Study: Heavy AI Use Linked to Significant Drop in Student Exam Scores

创业邦 · wechat · 2026-08-24

Based on a 30-month study tracking 27,000 Chinese students, The Economist reveals the profound negative impact of AI on learning:

Performance Reversal: Students using AI for homework saw average scores rise by 18% with 30% less time spent. However, their closed-book exam scores were 20% lower (1.4 standard deviations) than non-AI users.
Homework as a False Signal: Homework scores ceased to be a reliable predictor of exam success. Deeper AI usage correlated with steeper exam declines, affecting top-performing students the most.
The "Cognitive Offloading" Trap: AI crowds out high-intensity cognitive effort. Students shift from problem-solvers to mere auditors, but auditing without internal understanding leads to "black box" oversight. After 6 months, almost all students slid into fully outsourcing homework, proving that "correct usage" is unsustainable.
Warning for Adults: Similar patterns are observed in doctors, writers, knowledge workers, and programmers. While AI boosts immediate output, it causes skill degradation when AI is removed (e.g., doctors' adenoma detection rates dropped by 20%).
Conclusion: AI doesn't lower IQ but removes the training process required for intellectual growth. By outsourcing drafting and trial-and-error, the development of mental models is stunted.

More from AGI Musings