Berkeley mathematician drops homework grades as AI makes cheating hard to spot
UC Berkeley professor Richard Borcherds has stopped grading homework, citing AI-written work that is nearly impossible to tell apart from human writing, TechRadar reports.
According to TechRadar, Richard Borcherds, a mathematics professor at the University of California, Berkeley, has stopped grading homework for his students. The outlet reports that the change appears to follow improvements in AI-generated homework, which it says have made it extremely difficult for instructors to identify submissions written by people.
TechRadar says Borcherds described it as nearly impossible to spot a human-written submission among work produced with AI. The outlet's headline goes further, calling online exams a waste of time and saying students are increasingly using AI to cheat, though the coverage is brief on how widespread that trend is.
The report says grading in Borcherds' classes now relies on in-person exams, taken without electronic devices. It also notes that he is British-born and that his work has been recognized with awards. The available summary cuts off before listing those honors or other details of his career.
Beyond what TechRadar reported, this coverage does not say how many courses or students are affected, whether other Berkeley faculty have made similar changes, or how the university has responded.
Background
Since widely available generative AI tools arrived in late 2022, students have been able to produce essays, summaries and problem sets with a few typed prompts. Homework has long been a low-stakes way to practice, and in many courses it has been graded mainly for completion or effort. When that work can be outsourced to software, the grade no longer reliably reflects what a student knows.
Universities have responded in several ways. Some have moved high-stakes assessment into proctored, in-person exams, where students work without phones or laptops. Others have added oral examinations, handwritten "blue book" tests, or project defenses where a student must explain their reasoning live. Many instructors have also tried AI-detection software, but these tools have produced false positives and are difficult to rely on as evidence, which has made some educators cautious.
Mathematics presents a particular challenge. Modern AI systems have become much better at standard calculations and at writing solutions to familiar problems, and the polished written steps they produce can look like the work of a capable student. Instructors often rely on a sequence of problems that build understanding, so any shortcut that skips the struggle can weaken the link between homework and learning. Educators have been debating how to balance practice, feedback and fair measurement in this environment for several years.
Why it matters
A change like this one signals that assessment is becoming the central problem, not just the rules around cheating. If an instructor cannot tell whether a homework set reflects a student's own thinking, the grade no longer carries the same meaning. That affects students who want honest feedback, employers who rely on degrees as signals of ability, and institutions that must defend their standards.
The move also raises a tradeoff. Exams taken under supervised conditions can confirm individual skill, but they test a narrow kind of performance under time pressure. Homework, when it works well, encourages extended engagement with difficult material. Dropping it risks losing that practice, unless instructors find new ways to provide it, such as in-class problem solving, office hours or guided sessions where the work is visible.
For the wider AI conversation, this is a concrete example of a tool changing a routine academic practice. It shows that the question is less about whether AI can do schoolwork and more about how education can still measure and develop human understanding when the work itself is easy to outsource.
What to watch
Watch whether other mathematics and science departments at Berkeley and elsewhere adopt similar policies, and whether they pair them with new forms of practice. Pay attention to how institutions handle the cost of in-person exams, including accessibility for students with disabilities and those with conflicting schedules.
It will also be worth tracking whether universities revise their guidance on AI use, detection tools and oral assessment. Finally, watch for data or reporting that shows how common AI-assisted homework has become, since the current coverage relies on one professor's account as reported by a single outlet.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
It is worth asking what an in-person, device-free exam actually measures. A timed test can confirm that a student can recall methods and solve problems quickly under pressure, but it does not always show whether they understand the ideas behind them. If homework is dropped and exams become the only check, the course may end up rewarding speed and memorization more than the slow, messy process of working through a hard problem.
There is also a question about how far one account can be generalized. A single professor's experience, reported in a short piece, does not tell us how common this shift is across campuses or disciplines. Some instructors may find that their homework still works well, especially where feedback is given in person or through discussion rather than through final grades.
Finally, the claim that human and AI work are nearly impossible to distinguish deserves scrutiny. If that is true, the problem is not only about cheating; it is about whether the signals schools use to judge learning still mean what they once did. Those are open questions that deserve more evidence before the lesson is treated as settled.
The User what changes for people who use it
For students, the practical change is significant. Homework is often where people get their first honest feedback on whether they understand a topic. If that feedback disappears, students lose a cue that tells them when to slow down and revisit a concept before a midterm arrives. Many will feel less guided, especially in difficult courses where the gap between confidence and mastery can be wide.
The shift also lands differently depending on circumstances. Students who work evening shifts, care for family members, or have disabilities that make a rigid in-person test harder may find that a single supervised exam carries more weight than they can comfortably manage. Flexibility that homework once provided could quietly disappear.
Still, students can adapt. Study groups, office hours and practice exams become more important when graded homework is no longer the main rhythm of the term. The practical question for anyone in this situation is how to keep learning visible, not just how to get through the grade.
The Optimist the hopeful reading
There is a hopeful reading here. If homework can no longer be trusted as a grade, instructors may be pushed back toward the things that made learning work best in the first place: explaining reasoning aloud, defending a proof in person, and working through problems in front of others. Those methods can reveal far more about understanding than a completed worksheet ever did.
Removing the grade may also return homework to its original purpose, which is practice rather than judgment. When students know that their written answers are not the final verdict, they may feel freer to attempt hard problems, make mistakes and learn from them. That kind of low-pressure engagement is hard to build when every submission carries a score.
The change will not solve every challenge, but it invites a more thoughtful design of courses. For educators willing to experiment, this moment could lead to assessments that are harder to outsource and more useful to the people doing the learning.