Here’s a story that’s making the rounds in academic circles.
Roberto Serrano is a blind economics professor at Brown. He’s been teaching ECON 1170 for years — a tough course that usually attracts maybe 30 students. After a campus shooting in December 2025 shook him deeply, he decided to offer take-home exams for spring 2026. Enrollment jumped to 86.
The midterm results were remarkable. Average score: 96 out of 100. Forty students got perfect scores. Historically, the average for that exam was between 65 and 80. And Serrano says this exam was actually harder than previous ones.
Something felt off. The answers had a “very convoluted style,” he said. When he and his grad students ran the questions through ChatGPT, they got similar results.
So Serrano announced a change. The final exam would be in-person, proctored. He told the class he’d give them a chance to prove him wrong — if the final scores matched the midterm, he’d count both. Otherwise, the midterm would be voided.
Eighteen students dropped the course immediately. Nine more didn’t show up for the final. Of those 27, 22 had scored a perfect 100 on the midterm.
The students who did take the final? Average score: 48. Down from 96.
Serrano has since shared his story with El País and Inside Higher Ed. He’s not letting it go.
A Princeton survey earlier this year found that 29.9% of students admitted to cheating with AI on at least one exam. But this case offers a vivid look at what that actually means in practice — and how much AI might be replacing actual learning.
Brown recently released a provost-led report on AI in teaching and learning. It found professors aren’t the only ones who are worried.
