UC Berkeley mathematics professor Richard Borcherds, a 1998 Fields Medal winner, has stopped grading student homework for correctness. He now awards credit mainly for whether assignments are turned in on time, after concluding that artificial intelligence tools have made traditional marking unreliable.
Borcherds, 66, told the Daily Californian in a sit-down interview published October 23 that student use of AI for homework has become widespread. He said grading answers for accuracy creates a strong incentive for students to rely on tools instead of working through problems themselves and unfairly penalizes those who complete the work honestly. Paying graders to review submissions that may be AI-generated is, in his view, a waste of time and money.
“There is no easy way to tell the difference between human and AI-generated work,” Borcherds said. He noted that earlier AI outputs sometimes contained characteristic errors that gave them away, but the technology has improved enough that such mistakes are now rare. He described his current approach of grading solely on completion as imperfect but the only practical option he sees for now.
Most of his students’ final grades now rest on in-person exams where electronic devices are banned. Online exams, he told The California Post, have become “a waste of time” because average scores rise sharply when students can use AI without supervision. He cannot reliably identify which students are using the tools, only that the overall results improve noticeably.
Borcherds was born Richard Ewen Borcherds on November 29, 1959, in Cape Town, South Africa. His family moved to Birmingham, England, when he was an infant. He studied at Trinity College, Cambridge, under John Horton Conway, earned his doctorate in 1985, and has held faculty positions at both Cambridge and UC Berkeley. He joined the Berkeley mathematics department in 1993 and is a Fellow of the Royal Society. His Fields Medal recognized work on vertex algebras, the proof of the monstrous moonshine conjecture, and related advances in algebra and mathematical physics.
The interview with the Daily Californian came shortly after OpenAI announced that its systems had produced a claimed solution to the Navier-Stokes existence and smoothness problem, one of the long-standing Millennium Prize challenges. Borcherds said that if AI continues improving at its current rate, within a few years there may be little that human mathematicians can do better than machines. He added that he no longer knows what career advice to give students considering mathematics, a field he once viewed as relatively insulated from automation. When he asked ChatGPT the same question, the system offered no clear guidance either.
UC Berkeley requires faculty to include statements in their syllabi about expectations for AI use in coursework, according to Anne Marie Richard, assistant vice provost for undergraduate education. Actual policies vary widely by course and instructor. The campus has tested detection tools including Turnitin and Rumi but determined they are not reliable enough to serve as a standardized solution. Individual professors retain discretion over how they handle AI in their classes.
Borcherds compared the situation to earlier technological shifts that rendered certain human skills obsolete, such as hand weaving replaced by looms or horses displaced by automobiles. He observed that this is the first time such a displacement has appeared inside mathematics itself. He expects the academic system to change slowly, noting the continued existence of physical paper journals long after they became largely unnecessary. The shift in his own grading practices is one professor’s response to conditions that many university instructors across disciplines are now confronting.
