Brown professor Roberto Serrano: AI cheating, scores fell 50%

Roberto Serrano, the blind economics professor teaching ECON 1170 at Brown University, ordered an in-person final after a take-home midterm on March 5 produced an average score of 96 out of 100. The subsequent in-person final averaged 48, enrollment had jumped to 86, and 27 students either dropped the course or did not attend the final.

What happened in Serrano's ECON 1170 class?

The midterm, given as a take-home exam on March 5, produced an average of 96 with 40 students scoring a perfect 100; by contrast the in-person final averaged 48, and 18 students dropped the course while nine others did not sit the final, and El País noted that of those 27 students, 22 had scored a perfect 100 on the midterm. Serrano told Inside Higher Ed that historically the midterm average in this course has ranged between 65 and 80 percent, and he also said the take-home midterm was actually harder than past exams because take-home format allows more time to challenge the class. Serrano and his graduate students fed the midterm questions into ChatGPT and received similar answers, and many student solutions, even when correct, had a "very convoluted style," which raised his suspicions.

The course normally attracts very few students; Serrano has never had more than 30 students enrolled at a time, and on some occasions only eight. This semester the switch to take-home exams corresponded with 86 students enrolling. Serrano emailed the class that he would not immediately void the midterm but would give students a chance to "prove me wrong" by comparing the distributions of the final and midterm; he warned that he would declare the midterm void and reweigh the final if the distributions did not line up.

How widespread is student use of generative AI on campus?

A recent Princeton survey found that 29.9 percent of students admitted to cheating with AI on at least one exam or assignment. At Brown, a provost-led report on "Generative AI in Teaching and Learning" found that 56 percent of undergraduate respondents and 67 percent of graduate and medical student respondents reported intentionally using GenAI tools daily or weekly, while large majorities of students also expressed concerns about the impact of GenAI on their learning and feared negative consequences for their cognitive capacity. Brown has released that report as it wrestles with policy and pedagogy questions around generative AI.

Why it matters

The drop from a midterm average of 96 to a final average of 48 in the same course is a stark, concrete indicator that suspected AI-assisted work can mask whether students have actually learned material. Serrano framed the issue as more than policy: he said, "We cannot choose to become idiots," arguing that widespread acceptance of cheating risks eroding the duty of universities to defend human thought and learning. The case also highlights administrative strain: rapid increases in enrollment, ambiguous boundaries for acceptable GenAI use, and uneven faculty responses complicate efforts to preserve assessment integrity.

What to watch

Watch whether Brown or Serrano formally voids the midterm and reweighs grades, and whether the university adopts new enforcement or assessment policies following its provost-led report. Also watch for any follow-up reporting on disciplinary action tied to the 27 students who dropped or skipped the final, and for how other courses at Brown respond to suspected AI-assisted assessments.