Berkeley Professor Stops Grading AI Homework
Homework used to be simple: the teacher assigned problems, the student worked through them, and the grade showed how much had been learned. At the University of California, Berkeley, one celebrated mathematics professor says that chain has broken.
Professor Richard Borcherds, a 1998 Fields Medal winner, has changed how he marks homework. He no longer grades it for correctness; students receive credit for turning it in, because AI has made it nearly impossible to tell whether a polished answer came from a student’s thinking or a chatbot.
When the Right Answer Proves Nothing
Borcherds explained his decision in an interview with The California Post, saying student use of AI for homework is now widespread. If a grade rewards only the final answer, the honest student who struggles through a hard problem can end up with a lower mark than a classmate who pasted the question into an AI tool.
In other words, the old system now punishes honesty. Why wrestle with a proof for hours, Borcherds argues, when a machine will produce the answer in seconds and earn the same, or better, grade?
He admits his replacement system is far from perfect. Awarding credit for completion keeps students engaging with the work, but it does not measure understanding. Borcherds himself has called it “not a very good system,” with no better alternative for now.
The Numbers Behind the Worry
Borcherds is not acting on a hunch. A survey of 1,017 American teenagers, published by Common Sense Media in August 2026, found 70% had used AI for schoolwork. Among those users, 63% said they got answers from AI — using them as-is, rewriting them, or blending them with their own knowledge.
Strikingly, many students sense the downside: among AI users, 38% said the tools led them to fewer of their own ideas, and 39% felt they missed out on learning. Yet 66% said AI helped them understand schoolwork — a reminder it can tutor as well as shortcut.
A separate study is sharper still. Researchers examining 30 months of data from 26,811 students in grades seven through twelve in China found AI adoption linked to an 18% rise in homework scores and a 30% drop in completion time — but monthly exam scores fell 20% within six months, concentrated among students who appeared to hand homework over to AI.
A Question Bigger Than Homework
For Borcherds, the worry now stretches beyond the classroom. He has watched AI systems grow rapidly more capable at mathematics itself. In September 2026, OpenAI announced that an internal system had produced a proof addressing a version of the Navier-Stokes existence and smoothness problem, one of the famous Millennium Prize Problems that had resisted mathematicians for decades.
That breakthrough left Borcherds wondering what remains uniquely human in his field. If AI keeps improving at its current pace, he warned, there may soon be nothing human mathematicians can do that AI cannot do better. He even asked ChatGPT for advice for young mathematicians — and it had no idea either.
He compares it with earlier upheavals — weaving displaced by looms, horses by motor cars — except, he argues, this is the first time such a shift has arrived in mathematics itself.
What Should Schools Do Instead?
Online, opinion is split. Some students praise homework as a learning experience without a grade on every answer; others argue teachers should change assessment rather than abandon grading.
The most popular suggestion is a return to old-fashioned testing: in-person, closed-book, paper-and-pencil exams where no chatbot can sit beside you. Several commenters said their departments have already shifted grading weight toward timed exams, letting students use AI on problem sets as long as they can prove understanding under exam conditions.
Borcherds doubts universities will transform overnight. But his quiet rebellion has started a loud conversation: if homework no longer shows what a student knows, what should replace it?
