Student who turns in a perfect mathematics assignment may not really understand the solution. Richard Borcherds, a professor at the University of California, Berkeley, has stopped grading homework for accuracy after noticing that students using artificial intelligence were routinely getting top marks. His decision raises questions about the measurement of learning in universities, as artificial intelligence is able to do difficult assignments in seconds.
In an interview with The Daily Californian, published October 23, 2026, Borcherds, winner of the prestigious Fields Medal in 1998, explained his choice. He believes there is “no point” in having graders assess responses that students can easily create with artificial intelligence. Now, instead, students are credited for turning in their homework.
Why We Changed the Way AI Grades
Borcherds concedes that “completion-based grading is not a very good system,” but he has not found a good alternative yet. But he’s worried about more than students just using chatbots to do their work. Even mathematically correct submissions can be insufficient evidence of a student’s ability to independently understand a problem, solve a problem independently, or explain their reasoning.
Research at UC Berkeley adds another dimension to the concern. According to a study from October 9, 1,222 people did better on some math tasks after receiving AI help at first. But they had problems with accuracy and persistence when the help was taken away. Borcherds’ students were not involved in the experiments, but their results suggest a difference between successfully completing an assignment, and retaining the ability to solve similar problems independently.
The bigger issue for UC Berkeley is AI in education.
Borcherds does not have a university-wide grading policy; grading is a teaching decision made on a case by case basis. Instructors at UC Berkeley can set their own expectations for AI use, and the university recommends that course syllabi include explicit guidelines. Its teaching resources also talk about oral or written follow-up assessments as a means of checking student understanding.
Concerns have also been raised about the future of maths. Borcherds says he is no longer sure what advice to give to students who are considering careers in the field, especially as A.I. systems get better at doing advanced mathematical research.
His decision does not imply that homework has lost its educational merit. Students may not be graded on accuracy or completion but it is still helpful for students to work on difficult problems. Universities are finding it hard to design assessments that can tell the difference between real understanding and answers produced by technology.