Conditionalizing expectations

An information system characterizes a view of the world. Typically, this view is incomplete and subject to change, and, as a consequence, such systems use nonmonotonic reasoning to form expectations about the world, and they modify their expectations as new information is acquired. For instance, database systems use the closed world assumption (a very naive variety of nonmonotonic reasoning), and their expectations are updated when new information becomes available in the form of transactions. Gärdenfors and Makinson (1988, 1994) have shown that nonmonotonic inferences can be constructed from a preference ordering of expectations. In this paper, we adapt the process of conditionalization and techniques developed in the area of belief revision to handle changes in the nonmonotonic information encapsulated in an expectation ordering, thereby providing a mechanism for modeling the removal of old expectations, the incorporation of new expectations, as well as the raising and lowering of existing expectations. Changes to the expectation ordering using conditionalization are based on a relative measure of minimal change