The rapid proliferation of generative artificial intelligence across higher education has reached a significant turning point at one of the nation’s premier public institutions. A distinguished mathematics professor at the University of California, Berkeley has taken the drastic step of eliminating graded online homework and remote testing entirely, pointing to the widespread, unchecked reliance on artificial intelligence tools that has rendered traditional remote assessment virtually pointless.
According to reports, renowned mathematician Richard Borcherds made the decision after concluding that it is now practically impossible to distinguish between authentic, student-generated work and answers synthesized by advanced large language models and specialized problem-solving algorithms. In response to this fundamental shift in the academic landscape, Borcherds has restructured his evaluation methods, relying almost exclusively on traditional in-person exams administered without the presence of any electronic devices to accurately gauge student comprehension and academic performance.
The decision highlights a growing crisis of academic integrity and institutional evaluation in universities worldwide as generative AI tools become increasingly sophisticated at solving complex mathematical proofs, calculus equations, and theoretical problems in seconds. For educators like Borcherds, the pedagogical value of homework as a diagnostic tool for learning has been severely undermined by the temptation and accessibility of automated cheating.
Online exams are now a waste of time as several students just use AI to answer the questions, Borcherds explained in remarks to the press, highlighting the stark statistical anomalies that tipped him off to the systemic shift. He noted that while he cannot always pinpoint individual cheaters with absolute certainty, the broader trends are undeniable, pointing out that he notices the average score increases a lot in online exams compared to controlled environments.
The mathematician elaborated on the perverse incentives created by conventional grading structures in the age of generative AI. Grading for correctness just adds a huge incentive for them to use AI for homework instead of doing it themselves and penalizes students for honesty, Borcherds observed. Furthermore, he emphasized the administrative and financial inefficiency of the status quo, noting that paying to grade AI-generated answers is a waste of time and money for both the university and its teaching staff.
The dilemma faced by the UC Berkeley professor reflects a broader, systemic challenge confronting the global academic community. Since the widespread public release of advanced generative artificial intelligence platforms, educators across disciplines ranging from computer science and mathematics to the humanities have struggled to maintain traditional assessment models. While writing-intensive fields have grappled with AI-generated essays and research papers, quantitative disciplines face a parallel disruption as AI models become increasingly proficient at symbolic math, coding, and logical deduction.
For decades, homework assignments and online quizzes served as foundational pillars of the undergraduate learning experience, allowing students to practice problem-solving concepts at their own pace while providing instructors with continuous feedback on class comprehension. However, the advent of AI calculators and reasoning models capable of breaking down complex mathematical derivations step-by-step has short-circuited this feedback loop. When students use AI to bypass the struggle of grappling with difficult problems, they may secure high marks on remote assignments, but they often arrive at midterm and final examinations entirely unprepared for the rigors of independent problem-solving.
Borcherds’ pivot back to traditional, high-security, in-person testing methods represents a growing retrenchment among university faculty members who are choosing to abandon remote components altogether rather than engage in an escalating arms race with cheating software. By removing electronic devices and moving evaluations back into physical classrooms under direct supervision, instructors can once again establish a reliable baseline for student knowledge and ensure that grades accurately reflect individual mastery of the material rather than access to superior technology.
The situation at UC Berkeley underscores a difficult transition period for modern higher education as institutions grapple with the permanence of generative artificial intelligence. As professors continue to weigh the benefits of digital learning tools against the reality of widespread academic dishonesty, the approach taken by faculty members like Borcherds suggests that traditional, proctored in-person evaluations may see a strong resurgence as the most viable defense of academic standards in an AI-driven world.
