A Mixed-Initiative Approach to Teaching Agents to Do Things

Mihai Boicu, Dorin Marcu, Michael Bowman, and Gheorghe Tecuci

The research problem addressed in this paper is the development of knowledge bases and knowledge based agents by a subject matter expert that does not have knowledge engineering experience and receives limited support from a knowledge engineer. One possible solution to this problem is provided by the mixed-initiative Disciple approach where an expert teaches a learning agent while solving problems in cooperation. Disciple synergistically integrates the complementary human and automated reasoning to take advantage of their respective knowledge, reasoning styles and computational strengths. The general strategy behind the Disciple approach is to replace the difficult knowledge engineering tasks required to build a knowledge base, tasks that cannot be performed by a subject matter expert, with simpler tasks that can be performed by the expert. The teaching process is illustrated with examples involving the Disciple workaround agent that is able to plan how a convoy of military vehicles can work around obstacles in their path.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.