Artificial intelligence is shaking up college classrooms, leading some of California’s top professors to revert to a more traditional approach: using paper.
Professors from institutions like UCLA, USC, UC Berkeley, and Stanford shared their thoughts on AI’s impact in a recent conversation. While a few are eliminating technology from classroom assessments entirely—opting to grade only in-class written work—others are finding ways to leverage AI positively in their teaching.
“The technology has been shoved down all our throats, and we’re all just dealing with the fallout,” remarked Morgan Polikoff, an education professor at USC Rossier.
Taking a break from teaching, Polikoff mentioned his inclination to transition more of his undergraduate assignments to in-person, paper-based tests. He noted that he might implement a grading strategy similar to UC Berkeley math professor Richard Borcherds, who recently stopped grading homework for accuracy.
Polikoff recounted an incident from two years ago when he caught a student using AI for a paper—something he had specifically prohibited. Spotting it was straightforward, and the student admitted it when confronted, leading Polikoff to have him rewrite the paper from scratch.
“To assess knowledge and skills in this new landscape, we’ll need more standardized and AI-resistant methods, both for admissions and academic performance,” he said, emphasizing the growing difficulty in gauging what students truly understand. “I don’t see any value in students submitting AI-generated work. That’s just wasting everyone’s time,” he added.
At USC, AI usage rules are left to individual instructors, allowing them to develop course-specific guidelines. Polikoff’s PhD program is currently navigating a policy that still grants faculty autonomy.
The university invests about $3.1 million annually on a subscription to ChatGPT Edu for 80,000 users, which Polikoff dismissed as “pennies” compared to what they pay their football coach.
Across Los Angeles at UCLA, Thomas Maierhofer, who teaches statistics and data science, has adopted a mixed approach. “All exams are in-person and on-paper, with no phones or laptops allowed. Assignments outside the classroom are graded for completion,” he explained.
Maierhofer views AI as a potentially beneficial educational tool and is in the process of developing a custom chatbot designed to support learning. He believes his exam policy encourages honesty among students.
“If exams are taken in-person and on-paper, students are really only hurting themselves if they choose to use AI dishonestly, instead of using it to boost their learning,” he stated.
At UC Berkeley, Borcherds, known for his decision to stop grading homework for accuracy, now mostly assesses students based on in-person tests without electronic devices. “Online exams have become ineffective, as many students use AI to respond,” he noted. “I can’t always tell who’s using it, but I see many more students performing better in those formats.”
The University of California allows individual campuses and programs to enforce stricter AI regulations, particularly related to admissions and academic integrity.
At Stanford Law School, Professor Julian Nyarko takes a different view. “I think AI has tremendous potential. Too many humans aren’t great teachers; I actually have more skepticism about human abilities than I do about AI,” he said.
In a study led by Nyarko, 16 law professors anonymously evaluated responses to common student questions and favored Google’s AI answers over their own colleagues’ responses 75% of the time. However, researchers emphasized that this study focused on answer quality rather than long-term learning outcomes with AI support.
Nyarko cautioned against misuse of the technology. “Using AI irresponsibly, like asking it to write a paper or solve a problem, is harmful,” he stressed.
He mentioned that while he has always given in-class exams, many colleagues who used to assign lengthy take-home tests now feel unable to, fearing students will just ask AI for help. “Restricting access to AI seems impossible now; most students already have it at their fingertips,” he acknowledged. “This shift necessitates significant changes in our educational practices. We’re in a transition from thinking we understood how education worked to realizing we need to figure out how to teach effectively in light of AI.”
Meanwhile, Stanford is providing all students with free access to ChatGPT, Gemini, and Claude through a pilot program lasting until August 2027. Unless instructors impose stricter guidelines, students can use AI as a learning aid, provided they don’t submit AI-generated work as their own.
The discourse around AI continues to impact discussions for the upcoming state superintendent race. Sonja Shaw, the Republican candidate for California Superintendent of Public Instruction, argues that too much reliance on AI undermines crucial skills needed by students for college and employment. She advocates for a return to rigorous academic standards.
“AI-assisted work and grade inflation make it harder to discern who actually understood the material,” Shaw remarked. She believes that objective testing remains the best way to assess student abilities without the crutch of tools like ChatGPT. “AI should be viewed like a calculator—it can be a helpful tool, but it can be dangerous if you don’t grasp the foundational concepts first,” she added.
