Fields Medal professor ends homework due to AI cheating
The age-old battle between students and assignments just got a whole lot more complicated. A renowned mathematician at the University of California, Berkeley is essentially throwing in the towel on graded online homework, arguing that the rise of artificial intelligence has made the traditional system obsolete.
For professor Richard Borcherds, the issue isn’t just cheating; it’s the fundamental inability to distinguish between genuine student work and AI-generated submissions. He notes that it has become nearly impossible to tell if a student wrote an answer themselves or if an algorithm produced it. This reality has fundamentally undermined the purpose of online testing.
“Online exams are now a waste of time,” Borcherds explained. “I cannot tell which students are using AI, but I notice the average scores increase significantly in these online settings.” He further pointed out that the focus on grading for absolute correctness only pushes students toward using AI for homework instead of engaging in actual learning, effectively penalizing honesty.
This dilemma is not new, but the technology has outpaced the detection methods. When OpenAI released its first popular large language model, GPT-3.5, in late 2022, the potential for AI-assisted work was theoretical. Now, the sophistication of these tools has reached a point where simply grading answers for correctness no longer serves as a fair incentive for students to do the work themselves.
Researchers have independently corroborated the concerns about unmonitored testing. Studies have shown that students routinely achieve higher scores on online exams, suggesting that these assessments no longer accurately reflect what students have actually learned in a course.
The stakes have driven some institutions to rethink their entire testing philosophy. For instance, a professor at Brown University recently decided to shift to in-person final exams after observing a dramatic spike in student performance on take-home midterms. This shift resulted in far fewer students achieving top scores, underscoring the fact that true mastery is better measured face-to-face.
While the problem persists, the academic community is tackling the issue through policy rather than just detection software. The University of California, Berkeley, for example, determined that relying on AI detection tools was unreliable and opted instead to empower professors to establish their own clear AI policies within course syllabi.
Ultimately, the focus is shifting away from policing student output and toward defining learning in a world where artificial intelligence is an inescapable reality. As long as AI continues to evolve, the education system must adapt to measure understanding, not just the presence of text.
Buy on Amazon
Credit
