User-Agent Interactions in Mixed-Initiative Learning

Mixed-initiative learning integrates complementary human and automated reasoning, taking advantage of their respective reasoning styles and computational strengths in order to solve complex learning problems. Mixed-initiative learning is at the basis of the Disciple approach for developing intelligent agents where a subject matter expert teaches an agent how to perform complex problem solving tasks and the agent learns from the expert, building and refining its knowledge base. Implementation of practical mixed-initiative learning systems, such as those from the Disciple family, requires advanced user-agent interactions to facilitate user-agent communication, the distribution of tasks between them, and the shift of initiative and control. This paper discusses some of these user-agent interaction issues in the context of the mixed-initiative rule learning method of the most recent version of the Disciple system.