How Honest Are We When We Use AI?

Two pieces of research this year sit uncomfortably next to each other.

The first is a study from the University of Chicago, by Yier Ling, Alex Kale, and Alex Imas. They asked 338 students a simple question. Do you use AI for your schoolwork? About 60 percent said yes. Then they asked the same students about their classmates. Do they use AI? The number jumped to about 90 percent.

Same students. Same classrooms. A thirty point gap between what people admit about themselves and what they assume about everyone else.

The researchers call it social desirability bias. People answer in the way that makes them look good. When the students were asked to explain the gap, the reasons were telling. Admitting you used AI felt close to admitting you were lazy, or that you could not do the work on your own. So the number gets quietly shaved down, and the real rate of AI use disappears into that gap.

What makes the gap hard to close is that nobody in it is exactly lying. A student who says no is not inventing a story. They are answering a slightly different question in their head, something like, do I want to be the kind of person who used AI. That question has a socially correct answer, and it is not the true one.

The second piece is not a study but a story, and it puts the same instinct under pressure.

An economics professor at Brown University, Roberto Serrano, gave a take-home midterm for the first time in nearly twenty years. His students had asked for it, partly out of anxiety about sitting in a classroom after a shooting on campus, and he agreed it was fair. The scores came back beautiful. The class averaged 96 on an exam that historically lands somewhere between 65 and 80. It was, if anything, too good.

Something felt off. So he moved the final exam back into a room, with a clock and no help. The average fell to 48.6, the lowest he had ever recorded. Eighteen students dropped the course. Of the ones who stayed and sat the final, only a handful matched the scores they had earned at home.

Serrano did not soften what he thought it meant. He said a society cannot afford to have a large share of its best young minds believing that cheating is fine, because that leads to a declining society, a failed one. His words were, "We cannot choose to become idiots."

Put the two together and a pattern appears. One study measures the hiding. The other shows what happens when the hiding has nowhere left to go. A take-home exam did not create the problem. It removed the room that used to keep everyone honest, and what was underneath became visible all at once.

The strange part is how the shame travels.

Ask a child in school today about a teacher who used AI to grade their homework, and the reaction is quick. The teacher is lazy. Few stop to ask whether the grading was any good, or why a teacher with a stack of scripts and a full week of classes might reach for help. The same word the university students used about themselves, lazy, gets handed straight back to the adult at the front of the room.

So the judgment runs in every direction. Students hide their AI use. Teachers get judged for theirs. Parents are not sure what to tell either of them. And almost nobody stops on the quieter question underneath all of it, which is how much real learning actually happened, on either side of the desk.

That question turns out to be harder than it looks. A committee at Brown, writing about AI in teaching, ended up asking things most people have not settled for themselves. Is something in your own words if AI fixed the grammar. Are the thoughts your own if you formed them by talking an idea through with a machine, the way you might with a colleague. There is no clean line, and pretending there is one is part of why the shame sticks around.

None of this stays in school.

The same instinct shows up in offices, in studios, in research teams. People write with AI, run their first analysis with it, think out loud to it, and then feel a small pause before saying so. If the Chicago survey were handed to working adults instead of students, the gap would probably look familiar. The setting changes, the clothing changes, but the hesitation is the same one. It is the quiet worry that using the tool says something unflattering about the person using it.

And that worry has a cost. When people hide how they work, teams cannot learn from each other. When a real rate of use disappears into a thirty point gap, institutions plan around a picture that is not true. The shame does not stop the behavior. It only drives it underground, where it cannot be examined, or taught, or done well.

The interesting question was never whether the machine can write the essay, mark the exam, or draft the analysis. It can, and increasingly it will. The question is what people quietly stop doing once they no longer have to, and whether they notice in time.

Children growing up now will never know a version of this that was simple. They will have to decide, over and over, what is worth learning the hard way even when a shortcut is sitting right there. No adult can hand them a single rule for that, mostly because the adults are still working it out too.

The tools will keep changing. That decision will not move. No survey measures it, no exam room enforces it, and the people doing the work are the only ones who can keep making it, one honest answer at a time.

Yier Ling, Alex Kale, and Alex Imas. "Underreporting of AI Use: The Role of Social Desirability Bias." Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI 2026). https://dl.acm.org/doi/10.1145/3772318.3791073 (preprint: SSRN 5464215).

Emma Whitford. "Brown Professor Suspects Majority of His Class Used AI to Cheat." Inside Higher Ed, 8 July 2026. https://www.insidehighered.com/news/faculty/learning-assessment/2026/07/08/brown-professor-suspects-most-his-class-used-ai-cheat

Brown University Committee on Generative AI in Teaching and Learning. Inaugural Report, 2026. https://provost.brown.edu/sites/default/files/GAITL_Committee_Report_FNL.pdf