AI can now watch videos of students learning together — and catch what humans miss

Researchers compared AI systems at analyzing recorded classroom footage, testing whether multiple AI agents working together beat a single AI doing the work alone. The multi-agent approach won — it spotted student behaviors more accurately, which means universities and schools could eventually automate the tedious work of manually coding video data from collaborative learning sessions.

Right now, analyzing video of students working together requires humans to watch footage and write down what happened — who spoke, who was paying attention, who was stuck. It's slow and expensive. This paper shows that AI agents can do parts of this work better than previous AI approaches, which means institutions could scale up video analysis of learning without hiring more people to watch tapes. The catch: this is still a proof-of-concept. It works on curated clips with clear behaviors. Real classrooms are messier.