A Finite Base Belief Revision System

Theories of dynamic reasoning invariably are more elegant on paper than they are in an implementation. A major criticism of the AGM belief revision logic [Gärdenfors 1988] is that it relies on calculations using infinite belief sets to satisfy its postulates for rational belief revision. This has stopped it from being used in any practical systems for reasoning. [Williams 1993] proposes a theory of belief revision which uses finite theory bases to represent the belief sets, and an ordered partition of the base, called the ensconcement ordering, to represent the epistemic entrenchment ordering of the AGM model. Williams defines operators for expansion, contraction and revision which correspond to the AGM operators, and shows that these achieve a "best approximation" to the AGM logic, for a finite reasoning agent. We describe the implementation of Williams’ finite base belief revision theory, and discuss the practical issues involved in implementing a system which is guaranteed to terminate in reasonable time. We show some simple test cases, and give a brief analysis of the system’s performance on these examples.