UC Berkeley Professor Acknowledges AI Use in Editing Op-Ed on Student Math Skills

A math professor at the University of California, Berkeley, has sparked a significant debate after criticizing a notable lack of mathematical proficiency among students in an op-ed published in the San Francisco Standard. In her article, Professor Zvezdelina Stankova indicated that many of her students lack basic skills, stating that they are “five to eight years” behind in their understanding of fundamental concepts like fractions and algebra. She attributed this deficiency to the UC system’s decision to adopt a test-blind admissions policy, suggesting that the absence of standardized testing like the SAT has led to the admission of underprepared students into Berkeley’s rigorous mathematics program.

The controversy intensified when reporters from Berkeley’s student newspaper, the Daily Californian, noted that some phrases in the op-ed appeared to have characteristics typical of AI-generated text. Berkeley sophomore Francis Luo utilized AI detection software called Pangram, which indicated that approximately 33% of the text might have been generated or influenced by artificial intelligence.

In light of the allegations, Stankova confirmed her use of AI tools to assist with editing, emphasizing that her article resulted from an extensive effort that included “several hundred person-hours” of work, with about 80 hours directly attributed to her. She highlighted that while AI aided in locating relevant research documents, the analysis and conclusions were the product of her team’s expertise.

The San Francisco Standard responded to the situation by affirming their stance on editorial integrity, stating that while AI may assist, human authorship and responsibility are paramount for the articles they publish. They maintained that Stankova’s piece reflects her and her colleagues’ thorough original analysis.

Stankova’s admission has ignited discussions on social media about the appropriateness of using AI in academic environments. Critics of Stankova have expressed concerns regarding the irony of relying on AI to craft an op-ed focused on students’ inadequacies, arguing that such a piece should be entirely original. Conversely, some supporters have defended her decision, asserting that the use of AI should not overshadow the main argument she presented.

In her communication, Stankova mentioned that the discourse surrounding her AI usage is “orthogonal to” the larger initiative advocating for the reinstatement of standardized testing in college admissions.

The debate over the relevance of standardized tests like the SAT or ACT in the admissions process has grown increasingly contentious. In 2020, the UC Regents voted to eliminate these tests as application requirements, a move followed by notable universities such as Harvard, Stanford, MIT, and Yale. However, several of these institutions have since reinstated testing requirements.

Stankova’s op-ed included data that suggested a significant increase in students with a “severe deficit” in calculus preparedness since the transition to test-blind admissions. Leading up to her publication, Stankova was one of over 3,000 faculty members who signed a letter advocating for the return of admissions testing. She acknowledged that AI tools were also employed in the editing process for this letter.

Looking ahead, the UC Academic Senate has announced plans to review whether standardized tests should be reintroduced in admissions, with any potential changes affecting fall 2028 admissions at the earliest. Additionally, the UC system established its own “AI council” to develop initiatives surrounding AI and its application in academic settings. Camille Crittenden, a council member, noted that there are currently no specific guidelines regarding AI use in personal research or opinion pieces, adding that many AI detection tools, including Pangram, still lack reliability.

Pangram claims high accuracy, asserting that it identifies AI-generated documents with a 99.66% success rate, generating a false positive once every 24,000 documents.