'We cannot choose to become idiots': AI cheating probe follows sharp drop in Brown University exam scores

A Brown University economics professor said he uncovered widespread suspected use of generative artificial intelligence by students after unusually high scores on a take-home midterm exam prompted a comparison with ChatGPT-generated answers.

Professor Roberto Serrano said the issue arose during his advanced undergraduate welfare economics course in late 2025 and early 2026 after the university allowed changes to exam arrangements following a mass shooting on campus in December 2025, says Cybernews.

Students had expressed concerns about returning to physical classrooms after the shooting, and Serrano allowed them to complete the midterm exam from home. Enrollment in the course rose to 86 undergraduate students, compared with the typical class size of about 30.

The take-home midterm produced an average score of 96%, well above the historical average of between 65% and 80% for the course.

The unusually high results led Serrano and his teaching assistants to compare student responses with answers generated by ChatGPT. They found similarities that they believed indicated the use of AI.

Among the examples cited was a mathematical proof question in which many students used a contradiction argument, matching ChatGPT's response, even though a direct argument was the more conventional solution.

With approval from the university administration, Serrano required students to sit the final examination in person. He told students that if their final exam performance was consistent with their midterm score, the midterm grade would remain. Otherwise, the midterm would be voided and the final examination would account for their entire course grade.

Following the announcement, 18 students dropped the course, while another nine remained enrolled but did not take the final examination.

The in-person final produced an average score of 48.6%, the lowest recorded for the course. Three students received a score of zero, and only a small number of students achieved results comparable with their midterm performance.

Serrano said the case illustrates the growing challenge universities face as generative AI tools become more widely available and capable of producing convincing academic work.

"We cannot choose to become idiots," Serrano said, arguing that a culture that accepts cheating leads to institutional decline.

The case reflects broader concerns about the use of AI in higher education. A 2025 survey at Princeton University found that 27.7% of senior students said they had used ChatGPT in unauthorized ways, an increase from the previous year.