Study finds generative AI boosts homework results but lowers exam scores

A large study found generative AI improved homework results but was associated with weaker subsequent exam performance. Illustration: EdEvents News.

A study of 26,811 students in grades 7 to 12 in China has found that generative AI improved homework results and completion time, but was followed by lower performance in closed-book examinations.

Using 30 months of data across nine subjects, researchers found AI adoption raised homework scores by 18 per cent and reduced completion time by 30 per cent. Monthly exam scores fell by 20 per cent within six months.

The study also reported declines in high-stakes entrance-exam results, with the full effect emerging after about two years. The largest learning losses were concentrated among students whose very short completion times and high homework scores were consistent with outsourcing homework to AI.

Students who continued to spend a similar amount of time on homework as non-AI users experienced much smaller learning losses, suggesting the way the technology is used may be central to its effect.

Read the CEPR discussion paper