In a world where AI can generate research papers, solve equations or create art, educators worry about how college students may be using it, misusing it or missing out on it. Yet there have been few comprehensive studies of college students and their AI use.
Now, Igor Chirikov, a senior researcher at UC Berkeley’s Center for Studies in Higher Education, has published the largest study of generative AI use by undergraduates, in collaboration with researchers from the University of Technology Sydney and Cornell University. More than 95,000 students at 20 research-intensive public universities responded to questions about how they use AI, including whether they use it to cheat. The findings were published on May 21 in Science.
Key findings:
At least 9% use AI to cheat; number varies significantly by discipline with non-STEM majors cheating at higher rates.
Low-income, racially underrepresented and female students use AI less highlighting a digital divide.
Wealthier have the ability to pay for AI tools, thus gaining an unfair advantage (that’s not necessarily tied to their skills).
Employers actively looks for grads with AI experience, meaning poorer students risk falling behind in college and the workplace because of these financial barriers.
Researchers recommend universities teach responsible AI use while finding new ways to measure student knowledge that can’t be faked.
First of all, this is data from two years ago, relatively early in the cycle of AI adoption. In terms of GenAI capabilities and usage, it was definitely lower than what we’re seeing now.
Second, our numbers are conservative. We used an indirect survey method to make it easier for students to answer honestly about a sensitive behavior. But students still had to recognize when their use was not allowed, and that is challenging.
But even with those restrictions, it’s still a significant number of students. I show in a recent paper on grade inflation that when students use AI on assignments, an entire course’s grades may become inflated compared to what the students actually know.
AI cheating in an advanced Econ class at Brown, per professor
The temptation to use artificial intelligence (AI) to cheat is shaking up elite universities in the United States. Professor Roberto Serrano, who is the Harrison S. Kravis University Professor of Economics at Brown University, has detected a massive fraud in one of the classes he teaches, ECON 1170, an advanced undergraduate course in mathematical economics. He has conclusive evidence that at least 50 students cheated on the March midterm exam
When he reported the case to high-ranking officials at Brown, he got a cold reaction. The response from the provost, he said, was absolute silence. The dean did not comment either until Serrano took the case before the Academic Code Committee. At that point, he received a note acknowledging that what had happened in his classroom was “a wake-up call.” Serrano, a Madrid-born economist who has been at Brown for 34 years, believes this is not enough. “That cannot be the university’s position before an incident of this magnitude. Academic integrity is a value worth defending. The faculty cannot be left on its own in a battle that is decisive if we want to preserve the future of higher education,” explains the 61-year-old professor in a telephone conversation from Providence, Rhode Island. To prevent AI from ending the prestige and utility of teaching, he feels, it is necessary to adopt a different approach: “We need to publicly admit the seriousness of the situation and open up a broad debate about the real extent of the problem.”
The wake-up call is to emphasize, once again, that the issue is with the exams and current teaching methods. As I posted on another thread, the onus is on educators to embrace this new tool and teach students how to use AI to deepen their learning and thinking, not blame students for outsmarting our now outdated teaching methods and assignments. We need to reinvent what and how we teach such that the use of AI is integral to desired learning outcomes, not a way around them. If AI currently can be used to mail in assignments and exams, the problem is with the assignments and tests, not the students. This is a teaching problem, not a student problem.
In an AI world, it IS “necessary to adopt a different approach.” For example, here are some of the resources Harvard is offering to help faculty teach in the age of AI. Nobody has this all figured out yet, but the best argument against “cheating” I’ve heard is to solve the teaching problem.
It seems pretty obvious that any take home exams are going to have people who use AI to do their work. Students who follow the rules are at a huge disadvantage.
The solution, at least for now, is requiring in person exams.
I agree with this, specifically the exams. I’ve wondered if we might see more colleges using viva voce exams to confirm knowledge and understanding. So the student might write a large research paper, and then have a discussion or Q&A with a professor to defend it. Yes, this can probably only be done at more well-resourced colleges, but it would give those students and graduates a very clear signal that they understand the material. I’d expect a Brown, Harvard, Amherst, etc. to maybe try this path. Could be interesting to see.
Based on the article, the adminstration didn’t seem too concerned about it. But now the professor has exposed the cheating, it may force them to do something just like the rampant grade inflation at the top schools.
what % normally cheat in some way? is this higher or lower? no clue! I went to an ivy a million years ago and there was a big cheating scandal in an intro class when I was there when a large number of kids were caught up. I am not saying AI isn’t making it more likely (it seems logical it is…) but not sure this proof…
One difference between AI cheating and traditional cheating is that many students (at least now in 2026), even the best students, do not see why they shouldn’t make use of AI to do as much of their school work as possible. Students may also have adults in their lives like our own @ChoatieMom who hold the opinion that teachers should fully embrace students’ use of AI, rather than forbid or limit its use, and that the problem is not “cheating” but instead should be framed as outdated teaching methods.
In my D26’s AP Physics C class this year, for example, she said that most kids in the class were using AI to do their schoolwork during class time, during tests, and in every other context that they could, without the slightest shame about it (and the inexperienced teacher was simply flummoxed by the situation). In the fall, kids from this class will be attending Ivies, MIT, Stanford, UC Berkeley, etc. and most of them openly plan to use AI as much as they can get away with, regardless of the policies at their school; they even talked about this in class.
Using AI in school work is somewhat analogous to the use of calculators in math courses in the old days. Calculators are certainly useful in solving math problems, but the user needs to understand the math in order to know whether the result from the calculator is likely to be correct (as opposed to an incorrect answer based on miskeying the input values or framing the problem incorrectly).
The problem is that this is not always true. Yes, you need to understand math concepts to know how to use calculators correctly. The calculator is intended to help the user do more complex math with this foundational knowledge. And there might be some disciplines in which AI might work the same way. But students use AI indiscriminately, and in most disciplines, AI use does not depend on fundamental understanding of the discipline or material. If you can just plug in an essay question and a few sources and cut and paste whatever AI spits out, that’s not the same as using a calculator for math. Someone who uses AI to write an essay will not only never know how to write, but they also will never learn the foundational reading and analytical skills necessary for coherent thinking and interpretation. And as much as I sympathize with teachers who rely on in-class writing and oral exams to AI-proof their assignments, one result of this shift is that students are not learning to do the kind of sustained and focused reading, research, and writing that can’t be confined to class periods.
Perhaps teachers should include a question on a test given shortly after students turn in their papers: “Give a summary of the paper you turned in yesterday.” (Or something more specific to the topic of the paper.)
It isn’t because you-the-human need to input something into the calculator. If (like me during a physics exam) you divide instead of multipliying, the calculator provides a result but you fail. You need to actually understand what you’re being asked, conceptualize the process, in order to formulate a use for the tool.
AI does it. You enter what you got and it produces something. You don’t need to know or understand anything. The type of exam questions or exercises students enter into AI result in answers that “work”.
It’s analogous to dictating vs. writing. We haven’t stopped teaching holding a pen and handwriting because it’s the foundation for what comes next, even if presumably you could have children dictate stuff and never write a thing, as well as never read a thing. We’d create an entirely illiterate society. We know it’d be bad, even if adults dictate things in their every day lives.
We need to separate the learning process from an AI result clearly so that young people understand the process matters more.
The AI user still needs to understand the problem in order to frame the question to AI properly, as well as have some chance of knowing whether the AI answer is a hallucination based on common faulty information that it used as its training data.
The analogy still holds – just as a student needs to learn mathematical operations without a calculator before using a calculator to do those faster, a student needs to learn reading, research, and writing skills before using AI to do those faster.
The analogy might hold when you’re talking about some kinds of data processing (which is what a calculator does), but it does not hold when you’re talking about reading, research, writing, and reasoning. It just doesn’t – not for any discipline in which the process is at least as important as the product. If AI circumvents the process, then it invalidates the product.
Not when students are simply pasting a question from homework or an exam into AI. It can also slurp up all the course materials for context (textbook, syllabus, homework, past exams).
Kids concerned about their learning will put their course materials into the AI, and then ask the AI to teach / coach them through problems. But even well-meaning kids who are trying to learn the material can get frustrated or tired and ask the AI to go ahead and solve the problems when they get stuck, when problems get hard, or when things are graded. My daughter observed this multiple times with her friend who will be attending MIT. It remains to be seen whether this level of AI usage will serve the friend well at MIT, or will interfere with her learning and leave her unprepared when she has to struggle with a really hard problem unassisted or take a proctored exam.
The question is framed by the teacher to help the student process the concept and build understanding or skills or help structure the essay (etc).
Writing questions is a very important part of teaching - especially in middle/high school+ introductory college courses.
The issue is different when you reach an assumed level of conceptual autonomy that students can just frame their own question and proceed on their own.
However with AI, students don’t feel the need to understand the problem or to frame it.
They just copy/paste the question then copy/paste whatever they got.
Even if there is no hallucination, the purpose of learning is moot.
But the students focus on the results and don’t understand there’s a problem if they didn’t go through the process of writing themselves.