AI tools improved homework scores but lowered exam performance in large-scale student study

A study tracking 27,000 students aged 12 to 18 in China found that those who used AI tools saw their average homework scores rise by 18% over six months, according to research summarized by The Economist. However, when the same students were tested under exam conditions without access to AI tools, they scored 20% below peers who had not used AI during the study period.

The research, conducted by David Stromberg of Stockholm University and Victor Lei and Wu Yanhui of the University of Hong Kong, compared students who reported using AI models such as Doubao and DeepSeek with a control group of students who did not use such tools. Around 80% of surveyed students reported using AI tools, while the remaining 20% formed the control group.

The study’s findings align with a 2024 University of Pennsylvania study that observed similar short-term gains from AI use in math practice, but no advantage on subsequent closed-book tests. Educators have increasingly reported grading essays that resemble AI-generated content, though robust evidence on AI’s effects on learning outcomes had been limited prior to these studies.

The Stockholm University and University of Hong Kong researchers’ full study had not been independently verified by other institutions at the time of publication. Commenters on The Economist’s summary of the findings suggested the exam score gap may stem from students copying AI-generated answers without deeper engagement with the material.