A Paradigm Shift in Academic Assessment

As artificial intelligence tools continue to advance at a staggering pace, the academic world is grappling with profound questions regarding integrity and evaluation. Richard Borcherds, a distinguished mathematician and 1998 Fields Medal recipient at the University of California, Berkeley, has fundamentally altered how he grades student assignments after noticing a dramatic surge in the use of generative AI.

Borcherds revealed that he has stopped grading student homework based on the correctness of final answers. According to the professor, the ubiquity of advanced AI models means that many students can easily bypass the intellectual labor required to solve complex mathematical problems, receiving high marks without genuine comprehension.

The Limitations of AI Detection

Addressing the broader implications for higher education, Borcherds noted that distinguishing between authentic student work and sophisticated AI-generated output has become increasingly difficult. While earlier iterations of artificial intelligence often produced glaring, easily identifiable errors, current models have achieved a level of refinement that renders such mistakes rare.

Universities across the United States are currently wrestling with the technological shift. UC Berkeley has instructed faculty members to clearly outline their individual policies regarding artificial intelligence in course syllabi. However, institutional efforts to deploy automated detection tools—such as Turnitin and Rumi—have fallen short, with campus evaluations concluding they lack the reliability required for universal deployment.

Consequently, individual professors retain the autonomy to establish their own classroom guidelines.

Securing Academic Integrity

To ensure fair evaluation, Borcherds has shifted the emphasis of his grading framework toward in-person examinations where electronic devices are strictly prohibited. He dismissed online testing as largely ineffective, noting a stark discrepancy in average scores that suggests widespread unauthorized assistance.

Under the revised policy, homework completion rather than correctness serves as the primary metric, effectively protecting students who choose to complete their coursework honestly while mitigating the time and financial inefficiency of auditing AI-assisted submissions.

Looking beyond the immediate challenges of classroom management, Borcherds expressed broader concerns regarding the long-term trajectory of mathematical careers, suggesting that the rapid evolution of artificial intelligence may fundamentally reshape the discipline in the coming years.