Agents for Education Bringing Adaptive Behavior to an Internet Learning Community

Can adaptive behavior be useful in a human learning environment? The work presented here proposes that introducing adaptive agents to a human learning community can be e ective in several ways: 1. to guide selection of appropriate problems for human learners, 2. to embody \self-organizing" curricula, 3. to provide simple student models, 4. to impersonate human learning partners, and 5. to facilitate indirect human-human interaction. Taking a population-based approach and using evolutionary computation methodologies, we bring two types of software agents into an Internet learning community. Inside our CEL (Community of Evolving Learners) system, humans play competitive, educational games, openly or incognito, with each other and with software agents, while the populations of learners { human and agent { advance. The overarching goal of this research is to investigate the various factors that contribute to this type of environment and their e ects on both human and machine learners. We wish to demonstrate that the progress of each student can be tracked and that the agent populations adjust according to changes in human behavior, taken both on an individual and aggregate basis. Hopefully, the result will be that humans participants frequent the system over time and become active members of the learning community.