Seeking safety in knowledge

Knowledge demands more than accuracy: epistemologists are broadly agreed that those who know are non-accidentally right, satisfying some kind of safety condition. However, it is hard to formulate any adequate account of safety, and harder still to explain exactly why we care about it. This paper approaches the problem by looking at a concrete human cognitive capacity, face recognition, to see where epistemic safety shows up in it. Drawing on new models in artificial intelligence, and making a case that human face recognition is relevantly similar, I argue that safety is a natural feature of successful generalization. The concluding section of the paper applies this lesson to the problem of our intuitive recognition of knowledge.