A dramatic shift in exam performance at Brown University has cast a spotlight on the growing tension between artificial intelligence tools and academic integrity. In a spring economics course, what began as a record-setting round of midterm results ended with a stark reversal after the professor reintroduced in-person testing.
Suspicious Gains Prompt a Return to the Classroom
Professor Robert Serrano, who teaches ECON 1170, had historically led small classes of highly capable students, with enrollment never exceeding 30 and sometimes numbering as few as eight. This semester, however, a change in the university's evaluation system helped swell the roster to 86 students. The midterm exam, administered as a take-home assignment on March 5, produced eye-catching results: the average score reached 96 out of 100, and 40 students achieved a perfect score.
Serrano noted that the typical midterm average in the course had long fallen between 65 and 80 percent. He added that this particular exam was designed to be more challenging than previous versions precisely because the take-home format allowed students to work without time pressure. The surge in top scores, in his view, pointed toward unauthorized AI use.
A Sudden Exodus and a Collapse in Scores
Concerned by the pattern, Serrano informed students that the final exam would be held in person. The announcement triggered an immediate response. Eighteen students dropped the course entirely, and an additional nine were absent for the final exam. Of those 27 students, 22 had earned a perfect 100 on the take-home midterm.
For the students who did sit the in-person final, the collective performance underwent a jarring decline. The average score fell from 96 to 48, a drop of exactly 50 percentage points that underscored the instructor’s suspicions about the integrity of the earlier results.
The Broader Academic Response to AI
Universities around the world are rethinking their approaches to generative AI, with many shifting focus from outright bans to frameworks for responsible use. A common structure now separates AI engagement into tiered levels: some assessments permit full AI assistance with attribution, others allow it only for specific tasks such as editing or brainstorming, and many traditional exams require students to work entirely without AI.
Serrano has adopted a firmly critical position on the matter. He argues that accepting widespread dishonesty among talented students poses a societal risk, cautioning that normalizing cheating erodes the foundations of a healthy civic and intellectual culture. His experience at Brown adds an empirical edge to a debate that is rapidly reshaping how institutions evaluate learning in the age of AI.
Sources: english.elpais.com, arstechnica.com