What matters is how AI helps
What Lapmin does. Lapmin leaves solving the assignment and writing the text to be submitted to the student.
What the researchers found. Nearly 1,000 students in grades 9–11 at one school in Turkey practised maths. Whole classes were randomly assigned to three groups: no AI, unrestricted AI that gave complete answers when asked, or AI that gave hints only.
Unrestricted AI lowered scores on a later test taken without AI. The test scores of the group given hints did not differ significantly from those of students who had practised without AI.
Students using unrestricted AI most often asked directly for an answer, and the data showed signs of answers being copied. Students using AI that gave hints sent more messages, and their conversations dealt more often with the substance of the work.
During practice, scores were 48% higher with unrestricted AI and 127% higher with hint-giving AI than without AI. The researchers say students can easily use unrestricted AI as a crutch, making the tool’s design a matter for careful consideration when introducing AI.
Connection to Lapmin. The unrestricted AI in the study gave complete answers to assignments. Lapmin answers factual questions directly, but gives hints for problems students must solve and text they must write, even when a student repeatedly asks for the complete answer.
Limit. The study used a different system from Lapmin. The hint-giving AI had the teacher’s solutions to the assignments in its instructions. The study covered one school, maths and four practice sessions.
Source.
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. https://doi.org/10.1073/pnas.2422633122