AI Is Making You Dumb: Study Shows Using Students Using AI Can Score Higher In Homework, But Eventually Fail in Exams

Key Points

BE HONEST, how many of you used artificial intelligence (AI) tools to score better in your school or college assignments? It's no secret that this technology has weeded its way into classrooms, with students increasingly relying on this technology to complete their class assignments and boost their scores. And why wouldn’t they? AI tools — such as chatbots like ChatGPT and DeepSeek — offer students a free or low-cost way to finish their assignments timely and achieve good scores. These tools may also help students in understanding concepts better as per their liking, something they might not get to with their teachers. But this practice has a downside: a recent research study points that students using AI may score higher in their homework, but eventually they fail in exams. 

The research survey, titled “The Generative AI Learning Penalty: Evidence From Chinese Secondary Education” was conducted in June 2026. Researchers David Stromberg of Stockholm University and Victor Lei and Wu Yanhui of University of Hong Kong tracked 26,811 pupils aged between 12-18 years of age to assess the impact generative AI has on exam scores and learning. The study was conducted in areas where AI adoption has been the fastest.

See also: University of Chicago Bans AI in Core Social Sciences Courses; Restricts Use of Laptop, Mobile Phones in Classrooms

AI and Education

The use of artificial intelligence is becoming increasingly common among students and school-going children. A survey conducted last year by Chegg, an American ed-tech firm, found that 80% of university students in developed countries use AI for their studies. More recent surveys have reported even higher usage, with 94% of students in Britain and 93% in Germany saying they use the technology. 

Several countries have been quick to adopt AI in education. China, for instance, requires students to receive at least eight hours of AI education each year. In the Gulf region, the United Arab Emirates, Saudi Arabia, and Qatar have also started including AI in their school curricula. AI chatbots can help students complete their work more quickly and easily. However, what happens when students do not have access to the technology? How much of what they learned by doing their homework remains with the students, and how much of the work was essentially done by the machine? 

Faster Homework, Poorer Exam Performance? What The Research Study Reveals

To explore this question, researchers David Stromberg and Yanhui Wu conducted one of the largest studies of its kind by tracking around 26,811 students between the ages of 12 and 18 in China over a period of 30 months. Their research compiled data from homework, monthly closed-book examinations and high-stakes entrance tests covering nine subjects. The findings showed a clear difference between students completing schoolwork and them actually retaining what was learned. 

The findings are striking. After six months of using AI, students recorded an 18% increase in their average homework scores across subjects. At the same time, the average time spent completing each assignment dropped from 64 minutes to 45 minutes (top and middle graphs). However, the improvement did not carry over to exam results. The students who used generative AI to complete homework scored 20% lower in exam results than those students who had not. (bottom graph). 

See also: What's On Gen Z’s Mind About Politics? C-Voter Survey Reveals Majority Feel Political Parties Don't Understand Their Concerns, Social Media Key Factor Behind Voting Decisions

What Does The Findings Mean For Students and Educators?

Using AI to do homework can make things look better than they really are. Scores go up and work may get done faster, but the exam results show that real learning may not be happening. Homework is meant for students to practice and understand their curriculum better, not just merely a task to be completed. If AI does all the thinking and completes assignments, it leaves no room for students to actually learn and understand their curriculum.

This study suggests homework scores might not truly show how much a student has learned, since AI can boost them artificially. Teachers may need to rethink how they use homework grades, since they could end up rewarding AI use instead of actual understanding. It might help to add more in-class work, oral presentations, or other tasks. Exams still seem to show a truer picture of what students really know.

(Edited by Harsh Pandey)

Suggested reading: