The opinion section of the San Francisco Standard published an op-ed over the weekend that quickly reignited the ongoing debate over SAT and ACT requirements in UC admissions.
In about 2,000 words, UC Berkeley math professor Zvezdelina Stankova argues current UC system admits are underprepared, as selected students are not evaluated against baseline test scores in a piece titled, “I teach calculus at Berkeley. Some of my students can’t do middle school math.”
The op-ed has spawned a conversation online about students’ proficiency in math. The New York Post ran a story on the article, and a post about it by UC Berkeley Law professor Chris Hoofnagle has almost 3 million views on X.
The article was flagged by the AI detector Pangram as 33% AI-generated or AI-assisted writing. A similar result was found in a June open letter from STEM faculty advocating for standardized testing.
“The faculty team and a number of journalists worked intensively on my draft over a period of three weeks with multiple drafts, meetings,” Stankova said in an email. She added that she used AI to help edit the piece but that the article is the result of “several hundred person-hours of intensive human work and deliberation, of which about 80 hours are my own.”
The June open letter from STEM faculty advocating for standardized testing in UC admissions was signed by thousands of academics, including five Nobel laureates and was partially written by Stankova.
Stankova and a team of writers, including Hoofnagle, physics and math professor Mina Aganagic and math professors Svetlana Jitomirskaya and John Lott, said in a joint statement with nearly identical language to Stankova’s that the open letter was written over multiple weeks and in consultation with faculty across the UC system. AI, they said, was only used to edit the article.
In an email to The Daily Californian, Pangram CEO Max Spero said his company’s technology “has a false positive rate below 0.01% and can identify the boundaries between human-written and AI-written segments with high accuracy, particularly for long texts.”
However, the use of AI detectors such as Pangram in the classroom has been controversial.
UC Berkeley instructors do not have access to plagiarism checker Turnitin’s AI detection service due to concerns that it cannot offer conclusive proof of AI usage. A UC Berkeley webpage with guidance on using AI detection software cites studies finding that they are biased against non-native English speakers, and other research shows that detectors such as GPTZero can produce inaccurate results.
However, a University of Chicago study this year found that Pangram improved upon other detection models, achieving near-zero error rates.
Pangram did not detect AI usage in a similar letter by humanities and social sciences faculty. The Standard did not reply to requests for comment about how AI was used in the editing or creation of Stankova’s piece.
Hannes Bajohr, a UC Berkeley professor in the German department who researches the impact of generative AI on literary reading expectations, said he anticipates the norms around AI usage will change dramatically over the next few years.
“At the moment, this seems like deception. Or it seems like something that is at least dishonest to a certain reason,” Bajohr said, “Because it is the person that makes that statement in their capacity as an individual that underwrites what they have written.”
Bajohr added that he was doubtful that Pangram’s finding of about 30% AI-generated text meant much, and in 2023, he released a novel, “(Berlin, Miami),” which was co-written with AI.