Abstract

This paper distinguishes between "descriptive" and "normative" conceptions of Bayesian principles of rationality, both in the context of inference and in the context of decision (which of course are not unrelated). I emphasize an idea according to which, "You have to work with what you have to work with" - that is, that rationality is a relation among old beliefs, new information, and new beliefs (in the case of inference) and among beliefs, desires, preferences, and choices (in the case of decision). According to this conception of rationality, one's current beliefs and desires are not themselves subject to evaluation as to their rationality (except for some minimal, basically logical and "coherentist" constraints). From this perspective, rationality is about how we move from old beliefs (whatever they are) to new beliefs when confronted with evidence, and about how our preferences are structured given what we believe and what we want (whatever we currently happen to believe and want). I present some formal details of this perspective and discuss several criticisms of it.