An economics professor at Brown University suspected widespread AI use after students averaged 96 percent on a take-home exam, The Decoder reported. When the final was moved to an in-person format without AI, the average dropped to 48 percent.
The case illustrates how difficult AI has made assessment design. Take-home work can measure research and synthesis, but it can also be hard to separate student understanding from model assistance.
Universities are likely to keep experimenting with exams, oral defenses, supervised work, and explicit AI-use policies as generative tools become normal study aids.