A UC Berkeley mathematics professor has stopped grading his students’ homework for correctness, saying there’s little point after watching artificial intelligence tools help them rack up top scores. Richard Borcherds, a Fields Medal winner who currently teaches MATH 250A, a graduate-level course on abstract algebra, now hands out credit simply for turning assignments in.
According to the Daily Californian, Berkeley’s student newspaper, Borcherds made the switch after seeing how many of his students used AI to consistently receive top scores. He currently grades on completion if a student hands in their homework, the paper reports, rather than evaluating whether the answers are actually right.
Borcherds, who won the 1998 Fields Medal for his contributions to mathematics, isn’t pretending the new system is a fix. He has acknowledged that it is “not a very good system,” the Daily Californian reports, but said that at the moment, he has no alternative.
No Campus-Wide Rulebook for AI
Berkeley doesn’t hand professors a single playbook for handling AI in the classroom. The Berkeley Division of the Academic Senate asks that faculty include a statement about course expectations, including the use of AI tools for coursework, in their syllabi — but beyond that, AI policy for classes is up to the discretion of faculty, per the student newspaper's reporting. That means the direct policies on AI use can vary dramatically with course level and focus area, with one professor's abstract algebra seminar looking nothing like another department's writing-intensive lecture.
Berkeley has also looked at whether software could catch AI-generated work for instructors automatically, and the answer so far has been no. The campus explored AI-detection software including Turnitin and Rumi and determined these tools aren’t adequate enough to be standardized on campus, the same report notes. That decision tracks with Berkeley's own Research, Teaching, and Learning Commons, which said that based on the findings from a pilot program, along with numerous publications and studies, UC Berkeley has opted out of Turnitin's AI-writing-detection tool.
A Bigger Worry Than Cheating
For Borcherds, the homework policy is almost a side issue next to a deeper question hanging over his field. He has said that if things go like this, in a few years, humans doing math research will be redundant, according to a separate Daily Californian report on the Berkeley math department. He has gone further still, telling the outlet that in a few years, there'll be nothing human mathematicians can do that AI can't do better — and he has claimed AI is already better than beginner graduate students.
That prediction landed with extra weight after OpenAI announced that its AI agents had apparently solved the Navier-Stokes existence and smoothness problem, one of mathematics' major unsolved challenges, using approximately 10,000 AI agents, the Daily Californian reported. The announcement left Berkeley's math community, in the paper's framing, at a crossroads over what role is left for human researchers.
Borcherds has openly admitted he doesn't know what to tell students weighing a future in his field. “I have no idea what to advise,” he said, per the Daily Californian's reporting — a striking admission from a mathematician who holds one of the discipline's highest honors.
Borcherds Sees Parallels in Past Disruptions
Borcherds has drawn comparisons between the current moment and earlier technologies that wiped out entire lines of work, likening AI's potential impact on mathematics to the way machines displaced hand weaving and horses. He's also pointed out that institutions don't always move quickly even after their old purpose has faded, comparing the situation to academic journals that kept printing physical editions long after losing much of their original function.
Even so, he doesn't expect math departments to transform overnight. He has suggested that universities might not adapt immediately to rapid AI advances, predicting that departments could largely operate as they do now for a couple years even as the technology keeps improving. He's noted that just five years earlier, mathematics looked relatively insulated from the disruption hitting other professions — a sense of safety that has since evaporated. He even asked ChatGPT what would happen to the field and got no clear forecast back.
Part of a Wider Pattern Beyond Berkeley
Borcherds's dilemma fits into a much larger shift playing out at colleges nationwide. A survey of more than 6,600 students at seven UK universities found that 32 percent admitted some level of unpermitted AI use in assessments, according to the Financial Times. The same reporting noted that AI-detection tools have come under fire over their reliability, specifically regarding false positives and bias, pushing some schools to abandon them altogether.
Those accuracy concerns aren't abstract. MIT's Sloan Office of Educational Technology has reported that across seven AI detectors, 61 percent of TOEFL essays written by non-native English speakers were wrongly flagged as AI-generated, against about 5 percent of essays by native English writers. A separate 2023 study cited by Berkeley's own Research, Teaching, and Learning Commons found that more than half of non-native English writing samples were misclassified as AI-generated by seven commonly used GPT detectors.
Faculty elsewhere are responding by redesigning how they test students rather than policing them after the fact. Colleges are pushing professors to revamp assessments to keep AI out, according to Inside Higher Ed. A fall 2025 Pew Research Center survey found that 54 percent of 13- to 17-year-olds said they had used AI chatbots to help with schoolwork, and 59 percent said students at their school used chatbots to cheat somewhat or very often, according to the San Francisco Standard. That same reporting described a San Francisco high school English department head who stopped assigning take-home essays after papers appeared to have been written by AI, limiting essay writing to handwritten classroom work instead.
The tension over whose rules apply isn't confined to California either. At SUNY Geneseo, provost Mary Toale affirmed that a course's instructor is the one who decides whether, and in what capacity, AI is acceptable for students to use, according to the Chronicle of Higher Education. The school reportedly asked an AI tutoring company to clarify to 224 students that professors' AI policies apply in every course, after the company's CEO suggested otherwise — part of a SUNY-system pilot program that began in 2024, from which faculty have the option to opt out.
Berkeley's own guidance underscores that detection software was never meant to settle these disputes on its own. The campus has said that its AI-detection tool provides no conclusive proof and that its use in academic-integrity investigations remains unclear, according to Berkeley's Research, Teaching, and Learning Commons. That caution echoes the American Association of Colleges and Universities, which advises, per the Financial Times, that faculty should never use AI detection as “hard evidence” or “smoking gun proof.”
For now, Borcherds's workaround leaves a graduate-level algebra class graded almost entirely on whether work shows up, not whether it's right — an uneasy stopgap in a field he suspects may look very different within just a few years.
