The N+1 approach to education
Pavel Katunin on incremental teaching, reinforcement learning and the three models behind Glite’s approach.
Reinforcement learning in education
Teaching is a sequence of decisions. The next activity should depend on what a learner knows, what they need to learn and what earlier activities changed.
We use reinforcement learning as a way to formulate this problem: a teaching policy chooses an activity, observes the learner’s response and updates its estimate of their knowledge. The difficult part is defining a reward that reflects lasting learning, rather than a convenient short-term proxy. A useful formulation still needs empirical validation.
Related articles
Tutoring as a reinforcement learning task
Teaching as a sequence of decisions under uncertainty: a reinforcement learning perspective.
Faster and Safer Testing for AI Tutors
Can student models help us evaluate teaching strategies before trying them with learners?