Steps Towards Programs that Manage Uncertainty
This addresses the need for AI systems to actively manage uncertainty in domains like medicine and plant pathology, though it appears incremental as it builds on existing credibility assessment techniques.
The paper tackles the problem of what to do when uncertain in AI reasoning, developing MUM, a medical expert system that plans diagnostic sequences, and later MU, a framework for building such systems, demonstrated by reimplementing MUM and a plant pathology diagnostic system.
Reasoning under uncertainty in Al hats come to mean assessing the credibility of hypotheses inferred from evidence. But techniques for assessing credibility do not tell a problem solver what to do when it is uncertain. This is the focus of our current research. We have developed a medical expert system called MUM, for Managing Uncertainty in Medicine, that plans diagnostic sequences of questions, tests, and treatments. This paper describes the kinds of problems that MUM was designed to solve and gives a brief description of its architecture. More recently, we have built an empty version of MUM called MU, and used it to reimplement MUM and a small diagnostic system for plant pathology. The latter part of the paper describes the features of MU that make it appropriate for building expert systems that manage uncertainty.