In a striking development that signals a growing crisis of academic integrity in the age of generative artificial intelligence, Richard Borcherds, a distinguished professor of mathematics at the University of California, Berkeley, has officially abandoned the practice of assigning graded homework. Borcherds, a Fields Medalist and a titan in the field of mathematics, concluded that the widespread, unchecked use of AI tools by students has rendered traditional take-home assessments functionally obsolete.
The decision by one of the world’s most respected academic minds serves as a bellwether for a broader, industry-wide reckoning. As Large Language Models (LLMs) become increasingly sophisticated, capable of solving complex multivariable calculus problems and articulating nuanced proofs in seconds, the barrier between student effort and algorithmic generation has effectively evaporated. For Borcherds, the pedagogical contract—the unspoken agreement that students will engage with material to demonstrate mastery—has been fundamentally broken.
The Chronology of a Classroom Crisis
The erosion of the traditional homework model did not happen overnight, but rather followed a trajectory aligned with the rapid public deployment of generative AI.
2022-2023: The Early Warning Signs
Following the release of ChatGPT in late 2022, faculty members across the University of California system began reporting anomalous performance data. Students who had previously struggled with foundational concepts were suddenly submitting near-perfect assignments. While professors initially attributed this to "extra study time," the uniformity and linguistic polish of the submissions began to raise red flags.
2024: The "Optimization" Phase
By mid-2024, AI tools evolved to include specialized mathematical solvers and symbolic logic engines. During this period, Borcherds noted a decoupling of student performance. Online assessment scores soared, but the correlation between those high scores and performance on in-class, closed-book assessments plummeted. The gap between what a student could do in the comfort of their home versus what they could demonstrate under exam conditions became a chasm.
2025: The Breaking Point
Throughout the 2025 academic year, Borcherds observed that the administrative burden of verifying student work had become unsustainable. The pedagogical value of "grading for correctness" had been hollowed out. If a student uses an AI to generate a solution, the professor is not grading the student’s understanding; they are grading the output of a silicon-based engine.
2026: The New Normal
In late 2026, Borcherds made the definitive shift. He informed his department and his students that homework would no longer contribute to final grades. Instead, he transitioned exclusively to in-person, proctored examinations. The shift was swift, pragmatic, and designed to strip away the "artificial" performance that had come to dominate his grading spreadsheets.
Supporting Data: The Discrepancy Gap
The data driving Borcherds’ decision is consistent with broader trends observed in STEM departments across the country. According to reports from the New York Post, the disparity between remote and in-person assessments has reached a statistical peak.
- Average Score Inflation: In courses where homework remains a significant percentage of the final grade, average scores have increased by 20% to 30% since 2023. However, in-person midterms and finals—conducted without electronic devices—have seen a stagnant or declining trend in average scores.
- Detection Failure: Borcherds noted that even with the most advanced AI-detection software, the "false positive" and "false negative" rates are too high to be actionable. The "arms race" between AI-generation and AI-detection tools has essentially reached a stalemate, leaving professors unable to verify the authorship of any digital submission.
- The Cost of Inefficiency: Beyond the academic integrity issue, there is an economic argument. Borcherds noted that "paying to grade AI-generated answers is a waste of time and money." When graduate student instructors (GSIs) spend hours grading submissions that were generated in seconds by an algorithm, the university is effectively subsidizing the automation of cheating rather than fostering intellectual development.
Official Responses and Institutional Friction
The response from the UC Berkeley administration and the broader academic community has been one of cautious observation. While Berkeley has historically been a champion of innovative teaching, the institution is currently navigating a delicate balance between embracing new technology and maintaining the rigor of its degree programs.
"The goal of a university is to certify competence," a university spokesperson noted in a general statement regarding AI policies. "If our methods of assessment no longer provide that certification, we must adapt."
However, not all faculty members agree with Borcherds’ "total prohibition" approach. Some departments are advocating for a "human-in-the-loop" model, where students are required to document their AI usage, cite their prompts, and write reflections on how the AI assisted their learning process. This approach seeks to treat AI as a calculator—an instrument to be mastered rather than a threat to be banned.
Borcherds remains unconvinced. He argues that in the context of foundational mathematics, there is no substitute for the raw, unassisted struggle with a problem. "Grading for correctness just adds a huge incentive for them to use AI for homework instead of doing it themselves," he stated, highlighting the "perverse incentive" structure that punishes honest students who struggle with the material, while rewarding those who delegate their critical thinking to software.
Implications for the Future of Higher Education
The decision to move away from graded homework is not merely a change in administrative policy; it is a fundamental shift in the philosophy of education. If the "homework" model is dead, what replaces it?
1. The Return of the Oral Exam
As written homework becomes untrustworthy, there is a growing movement toward a return to oral examinations—a practice common in the mid-20th century. By requiring students to defend their reasoning in real-time, professors can quickly assess whether a student has truly grasped a concept or simply copied a solution from an LLM.
2. Devaluing the "Correct Answer"
The traditional model of education relies heavily on the "correct answer." AI has effectively commoditized the correct answer. The future of pedagogy, many argue, must shift toward the process. In-class whiteboarding, peer-led problem solving, and supervised labs may become the primary venues for grading, as they are the only spaces where the student’s internal cognitive process can be observed directly.
3. The Widening Gap Between Elite and Mass Education
This crisis may create a bifurcated system. Elite institutions like Berkeley, which have the resources to conduct rigorous, in-person proctored testing, may maintain high standards of certification. Meanwhile, institutions that rely heavily on online learning or high-volume automated testing may see their credentials suffer from "grade inflation" and a lack of market confidence, as employers begin to view degrees from such institutions as less representative of true skill.
4. The Ethics of Honesty
Borcherds’ point regarding "penalizing students for honesty" is perhaps the most poignant. In an environment where AI usage is rampant but unpoliced, the honest student is at a significant disadvantage. By removing the graded homework component, Borcherds is leveling the playing field. He is forcing every student to demonstrate their aptitude under the same, high-pressure conditions, thereby protecting the value of the final grade for those who put in the actual work.
Conclusion
Professor Richard Borcherds’ stand against the tide of AI-driven homework is a wake-up call for the entire global education sector. It forces us to confront an uncomfortable truth: our current assessment structures were built for an era of information scarcity, not an era where information—and the processing of that information—is ubiquitous and instantaneous.
As universities scramble to update their codes of conduct and pedagogical methods, one thing is clear: the era of the "take-home assignment" as a primary measure of intellectual mastery is drawing to a close. Whether this leads to a renaissance of in-person, human-centric teaching or a further slide into a credential-driven bureaucracy remains to be seen. For now, in the lecture halls of Berkeley, the silence of the exam room is the only sound that still rings with the truth of genuine understanding.
