Affective Behavior Control for Lifelike Pedagogical Agents

Abstract

Lifelike pedagogical agents should be especially effective in constructivist learning environments in which students participate in active problem solving. We can simulate such a constructivist setting with personal computing using a well-designed, evocative graphical interface and the rich multimedia -- audio, video and animation -- currently available. Beyond such an authentic problem-solving context, constructivist learning employs a social aspect, centered on the interaction of learner and mentor. We submit that an animated pedagogical agent who delivers contextualized problem-solving advice can play the part of the expert. Further, we propose that an added measure of believability and motivation would result from giving these agents the ability to express situationally appropriate emotions. To test the promise of such an affective agent, we first identified the cognitive emotion types most useful in a problem-solving dialog. We then devised a structure to store the details of the learner's situation in order to determine the appropriate emotion from the pedagogical agent. These enhancements have been instantiated in a full-scale implementation of the lifelike pedagogical agent of DESIGN-A-PLANT, a learning environment developed in the domain of botanical anatomy and physiology for middle-school students. Evaluation by a focus group of students was encouraging. They preferred the emotional version of the agent and reported that his affective behavior was helpful in their problem solving.