Professional Documents
Culture Documents
Gallois 1996
Gallois 1996
Why call the knowledge mentioned in (1) a priori? 2 Why suppose that
one can have a priori knowledge of what one thinks? In the current
literature 'a priori' is used in at least two different senses. In the first
sense knowledge is a priori only if it is invulnerable to counterevidence.
In the second, knowledge is a priori only if it does not depend on
observation. There is little plausibility to the claim that, in the first sense,
we can have a priori knowledge of any of our thoughts. In contrast, the
claim that, in the second sense, we can, and do, have a priori knowledge
of our thoughts is very plausible. Your knowledge that you think you
are reading a paper is not, in any obvious way, based on observation.
The next premisses is:
(2) ETC --+ (~(3x) (x knows a priori that ETC)
it follows that:
(4) 0(3x) (x knows a priori ETC).
Another premisses that is required is this. If (4) and (6) are both true,
there could be someone who combines the a priori knowledge men-
tioned in (4) with the a priori knowledge mentioned in (6). That is:
(7) [(4) & (6)] ~ 0(3x) [x knows a priori that (ETC & (5))].
There seems to be no reason to dispute (7). Suppose (4) and (6) are
both true. What could preclude the possibility of someone combining
the a priori knowledge mentioned in (4) with the a priori knowledge
mentioned in (6)?
From (4), (6) and (7) we can derive:
(8) 0(3x) [x knows a priori that (ETC & (5))].
However, so construed, (5) does not exclude the possibility that you are
a brain in a vat presently inhabiting a world devoid of water. If it is to
tell against that possibility (5) must be construed as: (x) (t) (at t: x thinks
that x lives in a watery world --+ at t: x lives in a watery world). The
problem is that, so construed, (5) is false.
This objection to incorporating (5) as a premisses in the main argu-
ment has little force. It would be a striking enough result if the main
argument shows that someone could know a priori that there is some
water around at some time.
A better objection to (5) is advanced by Anthony Brueckner in his
reply to Michael McKinsey's version of the main argument. 5 Brueckner
points out that Tyler Burge, a prominent defender of ETC, does not
accept (5). The reason is this. One can think about a non-existent natural
kind by indulging in some sophisticated theorizing. For example, in a
world where hydrogen and oxygen exist, but do not bond together, one
could conjecture about the stuff that would be formed by two hydrogen
atoms bonding with one oxygen. If one conjectured that such stuff would
behave like water than one would arguably have the concept of water,
and a capacity to entertain thoughts about water. 6
In order to accommodate this point (5) should be replaced with:
(5') ETC --+ tO(x) (x thinks x lives in a watery world ~ (no one
has theorized about water --+ x lives in a watery world))
(7) with:
The revised main argument, the argument from (1)-(8'), gives no cre-
dence to the possibility of someone having a priori knowledge of the
existence of water. In order for it to do so, one would have to be able
to know a priori that no one has theorized about water. Nevertheless, if
one can know a priori that (8 ~) is true then one could be in a position
that it is very counterintuitive to suppose that anyone could be in. One
could combine one's a priori knowledge of (8 t) with one's a posteriori
knowledge that no one has theorized about water to deduce that water
exists.
(8 ~) is perhaps slightly less counterintuitive than (8). However, (8 ~)
is sufficiently counterintuitive to be rejected. If we reject (8 ~) then we
have to reject one of the revised main argument's premisses. Suppose
we are not going to reject (2). In addition, we wish to retain (5 ~) and
(61). In that case, we have to choose between (1) and (3). Once again
we have to choose between restricting privileged access, or abandoning
ETC.
II
Provisionally grant that (5) is true. What of (6)? (6) is initially plausible
since (5) seems to just articulate an entailment of a philosophical thesis.
Despite this, we should not concede (6) even if we are prepared to
concede (5).
Consider the following quick argument against ETC. No reputable
scientist believes that phlogiston exists. However, the expression 'phlo-
giston' seems to function in the same way as 'heat' or 'water'. For
example, if, to our great surprise, phlogiston turns out to exist, we could
make a scientific discovery about its real essence. Hence ETC implies:
(10) E](x) (x thinks that phlogiston was believed in by eighteenth
century chemists -+ (no one has theorized about phlogiston
---+ phlogiston exists))
You know a priori that you think that phlogiston was believed in by
eighteenth century chemists. So, if you could know a priori that ETC
is true, you could know a priori that if no one has theorized about
EXTERNALISM AND SCEPTICISM
III
Call the view that the semantic properties of a term like 'water' depend
on whether water exists as a natural kind the semantic variation thesis.
In the present context the semantic variation thesis has a drawback.
What we are after is a refutation of the main argument which does not
call into question ETC, or our privileged access to the propositional
attitudes encompassed by ETC. We are after a refutation of the main
argument that does not tell against its first premisses. Arguably the
semantic variation thesis makes premisses (1) dubious. Here is why.
Consider the following sentence:
(11) 'There is some water somewhere.'
According to the semantic variation thesis, if we live in a waterless world
(11) expresses a thought which is different from the thought it expresses
if we live in a watery world. When you entertain the thought expressed
by (11) do you know a priori which thought you are entertaining? Not
if the semantic variation thesis is true. It follows from the semantic
variation thesis that which thought is expressed by (11) depends on
whether water exists. The existence of water can only be established
a posteriori. Hence, the semantic variation thesis implies that you can
only establish a posteriori which thought you are entertaining when you
entertain the thought expressed by (11).
Does this argument show that the semantic variation thesis conflicts
with premisses (1) of the main argument? The matter is a delicate one to
resolve. Donald Davidson and Tyler Burge have argued that a minimal
type of privileged access is compatible with (ETC). 7 Their basic idea
is this. Suppose the content of my belief that water is wet is externally
determined. Consider the following pair of beliefs:
(i) The belief that water is wet.
(ii) The belief that one believes that water is wet.
Can one hold (ii) falsely for the following reason? One believes one
holds a belief expressed by 'Water is wet' with the externally determined
content that p. In fact one holds the belief expressed by 'Water is wet'
with the quite different externally determined content that q. Davidson
EXTERNALISM AND SCEPTICISM
and Burge hold that this is not possible. Whatever externally determines
the content of (i) will also externally determine the content of (ii). So, the
content of (i) is bound to match the content of (ii). When one attributes
to oneself a certain belief, one is bound to be attributing to oneself a
belief with the right content.
Let us apply the Davidson Burge point to the semantic variation
thesis. According to the semantic variation thesis what content (i) has
depends on whether water exists as a natural kind. Let us say that if
water exists as a natural kind then (i) has natural kind content. On
the other hand, if water does not exist as a natural kind then (i) has
descriptive content since 'water' would then be synonymous with a non-
rigid description. The point is this. If (i) has natural kind content then
(ii) will have a corresponding natural kind content. If (i) has descriptive
content then (ii) will, again, have a corresponding descriptive content.
Even if the semantic variation thesis is true, there is no way that (ii)
could attribute a belief with the wrong content. In this sense, Davidson
and Burge allege, (1) is true.
Is the Davidson Burge minimalist conception of privileged access
adequate? According to that conception it is enough to secure privileged
access that a belief such as (i) cannot target on a belief with the wrong
content. The semantic variation thesis was invoked to show that pre-
misses (6) of the main argument is false. In turn, the Davidson Burge
conception of privileged access was invoked to show that the semantic
variation thesis is consistent with premisses (1) of the main argument.
However, if Davidson and Burge are right about privileged access there
may be a way of showing that the main argument is sound.
There is a semantic treatment of singular terms, associated with, for
example, John McDowell, which we have not yet addressed. 8 According
to that treatment, singular terms that fail to refer are contentless. Hence,
sentences in which they figure are contentless, and express no mental
states. On this view, if 'phlogiston' does not denote an existing natural
kind then there are no thoughts about phlogiston. We are, somehow,
under the illusion that we sometimes entertain such thoughts. Call this
the McDowell view.
How does (1) of the main argument fare if the McDowell view is
correct? If it is, and the Davidson Burge view of privileged access is also
10 ANDRI~ GALLOIS AND JOHN O'LEARY-HAWTHORNE
correct, then (1) is true. On the McDowell view if water does not exist
as a natural kind, I cannot falsely think that there is water. However, if
there is no water, on that view, I also cannot falsely think that I think
there is.
Now let us consider (5) and (6), the other crucial premisses of the
main argument. The McDowell view appears to underwrite (5). Accord-
ing to that view, if water never exists, the sentence expressing the
consequent of (5) is contentless. Moreover, the consequent of (5) can
only be false if water never exists. So, the McDowell view rules out
the antecedent of (5) being true, and its consequent false. Hence, if
McDowell view can be known to be true a priori, (5) can be known to
be true a priori. That is, (6) is true.
The McDowell view supports (1) of the main argument if Davidson
and Burge provide a sufficient condition for privileged access to, at least,
content. Do they? If they do then the following conditions are sufficient
for knowing something a priori:
Are (a) and (b) jointly sufficient for knowing something a priori? It
would seem not. Suppose that identities are necessary. In addition, sup-
pose that, though no one has any reason to think so, beliefs are type
identical with neural states. For example, the belief that grass is green,
is type identical with neural state N1, the belief that snow is white with
neural state N2, and so on. It happens that the belief that one is in neural
state Nn is type identical with neural state Nn. For no reason at all, Jones
believes she is in neural state Nn. So, Jones' belief satisfies (a) and (b).
Nevertheless, it seems clear that Jones does not know she is in Nn.
Something extra needs to be added to (a) and (b) in order to obtain
a priori knowledge. If the extra that is needed is added, the question
again arises whether the semantic variation view can be reconciled with
(1). Fortunately, we need not pursue it here, There is an alternative
way of refuting the main argument which does not impugn its validity,
ETC, or (1). What we will argue is the following. ETC is supported by
more famous thought experiments of which, perhaps, the best known is
EXTERNALISM AND SCEPTICISM 11
It does not tell us that Albertine fails to entertain the thought that there
is water in a world free of both water and water theorizers.
Comparing the following situations with help to clarify the point.
In the first Albertine sees a body of water, and decides to use 'water'
as a synonym for the non-rigid definite description 'the stuff that is
odourless, colourless, tasteless ect.' In this first situation Albertine has
no doubt that the antecedent of (12) is satisfied. She asks herself the
following question. Could I have entertained the thought that I live in
a watery world if water had never existed? Clearly, she should answer
yes,
In the second situation Albertine uses 'water' so that (12) is true. She
believes that she acquired the concept of water from interacting with
water. However, she takes seriously the possibility that she has been
permanently envatted without having any contact with water. She asks
herself the following question. If I am right, and I acquired my concept
of water from interacting with it, could I have entertained in a waterless
world the thought that there is water? Since the antecedent of (12) is
satisfied she should answer no.
12 ANDRt~ GALLOIS AND JOHN O'LEARY-HAWTHORNE
sidered. Albertine thinks that there is water. What does the proposition
she is entertaining imply? Does it imply that there is something which
is necessarily water? Does it imply that there is some natural kind?
Does it imply that there is a coloufless, odoufless liquid distributed over
the Earth's surface? If externalism and the semantic variation thesis are
both true, Albertine cannot say. She does not know whether 'Water'
has to be synonymous with a non-rigid definite description in order for
'Water exists' to express the proposition she is entertaining. On the other
hand if the present view is correct, Albertine can know which of the
implications in question hold without having to know whether there is
any water. To that extent it is less open to dispute that Albertine knows
which proposition she is entertaining when she thinks that there is water.
Before proceeding a further observation needs to be made. There is
a problem with replacing (5) with (5*). Phlogiston does not exist, but
it might have. In the actual world Wa, without anyone theorizing about
it, Albertine entertains a thought about non-existent phlogiston. Let W l
be a world in which Albertine derives her concept of phlogiston from
interacting with it. Surely, it is true in Wa that there is such a world as
W1. On the other hand, it is not true in Wl that there is any such world
as Wa. Wl is accessible toWa, butWa is not accessible to W1. It seems
that the objection we have given to (6) commits us to denying that the
accessibility relation is symmetrical. Hence we seem to be committed
to no stronger modal logic than $4.
Our reply to this objection is suggested by some comments of Saul
Kripke's. 9 The objection depends on the assumption that if phlogiston
does not exist, it could h a v e - Kripke maintains that if fictional creatures
such as unicorns do not exist, they could not exist. What goes for a
unicorn goes for phlogiston. If the fictional substance phlogiston does
not exist, it could not exist.
IV
We have noted that (4) is fairly trivial and hence (5) is very plausible.
The argument thus enjoys advantages over the analogous argument in
the earlier part of our paper. We shall grant (1) though not without
reservations. Sometimes the principle of charity is deployed in a way
that seems to ensure only that we have a large variety of general beliefs
that are true. This seems to be what is implied by the Davidsonian
point that an error about the environment - say, that stars and holes
in the sky - can only be made sense of against the background of a
bunch of true general beliefs (in this case, about what holes are, what
the sky is, and so on). Understood this way, the principle offers no
potential rebuttal of the brain in a vat hypothesis. It merely entails that
even if we are brains in vats (or deceived by a malevolent demon),
we still have a bunch of true general beliefs. (Perhaps one could make
more headway here by combining a charity principle with the Kripkean
insights about reference mentioned earlier, though we won't pursue the
matter here.) In what follows, we shall assume, as Davidson seems to,
that the principle of charity is incompatible with the falsity of our belief
in physical objects. We urge the reader to remember that it is not clear
to us that this is so. We shall also grant Davidson the modest principles
of epistemic closure and so on whose necessary truth is required if the
above argument is to be valid.
16 ANDR]~ GALLOIS AND JOHN O'LEARY-HAWTHORNE
consideration. Let us grant that the best interpretation that we can find
given only behavioral data will always be a charitable one. Still it might
be that the meaning that the speaker really assigns and the beliefs she
really has, which we would know if we knew her inner mental life, does
not correspond to the one that is most reasonable given the behavioral
data alone. The primary challenge, then, (though not the onl), challenge)
when it comes to defending a constitutive status for charity is to close
the gap between what a speaker really means and believes and what the
behavioral evidence suggests. Davidson has always tried to remove that
gap by appealing to the social character of language:
As Ludwig Wittgenstein, not to mention Dewey, G. H. Mead, Quine and others have
insisted, language is intrinsically social. This does not entail that truth and meaning can
be defined in terms of observable behavior, or that it is nothing but observable behavior;,
but it does imply that meaning is entirely determined by observable behavior, even
readily observable behavior. That meanings are decipherable is not a matter of luck;
public availability is a constitutive aspect of language.12
his thinking is that such states don't form part of our ordinary evidential
basis for interpretation and so cannot constitute meaning even if they
are had.
Consider now the following rather fantastic scenario (which while
bizarre is no more bizarre than the sorts of scenarios described in ordi-
nary discussions of scepticism). People have a language of thought
which causally produces utterances. When someone speaks, a demon
looks at that language of thought, determines the biological functions
of the tokens in the language and whispers in the bearer's ear, so to
speak, providing information about the mental representations of the
speaker that causally produced the utterance. Such information will
include details about the structure of the internal language and infor-
mation about the biological function of bits of the internal language.
(Note that we are not assuming that the demon or anyone else has a
direct view of "the meanings" of people's inner states.) Hearers then
describe the overt utterances in ways that are responsive to the informa-
tion they have received. Of course, the hearers are not consciously aware
of the process whereby the demon provides them with such input, rather
in the way that we are not aware of the information processing that
delivers grammatical judgment as output. Nevertheless, the hearers
enjoy subconscious information processing mechanisms which make
use of input about speakers' inner properties as part of the data that
yields interpretative outputs. So we might imagine, for example, that
a hearer subconsciously takes in information to the effect that it is the
biological function of a certain type of syntactic item to be tokened
in the presence of F's and in processing that biological information,
construes it as, at the very least, evidence that the item is about F's. ~3
(We can similarly imagine possible worlds where people have access
to properties that are not publicly observable by a telepathic faculty
and that such information serves as input into information processing
concerning interpretation. What we have to say about the demon worlds
applies mutatis mutandis to telepathy worlds.)
Let us imagine a possible world where the scenario takes place.
What should Davidson say about it? Let us suppose the inhabitants of
this possible community use a language that is phonetically similar to
English to express themselves. It may well be that were we to translate
EXTERNALISM AND SCEPTICISM 19
NOTES
1 One who adopts this view is Anthony Brueckner in "Brains in a Vat", The Journal of
Philosophy, 83 (1986): 148-67.
z In calling it a priori we are following Michael McKinsey. The main argument is
essentially an elaboration of the argument given by McKinsey in "Anti-Individualism
and Privileged Access",Analysis, 51, No. 1 (January 1991): 9-16.
3 For a critical discussion of this epistemic closure principle see Robert Nozick Philo-
sophicalExplanations (Oxford 1981) chapter three.
4 Should we be so complacent about ETC? There is a bad idea that tends to be confused
with good versions of ETC. It is that a term picks out the cause of its utterance, no
matter how unnatural or gerrymandered the cause is. This thesis would no doubt help
with external world scepticism, but it is hopeless. On that view, it is so easy for terms
to refer that we would have a virtual guarantee that, say, 'water' succeeds in referring.
Apart from the problem that there are many causes of an utterance of any word, this
view has the unpalatable consequence that phlogiston exists. Moral: the constraints on
reference can't be so generous that all our natural kind terms refer come what may. It
is only those ultra generous constraints on reference that make it so obvious that the
natural kind terms of a brain in a vat refer to something or other. Abandoningsuch ultra
generous constraints does not vitiate the good externalist point that there may well be
propositions that we entertain that a brain in a vat duplicate cannot.
5 AnthonyBrueckner"WhatanIndividualistKnowsAPriori, Analysis,52,No. 2(April
1992): 111-118. Brueckner is replying to Mckinsey's argument in Mckinsey ibid.
6 In this case theorizing about something non-existentproceeds by envisaging the unre-
alized product of combining existing things. The case is similar to the following. One
succeeds in talking and thinking about someone who does not exist by referring to her as
the product of a separately existing sperm and egg. Perhaps one need not tie down one's
thought and talk about the non-existentto what already existed in this way. Perhaps one
could entertain water thoughts in a world where matter is non-atomic. Someone in such
a world could speculate about matter being atomic, and go on to speculate about how it
would behave if certain hypothetical atoms were combined in certain ways.
7 For Davidson and Burge's views on these matters see Donald Davidson "First Per-
son Authority", Dialectica, 38, Nos, 2-3 (1984): 101-111, and "Knowing One's Own
Mind", Proceedings of the American Philosophical Association (1986): 441-458: and
Tyler Burge "Individuation and Self Knowledge", The Journal of Philosophy, 85, 1
(November 1988): 649-663.
8 See John McDowell "Singular Thought and The Extent of Inner Space" in Subject,
Thought, and Context ed. John McDowell and Phillip Pettit (Oxford) 1986: 137-168.
EXTERNALISM AND SCEPTICISM 25
connect with theories of meaning may turn out to be a profound one; we should merely
be suspicious of any claim of epistemic priority for the latter over the former. The idea
of transcendental argument that thought, inquiry and communication have fundamental
preconditions may not be wrong, what is wrong it the idea that those preconditions can
be known a priori. A defense of these more general theses is beyond the scope of this
paper.
16 See "Reply to Chomsky" in Davidson and Hintikka, eds. Words and Objections:
Essays on the Work ofW. V. Quine (Reidel: Dordrecht, 1969), p. 306. We are indebted
to Jose Benardete for drawing this remark of Quine's to our attention. See his "AI and
the Synthetic A Priori", in Philosophy in Mind, eds. O'Leary-Hawthorne and Michael
(Kluwer Academic Publishers, 1994).
L7 "Knowing One's Own Mind", Proceedings and Addresses of the American Philo-
sophicalAssociation, 60 (1987).
18 It also cuts against using Swampman considerations to provide an a priori refutation
of the sceptical hypothesis that the world was created two seconds ago.
19 We are grateful to audiences at the University of Virginia, University of New South
Wales, Purdue University and the Australian National University for their helpful com-
ments on an earlier version of this paper.