Education edition

An AI tutor only gets students thinking when the conversation itself is graded. Mike Kentz argues this in an opinion piece. He relies on a US study of Khanmigo, Khan Academy's AI tutor. At 18 high schools in Tennessee, every student had an AI tutor within reach for two years. Of the students, 96% tried the tutor at least once. But the median student used it on only a third of the days. The conversations were rarely about math, and the gains in math were modest.

According to Kentz, when help is voluntary, students do not choose the tutor that asks questions. They choose the chatbot that gives the answer. When he required students to use AI and said he would review the conversations, he saw more effort. That was true for him too. He found a role-play, in which the AI plays a difficult person, more engaging than a regular tutor. Above all, knowing that his teacher was watching pushed him to think better.

What does this mean for you? According to this piece, a bot with no obligation works poorly. If you want students to think, make the conversation itself the assignment and grade it yourself.

The evidence is weak. Kentz's own pilot was small, had no control group and consisted mostly of self-reported data. He also sells a product based on this idea: AI Friction Labs.

Source →