RED: a classificatory and abductive expert system
Jr. Jack W Smith · OhioLink ETD Center (Ohio Library and Information Network) · 1986
This dissertation addresses the development and refinement of knowledge engineering guidelines for the use of certain generic tasks in knowledge-based systems. Generic tasks are a characterization of problem-solving based on generic types of input/output, organizations of knowledge, and control of knowledge evocation. The knowledge engineering issues are illustrated by using examples from medical knowledge-based systems within the context of a proposed architecture for abduction. Through these systems, extensions to the guidelines for the epistemic analysis of a domain with the proposed architecture for abduction in mind are discussed. These guidelines are useful for identifying: (1) The nature of the problem-solving types to be undertaken by the knowledge-based system within the architecture; (2) And the portions of the knowledge that need to be allocated to the different portions of the architecture. The application of these guidelines is first illustrated by using them to describe the MDX1 system, a knowledge-based system for the diagnosis of liver disease. MDX1 performs sub-hypothesis selection using the generic task of classification. MDX1's inferencing database embodies the generic task of knowledge-directed information passing. This task provides appropriate data or data abstractions about the problem description. The use of the guidelines are then illustrated by the design and implementation of the RED/ID subsystem. RED/ID is a prototype subsystem of the RED knowledge-based system designed to perform the interpretations of antigrams for the purpose of antibody identification. RED/ID uses in addition to the MDX1 tasks the abductive assembly/criticism of explanatory hypotheses task. This task assembles the sub-hypotheses returned by sub-hypotheses selection into the overall best available conclusion for the case. Lessons learned from using the proposed architecture and guildelines for RED/ID are related to knowledge allocation considerations for other knowledge-based systems embodying the tasks in the proposed abduction architecture. A discussion in this regard demonstrates that MYCIN and INTERNIST-1 can be described from the perspective of the proposed abductive architecture and the previously discussed tasks.