Eric Horvitz, David Heckerman, Keung-Chi Ng, Bharat Nathwani
Medical Computer Science Group
Knowledge Systems Laboratory
Stanford University
Stanford, California 94305
Author Email: horvitz@microsoft.com
Click here to access postscript file.
Click here to access pdf file.
A criticism of diagnostic systems, which are based on the formal foundations of probability and utility, is that their reasoning strategies and recommendations are inflexible and unnatural. We have developed a facility that increases the flexibility of normative reasoning systems by providing multiple human-oriented perspectives on diagnostic problem solving. The method endows a system with the ability to reason about arbitrary classes of diagnostic entities and to control the level of abstraction at which inference occurs. The techniques have been integrated into Pathfinder, an expert system that performs hematopathology diagnosis. We explain the background and approach that we have taken, and describe how we use the techniques in Pathfinder to modulate information- and decision-theoretic reasoning with strategic scripts that are familiar to physicians.
Keywords: Understandability of probabilistic reasoning, explanation of Bayesian inference, Bayesian reasoning, expected value of information, human-oriented inference procedures.
E. Horvitz, D. Heckerman, K. Ng, B. Nathwani, Heuristic Abstraction in the Decision-Theoretic Pathfinder System, Proceedings of the
Symposium on Computers in Medical Care, Washington DC, IEEE Press: Silver Springs MD, November 1989.