Generative AI Spurs Reform in College Assessments

Higher education must rethink assessment practices in response to the growing integrity challenges posed by generative artificial intelligence (GenAI), say authors in this Policy Forum. They analyzed data on student use of this technology across 20 major public research universities in the United States. The impact of GenAI on higher education is highly debated. In many ways, the technology is making common forms of evaluation, such as tests, projects, or term papers, less reliable as a measure of student capability. This highlights the need for a better understanding of where GenAI use is most prevalent and where misuse is most likely to occur. Igor Chirikov and colleagues analyzed survey data from more than 95,000 students across 20 U.S. research universities during the 2023-2024 academic year. Their findings reveal widespread GenAI use among students: roughly two-thirds of students reported using it over the study period, with 37% using it regularly. However, usage patterns differed considerably by discipline, with higher adoption in STEM fields. For example, 62% of computer science students reported regular usage, compared with only 24% of students in the arts. Notably, some social science disciplines, such as business and economics, also demonstrated high levels of adoption. Patterns of GenAI-assisted cheating also varied across disciplines. Estimated rates of misuse were generally higher in non-STEM fields, with economics (17%) and journalism (16%) showing relatively high rates, whereas biology (5%) was among the lowest. The study also found significant demographic disparities in GenAI use, with higher adoption among male, White, and Asian students than among female and underrepresented minority students. Although differences tied to socioeconomic status and disability were smaller, the authors suggest that the findings raise concerns about unequal access to AI tools and literacy. Chirikov et al. propose several paths forward. They note that there is no single "AI-proof" assessment model and suggest reforms tailored to individual disciplines. They place a focus on preparing students to use AI responsibly in professional contexts.