Economist Report on 27,000 Chinese Students: AI Use Lifts Homework Scores 18% but Closed-Book Exam Scores Fall 20%

The Economist reported on August 18 a study covering 26,811 middle and high school students in a county in central China, tracked from September 2022 to June 2025. After using AI, average homework scores rose 18% and completion time fell from 64 to 45 minutes, while closed-book monthly exam scores were 20% lower than those of non-AI users, about 1.4 standard deviations. The study also says the outsourcing rate rose from 58% to 81% after five months of use, and the learning penalty narrowed from 25% in early 2023 to 16% in June 2025.

The study provides verifiable quantitative evidence for schools, teachers and parents judging the boundaries of student AI use, and directly affects how AI education products design process records and anti-outsourcing mechanisms. Two direct effects: schools need to reassess the reliability of homework scores as a learning outcome measure; AI learning tools that only optimize submission speed and answer quality may widen the gap between homework and exams. Evidence limits: the sample comes from one county in central China, and the paper's wording on high school and college entrance exam effects is more restrained than the widely circulated 24% and 18%, with an average effect of about 7% across all AI users. Schools and teachers need to reassess the reliability of homework scores as a learning outcome measure, as the divergence between closed-book exams and homework performance changes daily evaluation. AI learning products that only optimize answer generation and submission speed may worsen homework outsourcing, so product design needs process records and capability verification.