A large-scale study tracking 26,811 Chinese secondary students over 30 months found significant declines in exam performance associated with the use of generative AI tools for homework. The research, published by the Centre for Economic Policy Research as a discussion paper titled “The Generative AI Learning Penalty,” indicated that while AI adoption reduced homework completion time by approximately 30 percent and increased homework scores by 18 percent, monthly exam scores fell by about 20 percent within six months. Over two years, the decline in high-stakes entrance exam scores reached between 18 to 24 percent.
The study identified effort displacement as the core issue, with around 80 percent of the learning losses attributed to students completing assignments unusually quickly. This trend was linked to the outsourcing of cognitive work to AI, which contributed to higher homework scores but diminished engagement. Students who maintained completion times similar to non-AI users experienced only minor declines in their exam performances. The most significant drops were observed in social sciences, followed by STEM subjects and languages, with younger students, high-achievers, and boys being particularly affected.
Professor Ethan Mollick emphasized the nuanced findings, stating, “AI tutoring in support of classes is good, using AI to ‘help’ with homework is bad.” This reflects a growing understanding that the context in which AI is used plays a more critical role than its mere presence. A separate study published in “Nature” found that students using an AI tutor during class learned more efficiently and reported higher engagement.
AI adoption among Chinese students is accelerating rapidly, with over 60 percent of primary and secondary school students reported to have used AI, and 71 percent utilizing it for homework. In the United States, a recent NPR/Ipsos poll indicated that 55 percent of K-12 teachers believe AI primarily functions as a shortcut for students, detracting from their learning.
Educators and policymakers face challenges in redirecting AI usage toward supplemental tutoring rather than as a replacement for homework. Some U.S. college professors have started implementing oral exams and in-class assignments to limit AI-enabled shortcuts. The scalability of similar interventions in K-12 systems, especially in China, where the AI-powered educational technology market is valued at over $43 billion, remains uncertain.
