Logical Foundations for Decision-Theoretic Planning

John L. Pollock

In decision-theoretic planning, plans are compared in terms of their expected-values. It has been supposed that this is justified by classical decision theory, but in fact this is incompatible with classical decision theory. Plan-based decision theory must be defended by showing that classical decision theory is wrong in ways that necessitate appealing to plans instead of actions in isolation. This paper proceeds by raising difficulties for classical decision theory and suggesting ways of meeting them which eventually lead to a kind of plan-based decision theory. The resulting plan-based decision theory differs in two way from more familiar versions proposed in the literature on decision theoretic planning. First, it evaluates plans in terms of their "expected utilities", which is different from the expected-value of executing the plan. Second, the objective is not to find optimal plans. It is argued that most plans cannot be compared by comparing their expected-utilities, and this makes optimality ill-defined. Instead, plans must be evaluated in terms of their contribution to the expectedutility of the agent’s "master plan", which results from merging all of the agent’s currently adopted plans into a single plan.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.