BERKELEY, CA — A new study led by Igor Chirikov, a senior researcher at UC Berkeley’s Center for Studies in Higher Education (CSHE) reveals that two-thirds of undergraduates at major U.S. public research universities utilized generative artificial intelligence (GenAI) during the 2023–2024 academic year. Both AI adoption and AI-assisted misconduct vary dramatically depending on the student's academic field.
The paper, titled "Generative AI use and misuse call for assessment reform in higher education," published in Science, is co-authored by Igor Chirikov (Senior Researcher at CSHE, UC Berkeley), Ivan Smirnov (University of Technology Sydney / Complexity Science Hub, Vienna), and René F. Kizilcec (Cornell University).
Analyzing survey data from 95,513 students across 20 major U.S. public research universities, the researchers deployed an indirect-questioning method to accurately estimate sensitive behaviors like cheating.
Key Findings:
High Usage in STEM, Higher Misconduct in Non-STEM
-
Regular GenAI use (monthly or more) is highest in fields like Computer Science (62%), Mathematics (53%), and Business (51%).
-
While overall GenAI-assisted cheating is estimated at a relatively low 9% among users overall, the rates are significantly higher among daily users (26%). They are also higher in non-STEM fields and vary by discipline: for example, the estimated cheating rate is 17% in Economics and 16% in Journalism, compared to just 5% in Biology.
-
Misconduct is heavily concentrated among high-frequency users. Daily GenAI users are more than three times as likely to cheat (26%) compared to monthly users (7%).
"Our results suggest that reforms should be implemented at the discipline level, as each discipline combines distinct learning goals, assessment traditions, and ways in which GenAI can support or substitute for student work.”
Rising Gender and Sociodemographic Gaps
The study also highlights concerns regarding equitable access and AI literacy. Male students (45%) and White/Asian students (39%) are significantly more likely to use GenAI regularly than female students (33%) and underrepresented racial minorities (29%).
"These findings highlight concerns about equitable access to technological resources in higher education, as students from underrepresented backgrounds may have reduced access to or familiarity with GenAI."
A Roadmap for Universities
The authors urge universities, academic societies, and accreditors to prioritize long-term assessment redesign rather than relying on restrictive rules that are difficult to enforce. Recommended strategies include:
-
Selective Use of Controlled Settings: Utilizing in-class, oral, or practical exams when independent performance must be verified.
-
Process-Oriented Assessments: Structuring assignments to require students to document their reasoning, critique AI outputs, or justify their choices, shifting focus away from just the final artifact.
-
Faculty Development: Investing in training instructors to understand AI capabilities and design better coursework.
"Fair evaluation now depends not only on what is assessed but also on how evidence of learning is generated, interpreted, and bounded within particular disciplinary contexts," the authors conclude.
About the Center for Studies in Higher Education (CSHE)
Established in 1956 at the University of California, Berkeley, CSHE is the nation's first research institute devoted to the study of systems of higher education. The Center produces multi-disciplinary research to inform institutional strategy, public policy, and academic practice.
Media Contact:
Shanshan Jiang-Brittan
Center for Studies in Higher Education, UC Berkeley
Link to Paper