The Structure of Knowledge

The Structure of Knowledge

In responding to the charge of circularity, we should note first that the argument
defending inference to the best explanation does not beg the question against the
sceptic in the way that it would if one of its premises simply stated that the principle
of inference is sound or truth preserving. Any principle, indeed any conclusion,
could be ‘defended’ in this way, since any premise implies itself. Here the principle
of inference was used, not mentioned, in its own defence. Such use does not trivially
result in self-support, as the sceptic might suggest. We do not normally dream, for
example, that dreaming is a reliable source of knowledge or hallucinate that
hallucination is reliable. In the case of inference to the best explanation, it is not a
foregone conclusion that it will be self-supporting, and the argument from evolution
is controversial even if sound. But the principle, when applied with critical care,
does iterate, in that it is more explanatorily coherent to believe that it leads to truth.
It is doubtful whether there are any obviously unsound principles that iterate it this
way or cohere with other fundamental sources of knowledge.

This coherence with other principles or sources is also significant even if circular.
Crystal balls might predict their own reliability, but their predictions fail to be
regularly confirmed by perceptions. This kind of mutual support adds significantly
to the self-support of the principle, since it strongly suggests a common cause in
truth. Why should percepts confirm earlier inferences if they do not reflect the same
facts believed? The more seemingly independent facts that an explanation unifies,
the more reason there is for accepting the explanation. Of course, a non-sceptical
answer to the question just raised simply applies an inference to the best explanation
again. But the near ubiquitous nature of the principle does not more firmly support
a sceptical attitude toward it, although it does make more clear that any defence of
it will have to be circular. While we have not relied heavily on necessary truths, it
is worth pointing out here that it is a necessary truth that basic or fundamental
epistemic principles can be supported only by themselves or their products and other
fundamental principles (otherwise they would not be fundamental). We cannot
escape the totality of our cognitive resources to verify that they lead to truth. Such
confirmation can come only from within the circle of these resources. In the case
of inference to the best explanation, which is involved in the legitimisation of perception, memory, and simple induction, and in turn supported by them, the circle
is certainly not narrow. Can we demand more of the epistemologist than the
demonstration of such broad coherence in the set of our epistemic beliefs?

Circular reasoning is certainly not always vicious. We assess the skill of a tennis
player, for example, by noting his execution of many skilful shots. But we judge a
shot to be skilful and not merely lucky because it is made by a skilful player. The
same shot made by a beginner would be just lucky. The reasoning is circular but
sound and informative nonetheless (compare Sosa and Van Cleve 2001). That
inference to the best explanation is supported by its own products makes for a circle,
but not a vicious or self-defeating one. It is true that if the principle of inference did
not lead to truth, then its being more explanatorily coherent to believe that it does
would not seem to lend it much useful support. But in the absence of knowledge of
the antecedent, the coherence of the principle with its products and with other basic
epistemic sources can be seen to give us reason to accept it, just as inconsistency or
incoherence with other principles would give us reason to reject a principle.

Sceptics, however, will have a strong rejoinder to this response to the charge of
circularity. They will point out that perfect coherence in a set of beliefs, including
epistemic beliefs, even together with truth, does not suffice for knowledge. Brains
in vats may have coherent sets of beliefs, and when some of those beliefs about
objects outside the vats also happen to be true, this does not give them knowledge.
A coherent set of beliefs describes some possible world, but not necessarily the actual
one. If we took the entire set of a person’s beliefs and transferred them to another
person in a different set of circumstances, the set would be equally coherent, but
would contain much less knowledge (Sosa 1991, p. 203). It is clear once more from
these examples that, in order to constitute knowledge, a coherent set of beliefs must
be anchored to the actual world or surrounding environment in the right way.

Perceptual input in itself is insufficient to provide the required anchor. Such input
could be provided by a deceiving demon or by the programmers of the brains in
vats, and a complete madman could weirdly process perceptual input so as to make
it cohere with his other mad beliefs. Such input and such processing would not help
in the acquisition of knowledge even if it occasionally resulted in true belief. The
perceptual experience must be of the right kind, and it must result in the right kind
of belief to be of use in acquiring or demonstrating knowledge. But we cannot simply
check this experience itself or the causal chains by which it produces beliefs in order
to confirm that it and they are of the right kinds. The causal chains are largely
inaccessible, and the experience must be conceptualised correctly in order to be of
epistemic use. Experience itself need not be conceptualised (or it could not be the
source of concepts), but until it is, it can play only an inaccessible causal role. To
be of use in demonstrating knowledge, experiential input must be conceptualised
in such a way that the best explanations for the beliefs it produces appeal to the truth of those beliefs. It will be clear that the madman’s beliefs fail this test from
the fact that they fail to cohere with the beliefs of others, despite their internal
coherence.

In our own case, in order to demonstrate knowledge we must defend our beliefs
against sceptical challenge. Normally we defend a belief by citing evidence for it
that is part of an explanatory chain leading to the fact believed. We do so for simple
perceptual beliefs as well. For example, I defend my belief that there is a red object
before me by noting that I am appeared to redly (but not seemingly by red beams
of light). The claim that there is indeed a red appearance that explains my belief in
the red object can be defended not only by inferring it as the explanation for that
belief, but also directly by appeal to the belief that I am appeared to in that way
(which is not to imply that I normally infer the belief about the object from the
belief about the appearance). But here I arrive at a point in the demonstration where
I no longer defend my belief by appeal to evidence or an explanatory chain. The
best explanation for my belief that I am appeared to redly is simply that I am so
appeared to, that ‘red’ is the correct concept or term in my vocabulary to apply to
this experience, that I apply this term consistently to this type of experience. The
latter is all that is required for the truth of my belief, and the only evidence for its
being the case is my having formed the belief itself.

These beliefs about certain appearances form the foundations for the demonstration
of knowledge in two related senses. First, they are shown to constitute
knowledge without appeal to evidence or coherence with other beliefs. Second,
they make up a set of most certain beliefs which with others must cohere, picking
out one set of coherent beliefs as true of the actual world of the subject. Once this
anchor is in place, it is doubtful that there are equally coherent but incompatible
sets of beliefs with equal explanatory coherence, and the main objection to
demonstrating knowledge by showing coherence in a set of beliefs falls away.
In clarifying the sense in which these beliefs are foundations, several caveats must
also be mentioned.

First, the relevant beliefs about appearances are not infallible, in contrast to their
characterization by traditional foundationalism. In order to pick out patterns
in experience from which objective properties can be inferred as causes in the
demonstration of knowledge, they must refer to properties that are instantiated on
different occasions and in different objects. While ‘appears red’ (as opposed to
‘appears to be red’) refers to a phenomenal property, a property picked out by what
it is like to experience it, it is not defined ostensively on each occasion as referring
only to whatever property is present in the visual field. A belief employing the latter
ostensive concept might be infallible, but it would not be of any epistemic use. Beliefs
in reinstantiated or reinstantiable properties are always fallible, since the concepts
they employ can be misapplied or applied inconsistently. But for certain properties that are naturally salient in experience, the best explanation for beliefs about how
they appear will appeal only to the truth of these beliefs, as indicated earlier. The
second caveat is that this will be true only for beliefs about how these naturally salient
properties appear, for beliefs about red but not about C-sharp. The best explanation
for my belief that I am appeared to C-sharply appeals to my musical training, and
not simply to the fact that I am so appeared to. Third, it is not necessary in order
to serve the function of foundations that the defence of every other belief trace a
line back to one or more of these beliefs about appearances. Knowledge can have
a web-like instead of linear structure as long as it is anchored to the world at key
points.