A Decision-Theoretic Abductive Basis for Planning

David Poole and Keiji Kanazawa

in Working Notes AAAI Spring Symposium on Decision-Theoretic Planning , Stanford, March, 1994

Abstract

This paper presents a coherent synthesis of logic and decision theory and shows how it can be used. We allow an axiomatization of the world in definite clauses from a set of assumables. These assumables are partitioned into the set of controllable assumables and uncontrollable assumables. The uncontrollable assumables have probabilities associated with them. The logic allows multiple concurrent actions and lets us predict the effects for both the uncontrolled and controlled cases. We show an example of its use and argue that abduction, particularly probabilistic abduction, lays an important groundwork for decision theoretic and probabilistic planning. The main empirical claim is that uncertainty and choices can be represented as independent exogenous events using logic to give the consequences of the events, resulting in a powerful and yet simple representation.

You can get gzipped postscript.


Last updated 14 Oct 94 - David Poole