Harvard study shows that an AI tutor can outperform active learning in the classroom

In a randomized trial of physics students, an AI tutor designed specifically for teaching produced better learning outcomes than an active pedagogy classroom, which is known to be very effective.

This randomized controlled study, conducted at Harvard with 194 undergraduate physics students and published in Scientific Reports, compares the learning obtained via an AI tutor to that obtained in the active learning classroom — a pedagogy that has been documented for decades as clearly superior to the classic lecture, and taught here by highly rated teachers in a course specifically designed according to best pedagogical practices.

The students who worked with the AI tutor, called PS2 Pal and designed to never give the answer directly but to ask questions that guided the solution step by step, learned more than twice as much as those in the active class, with effect sizes between 0.73 and 1.3 standard deviations — and in less time.