James C. Lester, Charles B. Callaway, Brian A. Stone and Stuart G. Towns
Knowledge-based learning environments provide an ideal testbed for developing and evaluating computational models of mixed initiative interaction. In each problem-solving episode, learners incrementally develop solutions for problems posed by the environment. To maximize learning effectiveness and learning efficiency, we have been developing animated pedagogical agents that dynamically provide explanatory advice in response to changing problem-solving contexts in learning environments. Animated pedagogical agents monitor students’ problem-solving activities and intervene with explanations in appropriate contexts. When students reach impasses, agents take control of the interaction and provide appropriate assistance. Dynamically controlling complex problem-solving interactions requires animated agents to make runtime decisions about when to intervene, how to select the content, with what level of directness to present hints, and with which media the advice should be delivered. This paper outlines some fundamental issues in mixed initiative problem solving with animated pedagogical agents and presents implemented solutions to these problems.