A Priori”. Philosophical Studies, 92(1):25–43, 1998.
Gillian Russell (2008) Truth in Virtue of Meaning, Oxford University Press, Oxford.
Stewart Shapiro (1991), Foundations without Foundationalism: a case for second-order logic, Oxford
University Press, Oxford.
Gila Sher (1991), The Bounds of Logic, MIT Press, Massachussetts.
Peter Smith (2008), An Introduction to Gödel’s Theorems, Cambridge University Press, Cambridge.
Timothy Williamson (2003) “Blind Reasoning”. Aristotelian Society Supplementary Volume, 77(1):249–293.
Crispin Wright (1991) “Scepticism and Dreaming: Imploding the Demon.” Mind 100(1):87–116.
