An economics professor at Brown University has presented what many consider irrefutable evidence of mass AI cheating in higher education. Roberto Serrano, a seasoned instructor who has taught advanced economics for decades, watched in disbelief as his take-home midterm exam scores soared to an average of 96 out of 100—a number far beyond anything he had ever seen. When he subsequently switched the final exam to an in-person format, the class average plummeted to 48. The disparity, he argues, can only be explained by widespread use of generative AI tools like ChatGPT.
The Circumstances That Triggered the Experiment
Serrano's decision to offer take-home exams was born from tragedy. In December 2023, a gunman killed two students on the Brown University campus, leaving the community shaken. Many students expressed anxiety about sitting exams in crowded rooms, and Serrano sought to accommodate them by offering take-home midterm and final papers. It was the first time in decades that he relaxed his standard in-person testing protocol. The irony, as he later told reporters, stings deeply: the one time he loosened the rules in the name of compassion, a significant portion of the class chose to cheat.
His course, ECON 1170, is an advanced undergraduate economics class that typically draws a small, dedicated group of students. In previous years, enrollment had never exceeded 30, and on one occasion it was as few as eight. This semester, however, enrollment swelled to 86 students. Serrano believes the new take-home format attracted many who might otherwise have avoided a challenging upper-level course. The expanded class size itself was a warning sign, but the results were far more dramatic.
The Numbers That Exposed the Fraud
The midterm results were, in Serrano's own word, extraordinary. The class averaged 96 out of 100, with 40 students achieving a perfect score. Historical averages for the same course typically range between 65 and 80, and this particular midterm was designed to be more difficult than previous iterations. Serrano had reasoned that with unlimited time at home, students could delve deeper into the material and produce more thoughtful answers. Instead, he encountered a pattern that defied normal variation.
Beyond the raw scores, the quality of the answers themselves raised red flags. Many correct responses exhibited what Serrano described as a “very convoluted style”—verbose, error-free, but mechanically articulate in a way that seemed unnatural for human writing. To test his suspicions, Serrano and his teaching assistants fed the same exam questions into ChatGPT. The outputs were strikingly similar to the answers submitted by students. This was not mere coincidence; it was evidence of systematic outsourcing of cognitive work.
Setting a Trap: The In-Person Final
Armed with his suspicions but wanting to confirm them beyond doubt, Serrano designed a simple experiment. He announced to the class that the final exam would be held in person, under traditional proctoring conditions. He explicitly told students that he would compare the distribution of scores between the midterm and the final. If the two distributions matched, he would keep the midterm grades. If they differed significantly, he would void the midterm and reweight the final to count for the entire course grade.
The response was immediate and telling. Eighteen students dropped the course entirely, and nine more simply skipped the final exam without official withdrawal. Among those 27 students, El Pais reported that 22 had scored a perfect 100 on the midterm. The implication is clear: they knew they could not replicate that performance without AI assistance. Among the students who did sit for the in-person final, the average score fell from 96 to 48—a drop of exactly 48 points. By Serrano's conservative count, at least 50 of the 86 students cheated on the midterm. He calls the evidence overwhelming and has made the story public through interviews with El Pais and Inside Higher Ed, refusing to let the issue fade.
A Wider Reckoning in Academia
Serrano's case is a microcosm of a crisis that has been building since the public release of ChatGPT in late 2022. Universities across the United States and around the world are grappling with how to maintain academic integrity when powerful AI tools can complete assignments in seconds. A recent survey of Princeton students found that 29.9% admitted to cheating on at least one exam or assignment, with the vast majority of those cases involving AI. At Brown itself, a provost-led report revealed that most undergraduates use generative AI on a weekly or even daily basis. Yet the same report found that large majorities of students worry about the effect of such tools on their own learning, and fear a degradation of their “cognitive capacity.”
The issue extends beyond the classroom. Employers are increasingly wary of graduates who may have earned degrees without actually mastering the material. If Serrano’s class is any indication, the gap between apparent knowledge and actual understanding could be as wide as 48 points—a chasm that undermines the value of a university education. As AI reshapes hiring practices and workforce expectations, the ability to think critically and independently becomes more precious, yet harder to verify.
Why This Matters: The Societal Dimension
Serrano frames the problem in the starkest possible terms. “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay,” he told Inside Higher Ed. “That leads to a declining society, to a failed society. We cannot choose to become idiots.” His words reflect a growing alarm among educators that the widespread use of AI for academic dishonesty is not just a grading problem, but a threat to the foundational purposes of higher education: the development of critical thinking, reasoned argument, and independent inquiry.
The broader implications are unsettling. If students become accustomed to using AI to shortcut intellectual effort, they may never develop the cognitive muscles needed to solve complex problems, innovate, or lead. The phenomenon is not limited to economics courses; it spans disciplines, from humanities to engineering. Some universities have responded by banning AI outright, while others have chosen to integrate it into their pedagogy, teaching students how to use AI ethically as a tool rather than a crutch. But Serrano’s experiment suggests that when given the choice, a substantial number of students will opt for the easy route, even at the expense of their own education.
Historical Context and the Evolution of Academic Dishonesty
Cheating in academia is hardly new. Students have plagiarized, copied, and collaborated illicitly for as long as formal education has existed. But AI represents a quantum leap in both the ease and undetectability of cheating. Earlier technologies—such as essay mills or answer keys—required human effort or payment. ChatGPT can generate a passable essay in seconds for free. The result is that the temptation to cheat has never been stronger, and the ability to detect it has never been more difficult.
Brown University has not been passive. The administration has invested in detection software, revised honor codes, and encouraged faculty to redesign assessments to minimize AI vulnerability. Yet Serrano’s experience shows that these measures may be insufficient, especially when instructors try to be compassionate. The shift from in-person to take-home exams, even with good intentions, opened the door wide.
Other institutions are also experimenting with new approaches. Some have returned to oral exams, in-class timed writing, or project-based assessments that require original synthesis. Others have embraced AI by assigning tasks that explicitly involve AI but demand human oversight and critique. There is no consensus, and Serrano’s stark data point adds urgency to the search for solutions.
Ultimately, the Brown professor’s small experiment—one class, one term—turns a fuzzy anxiety into a concrete number. Take the AI away, and half the apparent knowledge disappears. That is the figure universities now have to sit with. As they ponder the future of assessment, they must also consider the future of learning itself. If the goal is to produce educated citizens capable of independent thought, then the presence of AI in the classroom must be managed with both vigilance and imagination.