holland@GMR.CSNET (Steve Holland) (03/04/86)
Seminar at General Motors Research Laboratories, Warren, Michigan: Belief Functions in Artificial Intelligence Prof. Glenn Shafer University of Kansas Lawrence, Kansas 66045 Thursday, March 6, 1986 ABSTRACT The theory of belief functions, or the Dempster-Shafer theory, has attracted wide interest as a tool for the management of uncertainty in artificial intelligence. What are the advantages and disadvantages of belief functions when they are compared with numerical alternatives such as Bayesian probability and fuzzy logic or with non-numerical alternatives such as default logic and the calculus of endorsements? What are the current prospects for sensible use of belief functions in expert systems? In this talk, I will offer some general judgments on these questions. I will emphasize the need for interactive tools for the construction of probability arguments, and I will speculate on long-term possibilities for probability judgment using man-made associateve memories. -Steve Holland, Computer Science Department