Mixed-Initiative Agent Teaching and Learning

This paper presents a mixed-initiative teaching and learning method that allows a subject matter expert, who does not have knowledge engineering experience, to transfer his problem solving expertise to a learning agent. The expert is teaching the agent to perform domain specific tasks in a way that resembles how the expert would teach an apprentice while solving problems in cooperation. The method integrates synergistically the complementary knowledge and reasoning styles of the expert and the agent, where the expert helps the agent to learn and the agent helps the expert to teach it. This approach is illustrated with examples of teaching the Disciple-COA agent to critique courses of actions, a challenge problem defined in the DARPA's High Performance Knowledge Bases program.