In recent years many epistemologists have urged that the concept of knowledge be explicated 'naturalistically', as empirically reliable true belief. 1 According to such accounts, S knows that p when not only is his/her belief that p true in the actual world, but when in alternative possible worlds, worlds in which it is not the case that p, S does not believe that p. Just as a thermometer is reliable when differing states of the environment would produce differing readings, so is a person's belief reliable, and constitutive of knowledge, when differing states-of-affairs would produce differing cognitive responses.
The reliability theorist does not wish to demand, however, that, in any and every alternative world in which it is not the case that p, S does not believe that p. Such a demand, surely, results in skepticism; in skeptical scenarios, S believes that p even though it is not the case that p. Instead, the reliability theorist wants to restrict the scope of the reliability demanded of the knower to those alternatives that are relevant. Thus:
(RA) S knows that p iff (i) S believes that p (ii) It is true that p (iii) S discriminates, or is able to discriminate, between the stateof-affairs that p, and relevant alternatives. That is, for every relevant alternative R to the state-of-affairs that p, if R were to obtain, S would not believe that p.2
The foregoing suggests that a 'relevant-alternatives' approach to knowledge evolves rather smoothly out of the core ideas of the reliability account. It is my view, however, that, ira relevant-alternatives approach is to be viable, it must be pursued in independence from reliability accounts. To establish this view, I propose to argue for two related positions. The first is that, even if a knower must discriminate between the actual state-of-affairs and relevant alternatives, a naturalistic reliability account will generate a particularly implausible thesis as to when an alternative is relevant; a plausible thesis, on