The new study says professors and universities need to guide students better on how to use AI in coursework.
getty
No one really knows (yet) how much cheating is going on in colleges among students using generative AI. “Some claim it’s everywhere. Some say there’s no change,” says Rene Kizilcec, associate professor of information science at Cornell University and director of the Cornell Future of Learning Lab. “It’s really interesting to try to figure out what’s closer to the truth.”
That’s what a new study being posted online today by the journal Science, and coauthored by Kizilcec, sought to uncover. The researchers examined how students use AI across different academic disciplines and found that use was highest in quantitative fields like computer science, business and economics — not the humanities, where public anxiety about cheating has largely focused on students using AI to write essays.
The findings are based on survey responses from 95,500 students at public U.S. research universities, which have a total enrollment of more than 2.6 million undergraduates and award more than half of all bachelor’s degrees in STEM subjects (Science, Technology, Engineering and Math).
AI cheating appears less widespread than many professors and parents fear, Kizilcec says. While two-thirds of students in the survey reported using generative AI during the 2023-24 academic year, only 9% admitted knowingly submitting AI-generated work when it may not have been allowed. Among daily AI users, however, that figure rose to 26%. “That suggests that as students use it more and more, cheating with AI will also increase,’’ Kizilcec tells Forbes.
Kizilcec warns the findings point to a much larger problem: AI is exposing weaknesses in how colleges measure learning and whether degrees still reliably signal competence to employers. “The stakes are really high here,” he says. “It’s not just about a little bit of cheating, it is about the integrity of the institution and the credibility of the degrees that are issued.”
That concern is already reshaping higher education. Earlier this month, Princeton University faculty voted to require proctors for in-person exams starting July 1, ending a 133-year tradition of unsupervised testing conducted under the school’s honor code.
More Controlled Assessments
“Honor systems are a great idea and a great ideal,” Kizilcec says. But he argues AI has intensified existing competitive pressures in higher education, where grading on a curve and high-stakes outcomes can incentivize misuse. “In a world where you can get ahead by using something, and there’s a low chance of being caught, it’s no surprise the incentives cause students to act in that way,” he says.
In fact, The Harvard Crimson surveyed roughly 300 students and more than 40% admitted they regularly used AI for coursework in ways instructors might consider inappropriate.
Some colleges are already responding by reviving more tightly controlled testing environments. Across higher education, professors are increasingly experimenting with handwritten essays, in-class testing and oral exams that require students to explain their thinking in real time.
But Kizilcec cautions against treating blue book pen-and-paper style exams as a universal solution. Traditional exams often fail to capture broader professional skills, including judgment, collaboration and the ability to work effectively with AI tools students will likely use throughout their careers.
That tension is especially apparent in fields like computer science. At Columbia University, some professors are shifting away from evaluating polished final outputs alone and placing greater emphasis on whether students can explain, critique and improve AI-assisted work.
“The evidence is too clear that something needs to change about how we assess students in college,” Kizilcec says. “There are multiple accounts of how you could just pass all of your assignments without ever thinking about them.”
What Parents and Students Should Look For
For families evaluating colleges, the emerging AI debate may increasingly center less on whether schools ban ChatGPT and more on whether they have coherent strategies for adapting student assessment to the AI era.
One major challenge, Kizilcec says, is that many students remain unclear about what kinds of AI use are acceptable. For another project, his research lab analyzed course syllabi and found that many policies fall into vague extremes: “Use it however you want” or “don’t use it at all,” with little guidance in between.
Today’s Science report asserts that colleges need clearer, course-specific guidance about where AI can appropriately assist students — whether for brainstorming, editing, coding or feedback — and where it crosses the line into replacing core disciplinary thinking.
The study also calls for significantly more faculty training. Many professors are still learning about AI’s capabilities and limitations themselves, even as students rapidly adopt the technology. “All universities have some kinds of workshops for faculty, but that type of work needs to expand,” Kizilcec says.
“There are so many ways to use AI in completing a task, and some of them are really supportive of the learning process,” Kizilcec says. “Others are giving up agency in ways that undermine learning.”
For colleges, the challenge now may be less about forbidding students to use AI than redesigning assessments that can still reliably measure what students actually understand.
