Students are increasingly relying on AI for schoolwork, but research suggests this trend may harm their learning.
In 2024, Laurence Holt highlighted the “5 Percent Problem” in ed-tech, where only a small fraction of students benefit from programs as intended. AI, however, has seen widespread adoption, with students using it extensively.
Despite its popularity, studies on AI’s impact on learning remain limited. A recent Stanford analysis found only 20 out of 800 studies provided credible evidence.
Evidence of AI’s Impact on Learning
A recent study, “The Generative AI Learning Penalty: Evidence from Chinese Secondary Education,” analyzed data from approximately 27,000 Chinese students in grades seven to 12. Between 2023 and 2025, about 80% of students started using generative AI. Within five months, approximately 50% fully outsourced their homework to AI, leading to a 24% drop in high school exam scores and 18% in college exams.
The research found no evidence of learning benefits from AI use, instead identifying a “Generative AI penalty,” where students who use AI perform worse over time.
The study’s data reveals a stark contrast between students who used AI and those who did not. AI users spent less time on homework, received high scores matching AI capability, yet performed poorly on exams.
Peer Pressure and AI Adoption in Schools
Students often feel compelled to use AI when they see classmates achieving higher grades with its help. This dynamic can create a cycle where even initially resistant students adopt AI to keep up academically.
Addressing the AI Learning Penalty
Critics of the study’s findings struggle to provide comparable evidence of AI’s benefits in education. The data shows no discernible advantage for AI users, even when they study equally.
