For the second edition of Contemporary Theories of Knowledge, John Pollock and I have updated Pollock's influential earlier discussion (1986) of the main themes of analytic epistemology. Our focus is on the nature of rational (or justified) belief, as this has been thought to be a crucial component of knowledge and human intellectual flourishing. We retained the structure of the first edition, where we provide a taxonomy of theories of justification and discuss a range general features of proposals on rationality within that taxonomy. Thus, we offer treatments of internalist theories like foundationalism and coherentism, and externalist theories like probabilism, reliabilism, and the theory of proper functions. Throughout that discussion, we attend to the major specific contributions to the theory of knowledge of last thirty years. We then present our own view. This gives us an opportunity to reflect on epistemic normativity and on the relation between epistemology and cognitive science.

The book is distinctive in offering an extended analysis of the limits of probability theory in an account of rationality, in distinguishing between multiple levels of epistemological theorizing, in giving a general account of epistemic norms, and in developing a naturalistic epistemology that is not externalist.

Our view is that epistemic norms are descriptions of (probably innate) patterns of belief formation that human beings recognize as rational. These norms describe extremely abstract or general features of reasoning about perception, memory, time, or generalization from a limited sample. We are most certainly not proposing that complicated and sophisticated reasoning of the kind found in politics, religion, or cultural practice is in any simple sense innate, where some conclusions are somehow necessarily correct. We do maintain that sophisticated reasoning is composed of the simpler reasoning that we identify, but — in the case of sophisticated reasoning — the interaction between the norms and the diversity of starting premises yields a variety of intellectual commitments.

On another front, we also recognize that human beings often form beliefs irrationally. Thus, we treat the set of rational norms as a competence theory rather than a performance theory. This distinction is analogous to the competence/performance distinction in linguistics, where competent language users are held to possess a competence theory of a grammar even if actual performance of the language is often ungrammatical.

The traditional methodology of analytic epistemology is to test intuitions about cases in order to frame a theory of justification or knowledge. We claim that this is one way to uncover some of our epistemic norms, but it is not a very good way. Intuition reveals some of the contours of the patterns of reasoning we find rational, but it breaks down under test cases that have multiple defeaters. So, we propose that artificial intelligence (AI) realizations of epistemic norms will shed brighter light on our patterns of reasoning. By encoding a range of epistemic norms in an AI, we can experiment with the interaction of those norms in order to see whether the resulting belief is one we find reasonable. If it is not, then something has gone wrong with the account of epistemic norms, their interaction, or both. We view AI as a tool for better uncovering the nature of our norms.

In order for this to be illuminating, one need not be committed to the thesis that encoding epistemic norms in an AI is creating intelligence. Pollock does have that commitment, but the argument for it does not appear in Contemporary Theories of Knowledge, nor is it crucial to the central themes of the book. The parts of the book that require technical programming knowledge are brief and easily omitted. On the other hand, we have sought to provide enough detail so that cognitive scientists can use our framework to test their proposals for epistemic norms.