A Methodology for Uncertainty in Knowledge-Based Systems

In this book the consequent use of probability theory is proposed for handling uncertainty in expert systems. It is shown that methods violating this suggestion may have dangerous consequences (e. g., the Dempster-Shafer rule and the method used in MYCIN). The necessity of some requirements for a correct combining of uncertain information in expert systems is demonstrated and suitable rules are provided. The possibility is taken into account that interval estimates are given instead of exact information about probabilities. For combining information containing interval estimates rules are provided which are useful in many cases.