Knowledge Acquisition with a Knowledge-Intensive Machine Learning System

In this paper, we investigate the integration of knowledge acquisition and machine learning techniques. We argue that existing machine learning techniques can be made more useful as knowledge acquisition tools by allowing the expert to have greater control over and interaction with the learning process. We describe a number of extensions to FOCL (a multistrategy Horn-clause learning program) that have greatly enhanced its power as a knowledge acquisition tool, paying particular attention to the utility of maintaining a connection between a rule and the set of examples explained by the rule. The objective of this research is to make the modification of a domain theory analogous to the use of a spread sheet. A prototype knowledge acquisition tool, FOCL-1-2-3, has been constructed in order to evaluate the strengths and weaknesses of this approach.