UC Berkeley Professor Stops Grading Homework for Correctness
A professor at UC Berkeley has decided to stop grading homework based on correctness, after observing that artificial intelligence tools were helping students achieve high grades. Richard Borcherds, a notable mathematician, expressed that he no longer finds it sensible to pay graders to check the accuracy of students’ work when AI can increasingly handle that task.
This decision from Borcherds follows a recent announcement from OpenAI, which revealed that its AI agents had seemingly resolved the difficult Navier-Stokes existence and smoothness problem, considered one of the major unresolved problems in mathematics.
“If AI keeps advancing at this pace, in a few years, there won’t be anything human mathematicians can do that AI can’t outperform,” Borcherds shared with Berkeley’s student newspaper, the Daily Californian.
Having received the prestigious Fields Medal in 1998 for his significant contributions to the field, Borcherds now assigns credit based on whether students complete their assignments rather than the correctness of their answers.
He noted that while this approach isn’t ideal, he feels he has limited options at the moment. This shift illustrates the increasing challenges faced by universities as students gain access to AI systems capable of performing complex academic tasks.
Borcherds also conveyed his uncertainty regarding whether prospective mathematicians should continue pursuing the field. Just five years earlier, mathematics seemed to offer robust protection against disruptions that other professions were experiencing. “I have no idea what advice to give,” he admitted. Interestingly, he even consulted ChatGPT on this question, and it, too, didn’t have a clear answer.
The professor pointed out that universities might not swiftly adjust to rapid AI advancements, noting a typical hesitance to change within academic institutions. He predicts that mathematics departments could continue with their current practices for several more years, despite the progression of AI technologies.
Borcherds likened the situation to academic journals that persist in publishing physical copies, even though he believes they long ago lost much of their relevance.
“We’ve seen many technological advances in the past—like looms replacing hand weaving and cars taking over from horses—but this is the first instance of something like this impacting mathematics,” he remarked.
Berkeley has already put guidance in place requiring faculty to address AI usage in their courses. Anne Marie Richard, the university’s assistant vice provost for undergraduate education, mentioned that the Berkeley Division of the Academic Senate advises instructors to include their expectations regarding AI tools in their syllabi. However, the specific limitations may vary significantly from class to class.
While the university has looked into AI-detection software like Turnitin and Rumi, it ultimately found these tools unreliable for campus-wide application. Thus, individual instructors maintain the authority to set their own AI policies within their classrooms.
