The Generative AI Learning Penalty: Evidence from Chinese Secondary Education (cepr.org)

A recent study analyzing data from over 26,000 Chinese secondary students has uncovered a significant trend regarding the impact of generative AI on educational outcomes. While the use of AI tools boosted homework scores by 18% and reduced completion time by 30%, it led to a marked decline in monthly exam scores by 20% within just six months. The research indicates that high-stakes entrance-exam scores experienced declines of 18 to 24%, with these detrimental effects becoming more pronounced over time, particularly among junior students, high achievers, and boys. This study reveals a "generative AI learning penalty," suggesting that excessive reliance on AI for homework may hinder deeper learning and retention of knowledge, especially in social sciences, STEM, and language subjects. The analysis highlights that approximately 80% of students who utilized AI exhibited behaviors indicative of outsourcing homework, which correlates with high scores but short completion times. Conversely, students who maintained similar homework habits to non-AI users experienced minor learning losses, emphasizing the need for a balanced approach to AI integration in education. This research underscores vital implications for educators and policymakers as they navigate the challenges posed by emerging AI technologies in the classroom.