THE CRUNCH
A Fields Medal-winning mathematician at UC Berkeley has stopped setting graded homework, saying widespread AI use has made it pointless. Richard Borcherds told the New York Post that it is now nearly impossible to distinguish a student's own submission from one produced with AI, so he grades primarily through in-person exams with no electronic devices allowed.
"Online exams are now a waste of time as several students just use AI to answer the questions," Borcherds said, noting that while he cannot tell who is using AI, average scores rise sharply in online settings. Grading homework for correctness, he argued, gives students a strong incentive to lean on AI and penalises those who are honest.
The pattern is not unique to Berkeley, if the wider evidence holds. Earlier this year a Brown University professor recorded an average of 96% on take-home midterms, well above the historical range of 65% to 80%. When he moved the final exam in person, only two students came within 10% of their midterm grades, and just one did better. Researchers have concluded that unmonitored online exams no longer reflect what students have actually learned, even though AI use has not been definitively proven in every case.
UC Berkeley has examined AI-detection software but found it unreliable, so it leaves AI policy to individual professors, provided they state their rules in course syllabi. Borcherds, for his part, says that if the worst AI does is complicate homework grading, he will be "very relieved".