AGNESS: a generalized network-based expert system shell

Abstract

AGNESS is an expert system shell developed at the University of Minnesota. AGNESS is more general than other shells. It uses a computation network to represent expert defined rules, and can handle any well-defined inference method. The system works with non-numeric as well as numeric data, and shares constructs whenever possible to achieve increased storage efficiency. AGNESS uses a menu-driven user interface, and has several features that make the system friendly and convenient to use. The system includes eight explanation queries designed to increase the amount of information available to the user, the expert, and the knowledge engineer while remaining simple enough to be included in most of today’s expert system shells. AGNESS has been tested on several domains ranging from simplified problems to real world medical analysis. I. lNTRODUCTlON The design of expert consultation systems has been a topic of growing interest in Artificial Intelligence (Al) research during the past decade. Numerous expert systems have been constructed to give consultations in a variety of application areas. Two prominent examples of this are MYCIN [l], a program for the diagnosis of infectious diseases, and PROSPECTOR [2], a mineral exploration system. The common aim of expert system technology is to represent and apply knowledge obtained from a specialist in the problem domain. Early in the history of this technology, people realized that rewriting the entire system for a new domain was both wasteful and unnecessary. Since most of the operational code can be separated from the domain specific knowledge, one program can be written to handle rule bases from several domains. Using this idea, a system can be developed for a new domain by simply changing the rules that the operational system handles. This operational system is called a skeletal system or an expert system shell. Many expert system shells have been implemented recently with varying degrees of success. The best known of these are KEE ( Knowledge Engineering Environment) from Intellicorp, LOOPS developed at the Xerox Palo Alto Research Center, and

Ähnliche Arbeiten