LEARNING BY TEACHING AGENTS

By
Thomas Katzlberger

The overall goal in this thesis was to develop an intelligent learning environment that adopted an explanatory and constructivist learning approach, and helped students learn and improve their problem solving activities in the mathematical domain on distance-rate-time problems. The learning environment implemented for this thesis combines the STAR-Legacy cycle str ucture (Schwartz, Lin et al. 1999), variants of Smart Tools (Owens et al. 1995), and Adventure Player (Crews et al. 1997) into a single learning environment that implements a learning by teaching approach. Smart Tools provided the interactive representations that users create for problem solving based on their knowledge or understanding of domain concepts. Adventure Player provided a problem-solving environment with planning and simulation tools for developing and verifying solutions to complex problems. Adventure Player was linked to the Rescue at Boones Meadow episode of the Adventures of Jasper Woodbury series that adopted an anchored instruction approach for teaching and learning (Cognition and Technology Group at Vanderbilt 1997). The outer cycle structure into which all of the above tools were embedded was adapted from the STAR-Legacy system to provide a structured sequence of increasingly difficult problems. Each cycle in our learning environment draws problems from a global anchoring context, organizes learning activities, and arranges formative and summative feedback during learning. We extended this environment with software agents that could be taught, to pursue the goal to improve motivation and the ability to transfer beyond the set of problems that students were taught to solve.