Learning to Teach: an Architecture for Self-aware Educational Games

In this proposal I describe a new kind of educational game architecture: the self-aware educational game. Instead of simply amplifying an educational game designer’s best guess at a teaching strategy, self-aware games will actively reason over uncertainty about how a subject should be taught, perform experiments, and meaningfully adapt to collected evidence on a global scale. Combining accepted educational theory, evidence-based educational practices, probabilistic inference carried out by machines, and deeply generative game design, self-aware games do not presume the effectiveness of any single teaching strategy; they learn to teach.