185 open opportunities · 21 added this week · 2 closing within 7 days See what's open
Back to all news

UC Berkeley Mathematics Professor Stops Grading Homework for Correctness as AI Changes Education

Richard Borcherds, a mathematics professor at the University of California, Berkeley and a winner of the prestigious Fields Medal, has stopped grading homework for correctness after seeing students use artificial intelligence to achieve high scores. His decision highlights a growing challenge for universities: how to measure students' understanding when AI can produce convincing solutions to difficult problems.

UC Berkeley Professor Changes Homework Grading Because of AI

A mathematics professor at the University of California, Berkeley, has changed how he grades homework after concluding that the widespread use of artificial intelligence has made traditional marking less meaningful.

Richard Borcherds, a professor of mathematics who received the Fields Medal in 1998, told The Daily Californian that he no longer sees much value in paying graders to assess homework solely on whether answers are correct.

According to the student newspaper's report, Borcherds noticed that many students were using AI tools to obtain consistently high scores on assignments. This made it increasingly difficult to determine whether their work demonstrated genuine understanding or reflected solutions generated with the help of artificial intelligence.

Under his current approach, students receive credit for completing and submitting their homework rather than being graded on correctness.

Borcherds acknowledged that the arrangement is not a good long-term solution, but said he had not found a better alternative.

His experience illustrates a problem now confronting educators: homework can still be completed and submitted, but the resulting score may no longer provide a reliable picture of what a student has learned independently.

Why Traditional Homework Is Becoming Harder to Assess

For decades, homework has played an important role in mathematics education. Students solve problems, practise techniques and demonstrate their understanding before taking examinations.

Teachers use the results to identify weaknesses and give feedback, while students use the exercises to prepare for more advanced topics.

AI tools have changed that process. Students can now ask chatbots to solve mathematical problems, explain formulas, generate proofs and provide step-by-step answers.

These tools can be useful for learning. A student who struggles with a concept can ask for another explanation or work through an example.

The difficulty arises when students submit AI-generated answers without understanding the underlying mathematics.

A correct solution may look similar whether it was produced independently or generated by an AI tool. As the tools improve, it becomes harder for a lecturer to determine the difference from the submitted answer alone.

Borcherds' decision reflects this problem. If students can use AI to produce highly accurate homework, grading the final answers may reward the use of the tools rather than the understanding the assignment was intended to measure.

He also expressed concern that students who complete their work independently could be disadvantaged when their classmates rely heavily on AI.

A Bigger Question: What Is the Future of Mathematics?

Borcherds' concern goes beyond marking homework. He has questioned how artificial intelligence might affect the future of mathematical research and whether universities should continue encouraging students to pursue careers in the field in the same way they have in the past.

The concern has become more pressing as AI systems demonstrate stronger capabilities in complex mathematical tasks.

In September 2026, OpenAI announced that an internal AI system had produced a proposed solution to the Navier–Stokes existence and smoothness problem, one of the Clay Mathematics Institute's Millennium Prize Problems. OpenAI published a written proof and a formalisation in Lean alongside its announcement.

The problem concerns the mathematical behaviour of equations used to describe fluid motion. It has remained one of the most difficult open challenges in mathematics for decades.

OpenAI's announcement does not mean every mathematical problem has been solved by AI, nor does it establish that human mathematicians are no longer needed. However, it has added to the debate about how quickly AI capabilities are progressing and what that could mean for research and education.

Borcherds suggested that if AI continues improving at its current pace, machines could eventually outperform human mathematicians across a much wider range of tasks.

He has also expressed uncertainty about what advice to give students considering a future in mathematics. His remarks reflect the difficulty of making reliable career predictions when the technology is advancing so rapidly.

Universities Are Still Working Out How to Respond

Borcherds' experience is part of a wider challenge for higher education institutions. Universities must decide whether to prohibit AI use in certain assignments, allow it with conditions or redesign assessments to measure learning in different ways.

At UC Berkeley, expectations for AI use can vary by course. Instructors communicate their policies to students, and some courses permit certain forms of assistance while others restrict it.

The university has also taken a cautious approach to AI-detection software.

According to UC Berkeley's Research, Teaching and Learning Commons, the university tested Turnitin's AI-writing detection tool in a pilot programme. The results were inconsistent, leading the university to decide against enabling the feature campus-wide.

This matters because AI detectors cannot always establish whether a student has used AI simply by examining their submitted work. Relying on uncertain detection results can create additional problems for both students and instructors.

The challenge for universities is therefore twofold: they must respond to AI-assisted work while avoiding methods that incorrectly accuse students or fail to measure their understanding.

What Could Replace Traditional Homework Grading?

Borcherds' approach is one response, but it does not solve the wider problem of assessing students fairly.

Universities and lecturers could explore several alternatives.

More supervised examinations: In-person assessments without access to AI tools can help lecturers evaluate what students know and can do independently.

Oral examinations: Asking students to explain a solution and answer follow-up questions can provide evidence of their understanding.

Project-based assessment: Longer projects can require students to document their decisions, explain their methods and demonstrate how they reached their conclusions.

Transparent AI-use rules: Courses can specify when AI is allowed, what students must disclose and which parts of an assignment must be completed independently.

Assessment redesign: Instead of focusing entirely on the final answer, instructors can assess the reasoning, methods and knowledge required to produce it.

These approaches have limitations and require time and resources. The most suitable assessment method may depend on the subject, class size and learning objectives.

The larger point is that universities may need to place greater emphasis on how students think, explain and apply what they have learned, rather than relying solely on submitted answers.

What This Means for Students

For students, the lesson is not that learning mathematics or developing expertise has become pointless. Rather, the value of understanding the material may become even more important as AI tools become easier to use.

Students who rely on AI without learning the underlying concepts may struggle when asked to solve unfamiliar problems, explain their reasoning or apply knowledge in situations where a chatbot cannot provide a dependable answer.

Using AI as a study partner can help students explore difficult concepts and practise problems, but they still need to verify answers and understand the reasoning behind them.

For those considering careers in mathematics, science, technology or research, it may also become increasingly important to combine subject knowledge with skills in critical thinking, problem-solving and working responsibly with AI tools.

Final Thoughts

Richard Borcherds' decision to grade homework based on completion rather than correctness highlights a real problem for education in the AI era. When students can generate convincing answers with technology, traditional homework scores become less useful as a measure of independent learning.

His approach is not a permanent solution, and it does not represent a university-wide policy at UC Berkeley. It is an example of one educator responding to a changing environment while acknowledging that the right answer remains unclear.

The broader challenge for universities is to adapt assessment so that students can benefit from AI without losing the ability to think, solve problems and understand the subjects they study.

The question is no longer simply whether students can find the correct answer. It is whether they understand why it is correct.

Official and related sources

The Daily Californian: https://www.dailycal.org/news/campus/academics/ai-use-leads-uc-berkeley-math-professor-to-see-no-point-in-grading-homework/article_bdf1eae2-c08b-491f-b9f1-81a7cea177de.html

UC Berkeley's explanation of its AI-detection policy: https://rtl.berkeley.edu/news/availability-turnitins-artificial-intelligence-detection

OpenAI's announcement about the Navier–Stokes problem: https://openai.com/index/navier-stokes-solution/

Richard Borcherds' UC Berkeley faculty profile: https://math.berkeley.edu/people/faculty/richard-e-borcherds

Never miss a new opportunity

Get instant alerts, join our WhatsApp community, or take a quick break with a game.