AI Use in Education: Homework Score Increases vs. Exam Score Declines

AI tools create a performance gap between homework and exams

Artificial intelligence tools increase short-term homework performance but correlate with a decline in long-term exam performance when AI assistance is removed. A study of 27,000 students aged 12 to 18 in China found that students using AI models—such as Doubao and DeepSeek—saw their average homework scores rise by 18% over six months. However, these same students scored 20% lower than their peers who did not use AI when taking exams under closed-book conditions.

Study methodology and findings

Researchers from Stockholm University and the University of Hong Kong tracked a large cohort of students to measure the impact of AI on learning outcomes. The study divided students into two groups: an AI-using group (approximately 80% of the students) and a control group of non-users (approximately 20%).

Key findings include:

• Homework Performance: AI users experienced an 18% increase in scores across all subjects over a six-month period.
• Exam Performance: In tests taken without AI assistance, AI users performed 20% worse than the control group.

Comparative research on AI in learning

These results align with a 2024 University of Pennsylvania study on mathematics education. In that research, students practicing with AI tools, including ChatGPT and AI tutoring programs, performed better during short-term practice sessions but failed to to carry that advantage over to subsequent closed-book tests.

Educational implications and risks

Overreliance on AI for completing assignments may prevent students from developing the necessary cognitive skills required for independent problem-solving. The Brookings Institution has cautioned that while AI can support learning when designed intentionally, using the technology to replace critical thinking, creativity, or social interaction can hinder the development of essential cognitive and social skills.

Educators have also noted a trend of "formulaic, similar-sounding essays" generated by chatbots, suggesting that students may be copying AI-generated answers without deeply engaging with the material.

Global adoption rates of AI in education

AI adoption among students is widespread across various demographics and regions:

• Wealthy Countries: A Chegg survey found that 80% of undergraduate students report using AI in their studies.
• Britain and Germany: More recent polling indicates adoption rates of 93% in Germany and 94% in Britain.