Roberto Serrano, an economics professor at Brown University, required an in-person final exam after unusually high take-home midterm results raised suspicions of generative AI use. After the supervised exam, the average score among students who took it fell from 96 to 48, suggesting widespread use of external assistance on the earlier assessment.
What happened?
The episode traces back to December 2025, when a shooting on Brown’s campus left two people dead, including someone who had recently introduced themselves to Serrano. Disturbed by the incident, Serrano allowed both the midterm and the final for his ECON 1170 course to be taken as take-home assessments during the Spring 2026 semester.
That change coincided with a surge in enrollment: 86 students signed up for the course, compared with the usual class sizes that rarely exceeded 30 and sometimes were as small as eight. The midterm held on March 5 produced striking results: an average score of 96 out of 100, and 40 students achieved a perfect 100.
Serrano noted that historically the midterm averages for this course ranged between 65 and 80 percent. He also stressed that the take-home format allowed him to set a harder exam because students had unlimited time.
Why did Serrano suspect AI use?
Besides the high scores, the style of many answers looked unusual—overly elaborate and convoluted. Serrano and his graduate students ran the exam prompts through ChatGPT and found very similar responses. To determine whether students could reproduce such performance under supervision, Serrano announced that the final exam would be held in person. He told the class he would not immediately invalidate the midterm but would compare distributions: if the final matched the midterm, he would count it; if not, he expected to invalidate the midterm and adjust final grades.
Following the announcement, 18 students immediately dropped the course and nine did not show up for the in-person final. Of these 27 students, 22 had previously scored a perfect 100 on the midterm. Among those who took the supervised final, the average score plunged from 96 to 48.
Serrano’s view and background
Serrano, who was born in Spain, spoke at length with El País and Inside Higher Ed about the incident. He expressed deep concern that AI-enabled shortcuts could deprive students of real learning and erode independent thought. He warned that tolerating such behavior among talented young people risks societal decline, calling it a path toward a “failed society.”
The reporting also noted Serrano’s personal background: he lost his sight at 17 due to retinal dystrophy, learned Braille, and went on to study at Harvard—experiences that shape his stance on discipline and adaptation.
Broader context: student use of generative AI
Brown University has itself been examining generative AI’s role in education. A university report titled “Generative AI in Teaching and Learning” cites a survey finding that 56 percent of undergraduate students and 67 percent of master’s and medical students use generative AI tools daily or weekly. Separately, a recent Princeton University survey found that 29.9 percent of students admitted to using AI to cheat on at least one exam or assignment.
Many students also express worries that relying on AI harms their learning and could impair cognitive abilities over time. Serrano shares those concerns and argues universities must defend independent human thinking.
Consequences and open questions
Serrano has not closed the matter, even though he says Brown’s administration responded cautiously. The case highlights how widespread access to generative AI complicates assessment design, academic integrity enforcement, and the preservation of authentic learning outcomes. It raises urgent questions about how institutions should adapt evaluation methods to ensure students are genuinely mastering course material while balancing fair use of new tools.
