Berkeley maths don revamps grading over rampant AI cheating

The increasing sophistication of artificial intelligence is compelling educators to fundamentally alter traditional assessment methods. Richard Borcherds, a UC Berkeley mathematician, has shifted to grading homework merely for completion and now relies on in-person exams without electronic devices. This change addresses the challenge of distinguishing human work from AI-generated submissions, a task Borcherds states has become "nearly impossible" with current AI tools.

This difficulty is compounded by the reported unreliability of AI detection software. UC Berkeley found tools like Turnitin unsuitable for campus-wide adoption. While earlier AI tools might have been identifiable by "characteristic mathematical errors," rapid technological advancements have made such mistakes rare, effectively neutralizing a key method for detection. This leaves individual professors to devise their own policies for managing AI use.

Academic research supports these observations, with studies in Discover Education and Teaching of Psychology indicating a significant rise in online exam scores after ChatGPT's launch, particularly for unmonitored students. This suggests online tests may inflate scores and not accurately reflect student abilities. Borcherds argues that grading for correctness under these conditions penalizes honest students and wastes resources on evaluating AI-generated answers, underscoring a shift from detection to a revised assessment paradigm.