Is generative AI helping or harming learning? A field experiment in a higher education institution

CRIS SCIENTIFIC SEMINAR 2025/2026

TALK WITH MARTIN ARANGUREN, CNRS Research Professor, Sciences Po - CRIS 

By mid 2026, the adoption of generative AI tools for self-study has become nearly universal among university students. Does AI use augment or stunt learning? More exactly: does the use of generative AI during self-study improve or worsen learning outcomes? 

The few experimental studies that in the last two years have started to address the question indicate either a negative or a null effect of AI use on learning gains (with none supporting a positive effect). 

To contribute to this emerging corpus, we used an elective masters’ course delivered in the Spring semester of 2026 as the vehicle of a field experiment. Students were randomly assigned to producing a mid-term essay either making unrestricted use of generative AI (e.g. a commercial chatbot such as ChatGPT) or making no use at all of this technology. For the rest they worked under the same instructions. A few weeks after submitting the essay, all students took the same final exam in a traditional in-person, written, and unassisted format. 

Did AI use at the time of preparing the mid-term essay have any effect on the grades of the final exam? The talk describes the design and implementation of this pilot study.

Visit the Researcher's Website (CRIS) - Sign Up for the Seminar