Brown University Professor Uncovers Widespread AI Cheating After Record Midterm Scores
Key Takeaways
- ▸Generative AI tools like ChatGPT enable academic cheating at scale, as evidenced by 96% average scores on a traditionally 65-80% average exam
- ▸Traditional take-home exams are insufficient protection against AI assistance; real-time proctoring may become necessary for high-stakes assessments
- ▸Universities lack standardized protocols for detecting and responding to AI-assisted cheating, with some administrators described as having inadequate responses
Summary
An economics professor at Brown University discovered that dozens of students in his welfare economics class had likely used ChatGPT to cheat on a take-home midterm exam, resulting in an unprecedented 96% class average—significantly higher than the historical 65-80% average. Professor Roberto Serrano suspected the suspicious uniformity of high scores when he ran the answers through ChatGPT and found nearly identical responses, identifying a distinctive 'convoluted style' that didn't match human writing patterns. He moved the final exam to an in-person format, which exposed the cheating: 18 students dropped the course, 9 didn't sit for the final, and the class average plummeted to 48.6%—the lowest in the course's history. The incident highlights universities' vulnerability to AI-assisted academic dishonesty and raises critical questions about how institutions should adapt their assessment methods and policies in the age of advanced AI tools.
- AI detection patterns emerged clearly when in-person exam results (48.6% average) starkly contradicted midterm performance, confirming the professor's suspicions
- Academic integrity policies and faculty training urgently need updating to address AI-enabled dishonesty before it becomes normalized across campuses
Editorial Opinion
This incident reveals a fundamental crisis in higher education's assessment infrastructure. While generative AI tools like ChatGPT represent genuine intellectual advancement, their accessibility has outpaced universities' ability to maintain academic integrity. The stark gap between the 96% midterm average and 48.6% final exam average isn't just damning evidence of cheating—it's a flashing warning light that traditional honor systems and remote testing models are now obsolete. Universities must urgently develop new policies, invest in AI literacy for faculty, and potentially redesign how they evaluate student learning, lest widespread cheating become the path of least resistance.
