AI boosted homework scores, then exam scores dropped: study
A new study finds students using AI tools saw homework grades rise significantly — but their exam performance fell, suggesting AI is doing the learning for them.
The 60-second version
Students using AI for homework get better grades on assignments but worse exam scores, suggesting AI substitutes for learning rather than supporting it.
Key points
- Homework scores rose for students with AI access, but exam scores fell compared to non-AI-using peers.
- The 'AI crutch effect' means students complete assignments without developing underlying skills.
- School responses vary from outright bans to integration as teaching aids.
- The challenge is designing assessments that measure genuine understanding, not AI-assisted output.
Verdict. AI is a powerful tool, but it can undermine learning when used as a substitute for cognitive effort. The solution isn't banning AI — it's redesigning how we teach and assess.
The findingAI helps homework, hurts exams
A new study covered by The Economist reveals a paradoxical effect of AI in education: students who use AI tools see their homework scores improve, but their exam scores decline. The pattern held across multiple subjects and student demographics.
The study tracked students who had access to AI tools for homework assignments but took traditional closed-book exams without AI. The results suggest that AI is substituting for — rather than supporting — the learning process.
The mechanismWhy AI use hurts long-term learning
The mechanism is straightforward: AI tools can solve problems, write explanations, and even generate code or math solutions step by step. When students use these tools for homework, they complete assignments successfully — but they never develop the underlying skills themselves.
This creates a learning illusion: students feel they understand the material because they're producing correct answers. But the understanding is shallow. When tested without AI, they can't reproduce the work.
The debateBan or integrate?
School districts and universities are divided on how to respond. Some have banned AI tools outright, citing plagiarism and academic integrity concerns. Others are trying to integrate AI as a teaching aid — using it for personalized tutoring, feedback, and practice problems.
The study suggests that the answer may depend on how AI is used. Passive AI use — letting the tool do the work — appears harmful. Active AI use — where the tool explains, questions, and guides — may be more beneficial. But designing assessments that measure genuine understanding, rather than AI-assisted output, is an ongoing challenge.
The bottom lineWhat this means for students and parents
For students: if you're using AI for homework, test yourself without it. Can you solve the problem from scratch? If not, the AI is doing the learning, not you.
For parents: check whether your child's homework is truly their own work. Ask them to explain the concepts without the tool. For educators: the study is a strong argument for in-class, closed-book assessments that test genuine understanding.
AI doesn't make learning obsolete — it makes genuine learning harder to distinguish from imitation.
Primary sourcesThe Economist·Hacker News discussion