Learning · Cognition and Behavior

Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics

Summary

A trial with nearly 1,000 high school math students compared a regular ChatGPT-style tutor to one prompted to safeguard learning. Both improved practice scores hugely, but on a test without AI the regular-tutor students did 17% worse than students who never had AI at all; the safeguarded version erased that gap, because the difference comes down to whether the student uses AI as a crutch to copy or as a teacher to learn from.

Topics: Cognition, Learning
Access the report →

Reference

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R.. (2024). Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics. SSRN Working Paper. https://doi.org/10.2139/ssrn.4895486