Landmark Study Reveals Generative AI Boosts Homework Grades But Lowers Exam Performance
A comprehensive 30-month academic study tracking 27,000 students aged 12 to 18 has revealed an alarming disparity between AI-assisted coursework and actual learning outcomes. Conducted by researchers from Stockholm University and the University of Hong Kong across nine core subjects, the landmark study evaluated homework performance, monthly closed-book tests, and high-stakes entrance examinations.
According to the findings published as schools reopen globally, students using generative AI models saw their average homework scores rise by 18 percent, while completing assignments nearly 30 percent faster—reducing average completion time from 64 minutes to 45 minutes. However, when these identical students sat for closed-book examinations without digital tools, their test scores plummeted by 20 percent within six months. The decline was also evident in long-term entrance examination performance.
Educational experts and international bodies, including the OECD in its Digital Education Outlook, note that using AI tools without structured instructional frameworks leads to metacognitive laziness. While generative AI chatbots deliver immediate solutions for routine homework, students bypass critical problem-solving and critical thinking steps essential for long-term knowledge retention.
The study's authors emphasize that schools should not outright ban artificial intelligence, but rather redesign educational policies. They urge institutions to guide students toward using AI as an interactive tutor for underlying concepts rather than an automated solution engine, ensuring students develop authentic understanding before taking standardized exams.
