AI boosts homework grades but lowers exam scores

A study indicates that artificial intelligence (AI) helps students achieve better scores on their classwork and spend less time completing it, but in the long run, they perform worse on exams.

These are the conclusions of The Penalty of Learning from Generative AI: Evidence from Chinese Secondary Education, recently published by David Strömberg, Victor Lei, and Yanhui Wu through the pan-European think tank Centre for Economic Policy Research (CEPR).

The researchers used 30 months of panel data on 26,811 Chinese students in grades 7 to 12 and studied how generative AI affects task productivity and learning. The information combines monthly closed-book exams, high school and university entrance exams, and task scores and completion times in nine subjects.

The experts found that the use of AI increased task grades by 18% and reduced completion time by 30%, but decreased monthly exam scores by 20% after six months.

The total penalty became evident after two years in entrance exam scores, as they dropped between 18 and 24%. The most affected areas of knowledge were social sciences, followed by STEM and languages, and the most impacted were first-year students, high-performing students, and males.

Additionally, the research observed that learning losses were concentrated in approximately 80% of AI users who used it to speed up tasks. AI users who maintained a task execution time similar to those who did not use it experienced small learning losses.