Berkeley study flags AI cheating

UC Berkeley researchers on May 21 released what Berkeley News described as the largest study yet of undergraduate use of generative artificial intelligence, finding that use and misuse differ sharply by subject area and by students’ socioeconomic backgrounds. The study’s release coincided with a separate Cornell report arguing that widespread AI misuse is already undermining how colleges measure student learning. A classroom dispute at Western University in Canada added a live example this week after a professor said he believed most students in one class had used AI to cheat on a final exam. Together, the cases have pushed academic integrity and assessment design back to the center of higher-education debate. ### How large was the Berkeley study, and what did it find? Berkeley News said the study found a “slippery slope” in student behavior: the more undergraduates used AI, the more likely they were to cheat with it. The article said researchers found disparities both in access to AI tools and in misuse, with patterns varying by discipline and by socioeconomic status. (news.berkeley.edu) The Berkeley report did not frame AI use as uniform across campus. Instead, it said students’ use and misuse differed by subject, and it highlighted unequal access alongside unequal cheating patterns. That made the findings broader than a simple count of how many students were trying chatbots for homework help. ### What numbers are universities focusing on? (news.berkeley.edu) Cornell said its researchers analyzed survey responses from more than 95,000 students at 20 public research universities in the United States. The university said 37% of students reported using AI at least monthly, with regular use reaching 62% among computer science students and 24% among students in the arts. Cornell also said the survey showed demographic differences in generative AI use. (news.berkeley.edu) The article linked those patterns to a broader concern about whether current assignments and take-home assessments still measure students’ own work in a reliable way. ### Why are faculty talking about assessment, not just cheating? Cornell’s report said the issue is not only misconduct but the validity of existing assessment models. “Even this early stage evidence shows that we have a very serious challenge on our hands, and universities need to address that,” the article said. (news.cornell.edu) A related version of the report said the study, titled “Generative AI Use and Misuse Call for Assessment Reform in Higher Education,” was published May 21 in the journal Science. A Cornell-linked account of the findings quoted researchers saying misuse of generative AI is “a problem for assessment validity” and therefore for “the credibility of university credentials.” That language has helped shift the discussion from whether students are breaking rules to whether colleges are still testing the right skills in the right format. ### What happened at Western University this week? (news.cornell.edu) CTV News London reported on May 20 that a Western University professor rejected exam results after alleging that most students in one class had used AI to cheat on a final. The report did not by itself establish the allegations, but it showed how quickly disputes over AI use can escalate from policy debate to contested grades and exam outcomes. A separate local CTV listing described the case as involving a Western professor who alleged students used AI to cheat on a final exam. (nationaltribune.com.au) The incident surfaced as universities were already digesting the Berkeley and Cornell findings. ### What comes next for colleges? Science publication on May 21 gives the Cornell study an immediate audience among university leaders, faculty senates and academic integrity offices. Berkeley’s findings and the Western dispute add pressure on institutions to revisit exam formats, assignment design and enforcement rules for AI-assisted work. (ctvnews.ca) Cornell’s article said universities need to change how they evaluate students. The next phase is likely to play out in campus policy meetings, course redesigns and disciplinary cases as schools decide whether existing honor codes and assessment models can survive widespread student access to generative AI. (news.cornell.edu) (news.berkeley.edu)