FAGAN@SUMEX-AIM.ARPA (Larry Fagan) (05/17/86)
A Decision-Theoretic Approach to Explaining Heuristic Planning
Curtis P. Langlotz
PhD Oral Exam
Medical Information Sciences
Stanford University
Thursday, May 22, 1:15 PM
Medical Center M-112
Many important planning problems are characterized by uncertainty
about the current situation and by uncertainty about the consequences
of future action. These problems also inevitably involve tradeoffs
between the costs and benefits associated with possible actions.
Decision theory is an extensively studied methodology for reasoning
under these conditions, but has not been explicitly and satisfactorily
integrated with artificial intelligence approaches to planning.
Likewise, many perceived practical limitations of decision theory,
such as problem solving results that are difficult to explain and
computational needs that are difficult to satisfy, can be overcome
through the use of artificial intelligence techniques. This thesis
explores the combination of decision-theoretic and artificial
intelligence approaches to planning, and shows that this combination
allows better explanation of planning decisions than either one alone.
In addition, the explicit representation of probabilities and
utilities allows flexibility in the construction of a planning system.
This means that assumptions made by such systems, which may be
critical for their performance, are more easily modified than in a
system that does not explicitly represent uncertainties and tradeoffs.