Can Machines Teach? A Review of AI Tutoring, Learning Analytics and Generative AI in Education

Abstract

Artificial intelligence has been promoted as a way to give every learner the benefits of a personal tutor. This review examines the evidence behind that promise across three generations of technology: intelligent tutoring systems, which model students' knowledge and guide them step by step through problems; learning analytics, which use data from digital learning environments to predict and support student success; and generative AI, whose conversational abilities have revived interest in automated tutoring. We organise the review around six questions that educators, researchers and policymakers commonly ask. The evidence suggests that well-designed intelligent tutoring systems produce meaningful learning gains, approaching those of human tutors in some comparisons, though effects on standardised tests are smaller. Learning analytics can identify students at risk but improve outcomes only when predictions lead to effective action. Early experimental evidence on generative AI shows that unrestricted access can raise performance during practice while harming learning, whereas tutors designed to guide rather than give answers can avoid this harm. We discuss equity, privacy and the role of teachers, and conclude that the value of AI in education depends less on technical capability than on pedagogical design.