Document Type
Thesis
Date of Award
1-31-1991
Degree Name
Master of Science in Computer Science - (M.S.)
Department
Computer and Information Science
First Advisor
James Geller
Abstract
In many domains, for example medical diagnosis, there are several sources of knowledge, e.g., several domain experts. These domain experts may have disagreements over some issues in the domain. Research in the area of human decision making suggests that a human decision is not always made based on the advice of a single expert. Many current expert systems, however, expect consistency in their knowledge bases. An expert system does not deliberate by reasoning that "on one hand..." but "on the other hand..." The knowledge engineer, who is not necessarily an expert in the same domain, ensures consistency by eliminating all conflict between any two pieces of knowledge. This consistency is undesirable because it leads to inflexible reasoning, but the lack of it could cause unpredictable results.
Therefore, it makes sense to design a system where the knowledge of all the experts is available at the decision making stage, even if it is inconsistent In this research we are investigating issues of Multiple Expert Systems where the knowledge of several experts is stored without any modification or alteration. No effort is made to resolve differences in opinion among experts at this stage. Conflicts, if any, are resolved only after a query is received and only for the specific query. Little work has been reported in the literature on expert systems that do not require resolution of conflict at the time of knowledge acquisition We argue that, among other advantages, this provides a more exact simulation of real-life decision making processes.
Specifically, we present theory and implementation of a Multiple Expert System shell based on Prolog which permits the partial integration of independently formulated sub-knowledge-bases of several experts. Our knowledge base is made up of knowledge of several experts in separate belief spaces and a unified format, without any modifications to remove inconsistencies among experts. For any query, relevant knowledge bases are preselected. The system reports results indicating agreement or disagreement among experts for the query. It overcomes differences among the vocabularies of the different experts involved. The shell also supports predicate transparency to alleviate the user from having to guess how a proposition is analyzed into a predicate and arguments. An application of the Multiple Expert System shell to Pascal program debugging is also presented.
Recommended Citation
Mittal, Umesh Kumar, "Multiple expert system" (1991). Theses. 2560.
https://digitalcommons.njit.edu/theses/2560
