Reliability, Reasons, and Belief Contexts

Abstract

Here’s a problem that any reliability theory must face, whether it’s one that holds that beliefs are justified just when they’re products of belief-forming mechanisms with the potential of having good records of yielding true beliefs (as in Goldman 1979 and 1986), or one that holds that a belief meets the standards for knowledge if and only if its causal basis rules out any relevant chance of mistake (as in Dretske 1971 and 1981, and Nozick 1981). The problem is made evident when cast in probabilistic terms. Let r be S’s reason for tokening the true belief that p under conditions c. Then, according to reliabilism (and with a number of other things being equal), Sis justified in believing that p under c iff r makes it sufficiently probable that p, whileS knows that p iff the conditional probability that p on r and c is unity. But how do we specify the belief context c to be counted as epistemically relevant? As urged recently by John Pollock (1984), on the face of it there seems to be no principled reason for excluding mention of the truth value of the proposition believed. As he says, ‘The only obvious way to construct an objective non-epistemic kind of definite probability is to make it conditional on everything that is true [at the belief’s tokening]’ (Pollock 1984, 110); and, as he notes, ‘one feature of the present circumstances is a characterization of the belief as a belief in [p ], and another feature is the truth value of [p]’ (109). But if we allow both the belief token and its truth value membership in the relevant belief context, we then have the hopeless task of explaining how the probability of the token’s being true can be anything but one or zero, one if what is believed is true, and zero if it’s false. This in tum would commit reliabilism to the foolish doctrines that a belief’s truth is sufficient for knowledge, while no false belief is ever even justified. No wonder, then, that Pollock says ‘there is no way to construct an intelligible notion of reliability which does the job required by the reliabilist’ (105). And he is surely right in this: Neither knowledge nor justification can be explained as the chance of a belief’s being true if the relevant belief context is required to mention both the belief tokened and its truth value. We would be engaged in a comparably pointless task were we to try to explain the likelihood of a wager’s being a good bet on a horse race when we had to include among the givens both which horse was picked and its finish, or were we to try to make sense of the assertion that a theory’s predictions have a good chance of being true when both what the theory predicts and its outcome had to be included among the givens. (Quine makes a similar point when he says, ‘If there is no distinguishing between a thing’s disposition to act in a certain way in certain circumstances and the mere fact of its so acting in those circumstances, then whatever the thing may do can be laid to a disposition, by defining the circumstances narrowly enough’ [Quine 1973, 5]. See also Goldman 1979, 12; Goldman 1986, 49-51; and Feldman 1985 for related discussions.)