Students who use AI score great on homework, but bad on tests, large-scale new study shows
See more of our coverage in your search results. Add The New York Post on Google.
See more of our coverage in your search results. Add The New York Post on Google.
Article outline
- What happened
- The key numbers
- Background
- The details
- A closer look
- The bottom line
Key points
- The study tracked 27, 000 students aged 12 to 18 in China, where AI adoption has been turbocharged over recent years.
- While the remaining 20% formed the control group, some 80% documented using Chinese AI models such as Doubao and DeepSeek.
- All the while, the time they took to complete each assignment dropped from 64 minutes to 45, according to the Economist.
- But when it came to exams, students who applied AI for their homework scored 20% below their classmates who hadn't, the study demonstrated.
- In another new study, students at Middlebury College in Vermont were taught an unfamiliar topic either with or without the support of an AI chatbot.
Artificial intelligence is great for school work – until you can't employ it on the test, a large-scale new study indicates.
Students who applied AI o write their homework secured better scores and finished their work in record time -but did worse exam than kids who did things the old-fashioned way, according to new research out of China.
After six months, the students using AI saw their homework scores rise by almost one-fifth (18%) throughout all subjects, a joint study by David Stromberg of Stockholm University of Victor Lei and Wu Yanhui of the University of Hong Kong identified.
Experts say the rise of AI may have broken the relationship between homework grades and the amount of knowledge retained for tests.
In practice, the study tracked 27, 000 students aged 12 to 18 in China, where AI adoption has been turbocharged over recent years.
In the West, AI pickup has been even more drastic, with roughly 80-90% of US undergrads, 94% in Britain, and 93% in Germany admitting to using it in their studies, according to a survey last year by ed-tech firm Chegg.
But the evidence doesn't indicate that AI is inherently a bad thing for students.
Those using AI scored higher on tests than the control group and had the same advantages when they repeated the task a week afterwards, according to research by Zara Contractor and German Reyes.
"AI can boost learning productivity, but only for those who use the technology intelligently, " documented the Economist.
"Students must resist the temptation to reach for an AI-generated answer before thinking things through for themselves, " continued the report.
In short, students who use AI score great on homework, but bad on tests, is the central thread here, and readers can expect follow-up reporting as the picture becomes clearer.




