Brown University Professor Flags AI Cheating After Grades Plummet Without Tools

A professor at Brown University has raised alarms about the potential impact of artificial intelligence on academic integrity after a stark contrast in student performance between take-home and proctored exams. Roberto Serrano, an economics professor, noted an average score of 96 percent on a take-home exam, a significant increase from the historical average of 65-80 percent for the course. His suspicions grew when running the exam questions through ChatGPT yielded nearly identical answers, including a specific, convoluted mathematical proof that many students also presented.

To address these concerns, Serrano administered a proctored, in-person final exam. The results were dramatic: 18 students dropped the course, nine did not appear for the test, and the class average plummeted to 48.6 percent, the lowest in the course’s history. Only a few students came close to their take-home scores, with 19 failing outright.

Key Facts

Detail	Take-home Exam	Proctored Exam
Average Score	96%	6%
Course Drops	N/A	18
No-shows	N/A	9
Historical Avg.	65-80%	N/A

University Response and Broader Implications

Serrano voided the midterm and weighted the final exam at 80 percent of the course grade. He described the university’s response as “meek,” with administrators advising him to report individual cheating cases, a process he found “ridiculous.” He advocates for a stronger institutional stance, emphasizing the societal risk of students believing cheating is acceptable.

This incident at Brown University is not isolated. Two significant studies corroborate Serrano’s observations, indicating a pattern where reliance on AI for homework correlates with lower scores on proctored exams. One study from central China, tracking over 26,000 students, found that after six months of AI use, homework scores increased by 18 percent while completion time decreased. However, exam scores dropped by 20 percent, with long-term losses on entrance exams ranging from 18 to 24 percent. The study noted that 81 percent of long-term AI users exhibited this pattern, with top students experiencing the most significant performance decline (24 percent).

AI’s Impact on Grades and Learning

Another study from UC Berkeley, analyzing over 500,000 grades at a large Texas research university, revealed a 13 percentage point jump in A grades in writing and programming-heavy courses after ChatGPT’s launch. This effect was concentrated in unsupervised homework assignments, with courses heavily reliant on homework seeing a 16 percentage point higher increase in A grades compared to those emphasizing proctored exams. These findings collectively suggest that while AI tools can expedite homework completion and potentially inflate grades in unmonitored settings, they may hinder genuine learning and lead to poorer performance when students are tested without AI assistance.

Why This Matters for Indian Readers

For Indian educators, students, and policymakers, these findings underscore a critical challenge. As AI tools become more accessible, the integrity of academic assessments and the true measure of student learning are at stake. Institutions in India must consider adapting their assessment methods, promoting digital literacy, and fostering ethical AI use to ensure that educational outcomes reflect genuine knowledge and skill development. The experience at Brown University and the corroborating studies highlight the urgent need for a balanced approach to integrating AI in education, focusing on critical thinking and problem-solving rather than mere reliance on automated solutions.

Source: The Decoder, https://the-decoder.com/grades-dropped-from-96-to-48-percent-when-a-brown-professor-made-students-take-the-exam-without-ai/