exclusive details
See more of our coverage in your search results.
A renowned UC Berkeley mathematician who stopped grading homework for correctness because of rampant AI use has revealed he is now using in-person exams with no electronic devices allowed to judge how well his students are doing.
Richard Borcherds, who won the prestigious Fields Medal in 1998, said he has no better solution as artificial intelligence makes it nearly impossible to tell class submissions from a chatbot’s.
“I now base grades mainly on in-person exams with no electronic devices allowed,” Borcherds told The California Post.
“Online exams are now a waste of time as several students just use AI to answer the questions. I cannot tell which students are using AI, but I notice the average score increases a lot in online exams.”
Research backs up what the professor has seen in his own classroom.
A 2024 study published in the journal Discover Education by Alexander Stanoyevitch analyzed years of exam scores from an introductory statistics course he teaches every semester, before and after ChatGPT arrived. It found that scores rose significantly when exams of similar difficulty moved from in-person to online.
A separate study published in the journal Teaching of Psychology compared two sections of an undergraduate personality psychology course.
In one, 153 students took the final exam in person. In the other, 160 students took it online without monitoring. The online group scored significantly higher.
Whether the exam was online or in person predicted students’ final scores better than how they had done earlier in the course.
The researchers concluded that unmonitored online exams inflate scores and may not reflect students’ true abilities.
Borcherds said telling human work from AI-generated work has become increasingly difficult.
AI used to give itself away with a few characteristic errors, he said, but the technology has improved so much that such mistakes are now rare.
That’s why he stopped grading homework on whether the answers are right. Students now get credit mainly for completing their assignments.
“Grading for correctness just adds a huge incentive for them to use AI for homework instead of doing it themselves, and penalizes students for honesty. And paying [readers/graders] to grade AI generated answers is a waste of time and money,” Borcherds said.
UC Berkeley requires instructors to spell out their AI rules in course syllabi.
The university looked into AI-detection software, including Turnitin and Rumi, but found the tools weren’t reliable enough to adopt campus-wide, the Daily Californian reported. That leaves individual professors to set their own policies.
Borcherds, however, remains philosophical about the bigger picture.
“If the worst effect of AI on society is that it is harder to assess homework then I will be very relieved,” he said.