An AI tutor that really helps
Recent studies show that AI can help or harm learning, depending on whether it gives the answers or makes students think.
ObjectiveDistinguish a use of AI that makes students learn from a use that replaces the effort, and set up the former.
Recent controlled trials give a nuanced picture. In a university physics course, an AI tutor designed according to precise pedagogical principles enabled students to learn more in less time than an in-class active learning session. In a secondary school, by contrast, students who freely used an assistant during the exercises did better on those exercises, then worse on the exam taken without assistance. A version designed to guide without giving the answer avoided this effect.
What makes the difference
- The tutor asks questions and gives hints before giving a solution.
- It draws on the teacher’s course, not on general knowledge.
- The student produces their own answer, then receives feedback.
- The final assessment is done without assistance, to measure what has been learned.
Try it in LearnyaYou are the tutor for my Year 9 pupils on the attached chapter. Never give the solution to an exercise. Ask a question to help them move forward, then give a hint if a pupil gets stuck twice. At the end, check that they can do a similar exercise on their own. Try in Learnya
References
- Kestin, Miller, Klales et al. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports 15, article 17458. doi.org/10.1038/s41598-025-97652-6
- Bastani, Bastani, Sungu et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS 122(26), e2422633122. doi.org/10.1073/pnas.2422633122