A study in elementary rule based reasoning

In an effort to provide sophisticated performance, Knowledge Based Systems limit their application to a narrow domain and instruct the computer how to "reason" within that domain. Instructions are typically represented as rules which express the conditions under which conclusions may be derived. By interpreting the rules within a given data base context the systems can plan, analyze and control inferencing in goal directed manners. This thesis documents a study of the Knowledge Based System methodology. Knowledge Bases are described in terms of structure and control. To exemplify design issues a rule interpreter is demonstrated and five experimental systems are surveyed. It is proposed that certain reasoning processes which are entailed in the generation of theories can be modeled as a rule governed system. Drawing upon the propose-evaluate-refine approach to learning, the model claims: 1. Hypotheses can be generated by detecting patterns in a data base and describing the observations as logical expressions. 2. By evaluating the generated hypotheses against all currently available data, relevant evidence can be assembled and classified. 3. By analyzing accrued evidence and patterns among the hypotheses, the system can determine how to revise its hypotheses so as to satisfy criteria required of a theory. (See more in text.)