In philosophical theories of knowledge (epistemology), justification is a desirable property that one's beliefs ought to have before they can be accepted as part of a rational discourse. Roughly, for internalists about justification, a belief is justified if the subject has or has available to him good reasons to believe it; while for externalists a belief is justified if there exist good reasons to believe it, regardless of whether the subject actually has or has access to those reasons. One such externalist view of justification is reliabilism, the position that a belief is justified if it has been produced by a properly functioning belief-forming mechanism (BFM). Some examples of BFMs available to human beings are sense perception, memory, and deduction.

Epistemic circularity is a notorious problem for reliabilism. If a belief is only justified if it was produced reliably by a certain BFM, how can I ever know for sure that a certain BFM is itself reliable? For instance, take the meta-belief that "sense perception is a reliable BFM." This belief is produced, at least in part, through sense perception itself, for example by analyzing the track record of my past sense perceptions and finding it to be in good order. But if a BFM is thus allowed to vouch for its own trustworthiness, then we have no way to discriminate between reliable and unreliable BFMs. After all, when trying to ascertain if a suspect in a murder case is sincere, it is quite irresponsible to ask the suspect himself. Thus, internalist critics complain, reliabilism is plagued by epistemic circularity and loses sight of the normative goal of epistemology.

Reliabilist responses to this serious charge have been of two kinds: (1) to show that epistemically circular arguments can be justificatory, and thus that BFMs can vouch for their own re-liability; or (2) to concede that epistemically circular arguments cannot be justificatory, but then to also insist that some higher-level circularity must be allowed in one's justificatory practices, or no beliefs at all can ever be justified. Here I argue that the first strategy fails and the second succeeds. Internalists are correct that epistemically circular arguments cannot be justificatory in the way that some reliabilists expect them to be, but they are incorrect that all circularity must be banished before our justificatory practices can be virtuous. To always allow circularity makes knowledge reprehensibly easy, but to never let it in at all is a kind of epistemic suicide.