Reliabilist theories propose to analyse epistemic justification in terms of reliability. This paper argues that if we pay attention to the details of probability theory we find that there is no concept of reliability that can possibly play the role required by reliabilist theories. A distinction is drawn between the general reliability of a process and the single case reliability of an individual belief, And it is argued that neither notion can serve the reliabilist adequately.

The reliability of a belief-producing process is a matter of how likely it is that the process will produce beliefs that are true. The term reliabilism may be used to refer to any position that makes this idea of reliability central to the explication of some important epistemic concept. I know of three such positions that appeal to some epistemologists: a reliabilist account of what makes a belief justified, a reliabilist account of what makes a true belief knowledge, and a (...) reliabilist answer to the question of the fourth condition, the question of what must be added to justified true belief to make knowledge. Obviously these are alternative positions rather than parts of a single coherent whole. I think of the first as reliabilism’s boldest stand, the second as the position to which it may retreat when the first is found untenable, and the third as its last refuge. I will criticize only the first two positions. (shrink)