Toward a Science of Expert Systems
Proceedings of the 18th Symposium on the Interface of Statistics and Computer Science, Fort Collins, Colorado, March 1986 |
Published by American Statistical Association: Washington, D.C.
Over the last several years, teams working on expert systems have been exploring formal approaches for belief revision and information acquisition. The formalization of major components of expert systems operation is useful for understanding and characterizing system behavior and for predicting changes with modification. Formalization also facilitates the involvement of investigators in more well-developed disciplines such as statistics. While the use of formal methodologies for diagnostic problem solving is attractive because of the generality, power, and axiomatic basis of inference, the methodologies have been criticized for making inferences that are difficult to understand and explain. I shall focus on the problem of explaining formal reasoning methodologies. The PATHFINDER system for pathology diagnosis is presented as an example of current research on aspects of the use of formal methodologies in expert systems. I will demonstrate that a formal system is amenable to controlled degradation to enhance its explanation capability.