The traditional landscape of higher education is undergoing a fundamental shift as the proliferation of generative artificial intelligence reaches a tipping point in the classroom. At the University of California, Berkeley, one of the world’s most prestigious public research institutions, the impact of this technology has led to a radical change in how students are assessed. Richard Borcherds, a renowned mathematician and professor at the university, has officially stopped providing graded homework assignments to his students, arguing that the rampant use of AI has rendered the practice effectively pointless.
For decades, the standard pedagogical model in mathematics and the sciences has relied on a combination of lecture-based learning, take-home problem sets, and periodic examinations. Homework was traditionally viewed as the primary mechanism for students to engage with complex material, allowing them to struggle with difficult concepts in a low-pressure environment before being tested. However, the emergence of advanced AI models has disrupted this cycle, making it nearly impossible for educators to ensure that the work being submitted reflects the actual knowledge or effort of the student.
According to a report by the New York Post, Borcherds has observed a dramatic shift in the quality and nature of student submissions. The mathematician, who is a Fields Medalist and a highly respected figure in his field, noted that the distinction between a response crafted by a human student and one generated by an artificial intelligence program has become virtually nonexistent. This loss of clarity has forced a reevaluation of the entire grading structure within his courses, as the traditional metrics for success no longer provide an accurate reflection of individual competence.
The decision to abandon graded homework stems from a pragmatic realization: the technological "arms race" between students using AI and professors trying to detect it has reached an impasse. While various software tools have been developed to identify AI-generated text, their efficacy in the realm of mathematics and technical problem-solving remains inconsistent. For a professor tasked with overseeing large cohorts of students, the time required to investigate every suspicious submission is prohibitive. Borcherds has concluded that the most effective way to address the issue is to remove the incentive for dishonesty altogether by decoupling homework from the final grade.
The shift is not limited to homework alone. Borcherds has also moved away from the use of online exams, a format that gained significant popularity during the COVID-19 pandemic and remained a staple of many modern university courses. In his view, the convenience of digital testing is now outweighed by the ease with which students can circumvent academic integrity protocols. The professor pointed out that online exams have essentially become a "waste of time" because of the sheer number of students who utilize AI to find answers or solve complex equations in real-time.
One of the most telling indicators of this trend, according to Borcherds, is the statistical anomaly found in grading data. He noted a significant and unnatural spike in average scores during online examinations compared to traditional, supervised settings. When the average performance of a class rises sharply without a corresponding change in the difficulty of the material or the quality of instruction, it often points to the use of external aids. For an educator, these inflated scores do not represent a sudden increase in student brilliance, but rather a systemic failure of the assessment method itself.
This creates a multifaceted problem for the university. First and foremost is the issue of academic integrity and the value of a degree. If a student can navigate a rigorous mathematics course at a top-tier university using AI to complete their assignments, the grade they receive ceases to be a reliable indicator of their skill level. Furthermore, the reliance on AI for homework creates a "perverse incentive" structure. Borcherds explained that when grades are based primarily on the correctness of take-home work, students who choose to work honestly are effectively penalized. They may spend hours struggling with a problem only to receive a lower grade than a peer who generated a perfect answer in seconds using a chatbot.
Beyond the ethical concerns, there is a significant financial and administrative burden associated with the current state of AI in academia. Borcherds highlighted the inefficiency of the traditional grading process in this new environment, noting that paying teaching assistants and readers to grade AI-generated answers is a waste of both time and money. In large university systems like UC Berkeley, the infrastructure for grading—which involves hiring graduate students and utilizing specialized software—represents a substantial investment. If the work being graded is not human-produced, that investment yields no educational return. It becomes an exercise in bureaucracy rather than pedagogy, where humans are essentially being paid to review the output of machines.
In response to these challenges, Borcherds has returned to what many might consider an "old school" approach to education. His grading criteria now rely almost exclusively on in-person examinations where the use of electronic devices is strictly prohibited. By returning to the physical classroom for assessments, the professor can ensure a level playing field where the only tools available to students are their own minds, a pen, and a piece of paper. This environment eliminates the possibility of AI intervention and forces students to demonstrate a genuine understanding of the mathematical principles being taught.
This transition reflects a broader debate within the global academic community. Educators across all disciplines are currently grappling with how to handle the integration of large language models and specialized AI tools. Some argue for the complete integration of AI, suggesting that students should be taught how to use these tools as part of their professional development. Others, like Borcherds, argue that the fundamental act of learning—particularly in a field as rigorous as mathematics—requires the kind of independent cognitive effort that AI is designed to replace. If the goal of a university education is to develop critical thinking and problem-solving skills, then the shortcut provided by AI may be fundamentally at odds with the mission of the institution.
The move to scrap graded homework does not necessarily mean that students are no longer expected to practice. Instead, it reframes the purpose of homework. In this new model, assignments are treated as a resource for self-study rather than a high-stakes performance. Students are encouraged to complete the work to prepare for the in-person exams, which now carry the full weight of their academic standing. This places the responsibility for learning squarely on the student. Those who use AI to "complete" their practice work may find themselves woefully unprepared when they sit down for a supervised exam without their digital assistants.
While this approach solves the immediate problem of grading integrity, it also highlights the growing divide between traditional educational methods and the rapidly evolving technological landscape. UC Berkeley has long been at the forefront of technological innovation, yet even its most brilliant minds are finding that some aspects of human development cannot be automated. The "correctness" of an answer, which was once the primary goal of a math problem, has become a commodity that can be generated for free. As a result, the value of education is shifting away from the final product and back toward the process of human thought.
As the academic year progresses, the results of this pedagogical shift will likely be watched closely by other departments and institutions. If Borcherds’ experience is any indication, the era of the take-home test and the graded problem set may be drawing to a close in the STEM fields. The return to the lecture hall and the blue book exam represents a defensive measure against a technology that is moving faster than institutional policy can adapt.
For the students at UC Berkeley, the message is clear: the era of "gaming the system" through digital shortcuts is being met with a return to supervised, manual rigor. While the world outside the classroom may be increasingly automated, the path to a degree in higher mathematics at Berkeley will continue to require the one thing AI cannot yet simulate: the unassisted, verified effort of the human brain. The decision made by Professor Borcherds serves as a stark reminder that in the age of artificial intelligence, the most valuable asset a student possesses is the knowledge they can prove they have when the screens are turned off.
