VAL@SAIL.STANFORD.EDU (Vladimir Lifschitz) (05/12/87)
SHOULD ED FEIGENBAUM AND I TALK TO EACH OTHER? Matt Ginsberg, Stanford (SJG@SAIL.STANFORD.EDU) Thursday, May 14, 4:15pm Bldg. 160, Room 161K In a previous talk, I argued on philosophical grounds that the time has come for the "neats" and "scruffies" in AI to begin to resolve their differences by working on problems of interest to each other. I suggested that, if one were to view the scruffy programs as performing two distinct tasks, one being conventional inference, and the other being some sort of "bookkeeping" with the results, insights could be obtained that would be of interest to both the formal and informal camps. In this talk, I discuss the application of this idea to problems of interest to the informal camp. Specifically, I will discuss the construction of a "flexible" expert sytem shell that can be easily tailored to solve problems using a variety of different methods, simply by changing an explicit set of bookkeeping functions. I will show the system using first-order logic to simulate a digital circuit, as suggested by Genesereth in his DART work, using an ATMS to diagnose the same digital circuit, as suggested recently by deKleer, solving a simple problem in default reasoning, and then solving the same problem more efficiently by using bookkeeping functions that include both default and justification information.