A study tracking 26,811 Chinese secondary students over 30 months found that generative AI tools used to expedite homework completion are linked to significantly lower exam performance. The research, published by the Centre for Economic Policy Research as a discussion paper titled “The Generative AI Learning Penalty,” indicates that AI adoption reduced homework completion time by approximately 30 percent and increased homework scores by 18 percent. However, in settings where AI was prohibited, monthly exam scores decreased by about 20 percent within six months, with high-stakes entrance exam penalties reaching 18 to 24 percent over two years.
The study identified that roughly 80 percent of the learning losses stemmed from students who completed assignments unusually fast with high homework scores. This pattern suggests that students were outsourcing cognitive tasks to AI instead of engaging with the material. In contrast, students who maintained homework completion times similar to non-users experienced only minor declines in performance. The sharpest declines in exam scores were observed in social sciences, followed by STEM and language subjects, and were notably higher among younger students, high-achievers, and boys.
Wharton professor Ethan Mollick noted that the context of AI use plays a critical role, stating, “AI tutoring in support of classes is good, using AI to ‘help’ with homework is bad.” This finding aligns with a separate study published in Nature, which reported that students using an AI tutor in class learned more efficiently and displayed higher engagement compared to traditional academic methods.
As AI adoption among students in China accelerates, a nationwide survey found that over 60 percent of primary and secondary school students have used AI, with 71 percent utilizing it for homework assistance. In the U.S., an NPR/Ipsos poll revealed that 55 percent of K-12 teachers believe AI primarily serves as a shortcut that allows students to evade more rigorous work.
The implications for educators and policymakers center on whether AI can be directed toward enhancing supplemental tutoring rather than replacing traditional homework practices. Some U.S. colleges have already adjusted their evaluation methods, with an increasing number of professors implementing oral exams and in-class assignments to mitigate AI-related shortcuts. However, the feasibility of scaling similar interventions across K-12 education systems, particularly in China, remains uncertain in a market reportedly valued at over $43 billion.
