Belief Revision in a Changing World

Abstract

Several authors have recently argued for a distinction in the way beliefs are updated with new information. They distinguish between information that tells the agent that the world has changed over time and the information that fills in or corrects the agent's picture of the world at a particular time. This chapter presents a representation of this distinction by means of a modal logic that combines epistemic and dynamic features. Moreover, it develops a completely declarative semantics for belief revision. This semantics enables one to deduce the result of revising a given body of beliefs in the light of new information, given simply the semantic content of the prior beliefs and of the new data. No purely procedural assumptions about the agent's epistemic policies or values (no information about priorities of defaults or degrees of entrenchment) are needed, beyond what is explicitly represented in the objects of the agent's beliefs. This is accomplished by distinguishing hard (incorrigible, unrevisable) belief and soft belief; further, the soft attitudes supervene on the hard level. A specific theory of nonmonotonic inference is used to generate soft attitudes from hard ones.