According to Ars Technica, economics professor Roberto Serrano grew concerned that generative AI was inflating student performance in his Brown University class. He moved the final from take-home to in-person proctored exam. The result: scores fell roughly 50 percent, turning a private suspicion about AI cheating into public evidence that something is broken in how universities measure learning.
If take-home assignments and online exams can be outsourced to a chatbot, grades stop measuring understanding and start measuring who is best at prompt engineering under honor codes written for a different era. That noise ripples outward: employers, graduate schools, and honest students all get a weaker signal about what a degree actually certifies.
Universities are treating AI cheating as a discipline problem. It's an assessment problem. When a course can be passed with work a model generates on demand, the course is testing the wrong thing, or testing it the wrong way. Serrano's 50 percent drop forces a choice schools have dodged since 2026: redesign classes around live reasoning, drafts, oral defense, and supervised work, or keep handing out polished transcripts with less learning behind them. The first path is harder. The second path is dishonest.
Filed to the Technology desk · Jul 8