Roberto Serrano, a Madrid-born economist and professor at Brown University who has been losing his sight since adolescence, made his mathematical economics midterm a take-home exam after a December incident in which a gunman entered Brown’s campus and killed two students. Serrano said he did not want to force students back into a classroom under those circumstances.
What happened on the midterm
After the take-home exam, Serrano noticed unusually high results: the class average was 96, whereas historical averages for this course typically ranged from 65 to 80. Forty students received perfect scores of 100 on the midterm. Serrano reported that many of the answers resembled the solutions produced by ChatGPT for the same problems, prompting suspicion of widespread AI-assisted cheating.
In response, Serrano required the final exam to be taken in person. The in-person final produced very different results: the course average dropped from 96 to 48. Many students who had enrolled knowing the course would be offered remotely performed much worse under supervised, face-to-face conditions.
Dropouts and no-shows
Serrano also reported that 27 students either dropped the course or did not attend the in-person final; of those, 22 had scored 100 on the midterm. These figures reinforced his assessment that a substantial portion of the midterm results were not the product of independent student work.
Responses and the broader debate
Serrano has framed the episode as more than a classroom integrity issue. He warned that normalizing cheating contributes to societal decline, arguing that when promising young people accept dishonesty, the larger community suffers. The case at Brown has reignited debate about how universities should respond to the rise of generative AI in education.
Options discussed by educators fall broadly into two camps. One strategy emphasizes defensive measures: return to in-person exams, oral defenses, and stricter enforcement of academic codes. Proponents argue that these approaches preserve the integrity of assessment and ensure that degrees remain meaningful.
The alternative is adaptation: accept that AI is part of the educational landscape and redesign pedagogy and assessment accordingly. That can include AI-literacy training, revised honor codes, and assignments that integrate or are resilient to AI assistance. Some institutions prefer incremental measures such as orientation workshops on AI ethics and revised course policies.
Serrano has been critical of what he sees as too-soft institutional responses, arguing they fail to recognize the broader societal implications of the technology.
A collective dilemma
Economists would note the situation embodies a classic collective-action problem: each student individually benefits from exploiting a professor’s trust to secure a high grade, but if everyone does so, the credential loses its value and trust in the system collapses — a dynamic comparable to the “tragedy of the commons.”
The Brown episode shows measurable consequences: steep drops in average performance and significant course attrition once supervised assessment was reintroduced. How universities choose to respond — stricter proctoring and oral exams versus curricular adaptation and AI literacy — will affect the credibility of degrees and the learning habits of future cohorts.
Conclusion
The incident at Brown University highlights that generative AI introduces not only technical challenges for academic integrity but also moral and institutional questions. Whether institutions will meet the challenge by reinforcing traditional assessment methods or by fundamentally redesigning teaching and evaluation to incorporate AI remains a pivotal decision for higher education and for the signaling value of academic credentials.
