After the midterm exams, Professor Roberto Serrano of Brown University communicated with his students, expressing his suspicion that many had resorted to artificial intelligence to cheat. As a result, he announced that the final exam would be conducted in person.
For the first time in nearly twenty years of teaching his Welfare Economics and Social Choice Theory course, Serrano opted for a take-home midterm exam this spring. He made this decision partly due to elevated student anxiety stemming from a tragic mass shooting in December that claimed two lives and left several injured at the university. Given these circumstances, he determined that it would be more appropriate for students to complete their exams from home.
However, by the semester’s end, Serrano felt regret over this choice. He observed that numerous students had potentially exploited AI technologies to receive perfect or nearly perfect scores on the midterm. Consequently, he decided to hold the final exam in person. This change led to over a dozen students withdrawing from the course and many others failing. Serrano criticized the administration's response to the cheating incident as “meek” and highlighted the pressing need for universities to effectively address AI-assisted cheating on a large scale.
Serrano's welfare economics class typically enrolls around 30 students, but this spring saw an increase to 86, a rise he links to the promise of take-home exams. When the midterm results arrived, the average score reached an astonishing 96 percent, raising alarms for Serrano. He noted, “Historically, the midterm average has ranged between 65 and 80 percent, and this exam was actually tougher than those in previous years.”
Suspecting something was amiss, he and his team analyzed midterm answers using ChatGPT, revealing striking similarities in the responses. The AI produced answers that, while somewhat correct, lacked the clarity expected from human students and often employed convoluted reasoning. For instance, when asked to prove a math statement, many students chose a "contradiction argument," which was technically correct but unnecessarily complicated compared to a straightforward approach.
In a communication to his class, Serrano indicated that he believed a number of students had used AI to cheat and, with the support of his dean, transitioned the final exam to an in-person format. He wrote to students that he would reserve judgment on the midterm scores for now, stating, “If the distribution of the final exam closely resembles that of the midterm, I may accept the midterm scores. Otherwise, as I anticipate, I will void the midterm and adjust the weight of the final exam accordingly.”
Despite his efforts, engagement from students was minimal, resulting in 18 withdrawals and nine additional students who did not take the final exam. The results aligned with Serrano's concerns: three students received a zero, and the average final score plummeted to 48.6 percent, a significant drop from previous years where it had never dipped below 65 percent. Only a few students maintained their midterm performance on the final exam.
Following these developments, Serrano informed the class that the midterm would be voided, adjusting the final exam's weight to 80 percent of the total grade. To facilitate passing, any student scoring 40 percent or higher on the final would receive a passing grade, compared to the previous threshold of 50 percent. Ultimately, 19 students did not pass the course.
In May, Serrano shared his findings with Brown’s Standing Committee on the Academic Code but received no immediate reaction. After going public with his experience in June, the committee requested that he lodge formal complaints against each suspected cheating student, including documentation of their exams. Serrano criticized this request, pointing out the reliance on AI-detection tools that often produce unreliable results.
A representative from Brown University, Brian Clark, stated that the institution takes allegations of academic integrity very seriously, asserting that the response procedures remain consistent regardless of the number of students involved. He noted that Serrano had not yet provided the requisite details for the committee to proceed.
The challenge of addressing widespread cheating is significant for faculty, as explained by Tricia Bertram Gallant, director of the Academic Integrity Office at the University of California, San Diego. She indicated that faculty members are not incentivized to report or prevent cheating due to a lack of compensation for their time spent on such matters.
Colleges must prioritize both fairness and efficiency when handling cheating allegations. Bertram Gallant recommends allowing students to acknowledge responsibility for cheating via email and enabling professors to submit complaints for multiple students simultaneously. She emphasizes that universities serious about upholding academic integrity should invest in dedicated staff and resources.
As the situation unfolded, a committee at Brown was also exploring the use of generative AI within the academic environment, aiming to adapt policies to this technological advancement. Their initial report indicated that a significant number of faculty members are concerned about AI-assisted cheating and recommended revisions to academic codes to address these challenges.
The report highlights the need for discussions surrounding the ethical use of AI in academic settings to evolve. It suggests that enforcing explicit guidelines on integrity related to AI while moving the conversation beyond punitive measures would foster a more open dialogue.
Amid these challenges, Serrano reflected on the long-term implications of academic dishonesty, stating, “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is acceptable. That leads to a declining, failed society. We cannot choose to become ignorant.”
