Back to feed
News Story
THE DECODERT2
1 sources

Grades dropped from 96 to 48 percent when a Brown professor made students take the exam without AI

Original

An economics professor at Brown University suspected most of his 86 students used AI to cheat on a take-home exam, which averaged 96%. When he administered an in-person final, 18 students dropped the course, nine didn't show up, and the average plummeted to 48.6%. Two large studies from China and UC Berkeley corroborate that students relying on AI for homework see significant drops in proctored exam scores.

SynthePulse Insight · AI deep reading

AI Cheating Suspicions: Open-Book Exam Average Plummets from 96% to 48% After Switching to Proctored Mode

Version 2 · 1 source

A Brown University economics professor found that the average score on an open-book midterm was 96%, but after switching to an in-person proctored final, the average dropped to 48.6%, with 18 students dropping the course. Two large-scale studies confirm a pattern of inflated homework scores and declining exam scores.

  • Brown University economics professor Roberto Serrano found that the open-book midterm average was 96%, far exceeding the historical range of 65-80%, and ChatGPT gave nearly identical answers to students.
  • When the final was changed to proctored, the average dropped to 48.6%, a historic low; 18 students dropped the course, 9 didn't show up, and 19 failed.
  • Two studies support this pattern: a Chinese study (26,000 students) found that after using AI, homework scores rose 18% and exam scores fell 20%; a UC Berkeley study (500,000 grades) found that after ChatGPT's release, the proportion of A grades jumped by 13 percentage points.
  • The professor criticized the university's 'weak' response, requiring him to report cheating cases one by one.

The Brown University Case

Roberto Serrano, an economics professor at Brown University, grew suspicious when his 86 students averaged 96% on an open-book midterm. Historically, the course average had been 65–80%. He input the exam questions into ChatGPT and got nearly identical answers, including complex mathematical proofs many students had used.

After warning students, Serrano changed the final to a proctored exam. The results were dramatic: 18 students dropped the course, 9 did not show up, and the average fell to 48.6%—the worst in the course's history. Only a few students scored near their open-book performance, and 19 failed outright. Serrano voided the midterm and set the final's weight to 80%.

Serrano described the university's response as 'weak', requiring him to report cheating cases individually. He advocated a stronger stance, warning that tolerating cheating leads to a 'failing society'.

Studies Corroborate

Two recent studies support Serrano's experience. A study in central China tracked 26,000 students in grades 7–12 for 30 months. After six months of AI use, homework scores rose 18%, completion time dropped from 64 to 45 minutes, but exam scores fell 20%, with long-term losses of 18–24% on entrance exams. About 81% of long-term users fit the pattern: faster homework, high homework scores, low exam scores; top students lost 24%.

The UC Berkeley study covered over 500,000 grades at a large Texas research university and showed that in courses with more writing and programming assignments, the proportion of A grades jumped by 13 percentage points after ChatGPT's release. The increase was concentrated in unsupervised assignments; courses relying on homework saw a 16-percentage-point larger increase than those relying on proctored exams.

Broader Impacts and Institutional Responses

The convergence of Serrano's case and large-scale studies suggests that AI-assisted cheating is widespread in unsupervised assignments and significantly undermines the validity of grades. However, the exact proportion of students using AI remains unclear, and whether the pattern is consistent across disciplines and educational stages is unknown.

The professor's call for stronger institutional responses highlights the tension between trusting students and maintaining academic integrity. Reporting cheating individually is resource-intensive and may be ineffective if the problem is systemic. The studies also raise concerns about long-term learning impacts: students relying on AI to complete assignments may fail to develop key skills, and the Chinese study found significant declines in entrance exam scores.

Uncertainties remain: the studies are observational and do not prove causation; the Chinese study is limited to one region; Berkeley data may not reflect the latest AI models. Nonetheless, the evidence points to an urgent need for educational institutions to reassess assessment design and integrity policies.

Credibility boundary

Main source is a The Decoder report citing Inside Higher Ed and two peer-reviewed studies. The professor's account is first-hand, and the studies are large-scale and published. However, the cheating cases are suspected, and the studies show correlation, not causation.

Insight takeaway

The Brown University incident, supported by two large-scale studies, strongly indicates that students using AI on unsupervised assignments inflate homework scores and depress proctored exam scores, posing a serious challenge to academic integrity that requires a systemic institutional response.

Sources for this version

  1. Grades dropped from 96 to 48 percent when a Brown professor made students take the exam without AI

    THE DECODER

Primary report

THE DECODERT2

Primary source