UC Berkeley professor admits to using AI in op-ed
UC Berkeley math professor Zvezdelina Stankova published an op-ed advocating for the return of SAT/ACT requirements in UC admissions, arguing that current admits are underprepared and some cannot perform middle school-level math The op-ed was flagged by AI detector Pangram as approximately 33% AI-generated or AI-assisted, reigniting debate over both standardized testing and AI use in academic writing Stankova and co-authors acknowledged using AI only for editing, emphasizing the piece resulted f
Analysis
TL;DR
- UC Berkeley math professor Zvezdelina Stankova published an op-ed advocating for the return of SAT/ACT requirements in UC admissions, arguing that current admits are underprepared and some cannot perform middle school-level math
- The op-ed was flagged by AI detector Pangram as approximately 33% AI-generated or AI-assisted, reigniting debate over both standardized testing and AI use in academic writing
- Stankova and co-authors acknowledged using AI only for editing, emphasizing the piece resulted from "several hundred person-hours" of human collaboration over three weeks
- The controversy highlights growing tensions around AI detection reliability, with UC Berkeley citing concerns about bias against non-native English speakers and inconclusive detection accuracy
- A University of Chicago study found Pangram achieved near-zero error rates, outperforming other detection models, though its use in classrooms remains controversial
Why It Matters
This article sits at the intersection of two major debates in higher education: the role of standardized testing in college admissions and the increasing use of AI tools in academic writing. For AI practitioners and researchers, it underscores the growing scrutiny of AI-generated content in scholarly and public discourse, as well as the limitations and biases inherent in current detection technologies.
Technical Details
- AI Detection: Pangram's detector flagged the op-ed at 33% AI-generated content; its CEO claims a false positive rate below 0.01% with high accuracy in identifying boundaries between human-written and AI-written segments in long texts
- Detection Controversy: UC Berkeley does not provide instructors access to Turnitin's AI detection service due to concerns about inconclusive proof; the university's own guidance cites research showing detectors are biased against non-native English speakers and can produce inaccurate results
- Comparative Performance: A 2025 University of Chicago study found Pangram outperformed other detection models with near-zero error rates, though it did not flag a similar letter by humanities and social sciences faculty
- Collaborative Authoring Process: The op-ed involved multiple faculty members (including Nobel laureates on the related open letter), journalists, and iterative drafting over three weeks, with AI used solely in an editorial capacity
- Academic Response: Professor Hannes Bajohr expressed skepticism about the significance of the 30% detection result and noted that norms around AI usage are expected to shift dramatically in the coming years
Industry Insight
- The incident illustrates the growing need for transparent AI usage policies in academic and professional publishing, as the line between editing assistance and content generation becomes increasingly blurred
- AI detection tools remain unreliable enough to warrant caution in high-stakes contexts; institutions should invest in clear guidelines rather than relying solely on detection technology
- The standardized testing debate, amplified by AI-related concerns about student preparedness, suggests that admissions policies may see a partial reversal toward test-optional frameworks, creating implications for edtech and assessment industries
Disclaimer: The above content is generated by AI and is for reference only.