Co-evolutionary Learning : Machines and Humans Schooling Together

We consider a new form of co-evolutionary learning in which human students and software tutors become partners in a cooperative learning process. While the students are learning from the tutors, the tutors will also be learning how to do their job more e ectively through their interactions with the students. Earlier work on game-playing machine learners has brought to light important issues for co-evolutionary learning; in particular, that these interactions can be modeled as a meta-game between teachers and students, and that learning may fail due to suboptimal equilibria { for example, because of collusion between the individual players { in this meta-game of learning. We discuss some of the challenges that these issues may raise for the designers of intelligent software tutoring systems and describe a prototype Java applet that we have built in an e ort to explore how these challenges may best be met.