Intelligent Tutoring Systems

Intelligent tutoring systems use evidence about a learner’s knowledge, progress, and behavior to choose what to teach and how much help to provide. Established approaches include Bayesian knowledge tracing, item response theory, knowledge graphs, and reinforcement learning for selecting problems, hints, or learning paths; newer systems add conversational language models, retrieval from course materials, and specialized agents. Researchers evaluate these systems with tutoring benchmarks, simulated learners, classroom studies, and measures such as mastery, engagement, and transfer to independent work. A central challenge is ensuring that fluent, correct responses are also pedagogically appropriate: systems can misdiagnose reasoning, reveal answers too soon, or perform well in simulations without producing durable learning gains.