What the Room Believed

On Monday, the Hechinger Report ran a column co-written by a University of Chicago undergraduate named Kristin Fasiang and Jill Barshay, the paper's longtime Proof Points columnist.¹ Fasiang had distributed an anonymous survey to three hundred and thirty-eight of her classmates. Sixty percent of them said they personally used AI tools like ChatGPT for their schoolwork. Ninety percent said they believed the average student on campus did.¹

The thirty point gap is the story. It is the same gap, in a new room.

An Old Misreading, A New Subject

The phenomenon is old enough to have a name. Sociologists call it pluralistic ignorance, the quiet condition in which a roomful of people each believes the others are doing more, or wanting more, or feeling more comfortable with a behavior than they actually are. The classical study was done at Princeton in 1993 on undergraduate drinking. Deborah Prentice and Dale Miller found that most students were privately uncomfortable with the campus drinking norm. Each thought every other student was fine with it. The norm nobody held became the norm everyone enforced.³

The same pattern has been documented in studies of sex on campus, of charitable giving, of climate concern, of how often a colleague answers email after hours. The pattern is so consistent that it is now its own small library, all of it showing that the imagined room is almost always different from the actual one, and that students will quietly conform to the imagined room before they realize it is imagined.

The Hechinger column adds AI to the list. The sixty percent who use ChatGPT do not know they are sixty percent. They think they are ninety. The forty percent who do not use it think they are alone. They are not alone. They are forty percent.

Why the Misread Matters

The reason this matters is not that students are lying on a survey. It is that the imagined norm is the norm that drives the behavior. KQED's MindShift, in its coverage of the survey, put the consequence cleanly. A perception of widespread AI use, left uncorrected, becomes self-fulfilling.²

The teacher cannot see who is using the tool and who is not. She sees the assignment, which has been folded and washed and pressed into the shape it has always arrived in. The students cannot see each other's work either, only its surface. The signal each of them uses to decide what is normal is the signal they themselves produce in public, and the signal in public is always the polished one. The classroom each of them thinks they are sitting in is the classroom no one is actually sitting in.

A Future That Reads the Room Honestly

The classrooms that will be sturdy through the next decade are the ones that find a way to make the actual room visible, not the imagined one. Not as surveillance. As a corrective to a misreading the imagined room produces on its own.

This is the part of the question Koan was quietly built for. Not the rule about whether AI is allowed on the next paper. The honest record of how the paper came to be. The pause before the student accepted the model's first sentence. The revision after she rejected the next one. The hour she sat with the problem before she opened the tool, and the half hour after she closed it. When that record exists, the student is able to see that other students sat with the problem too. The teacher is able to see that the imagined norm of frictionless AI use is not the actual one. The room is able, finally, to read itself.

The thirty point gap is not a moral problem. It is a measurement problem. Schools have always known how to gather the products of student work. They have not, until now, known how to gather the trail of it.

Fasiang ended her column with a small note about her classmates. They feared, she wrote, being seen as the one student who could not do the work without help.¹ That is the human fact. The technical fact, sitting underneath it, is that they cannot see anyone else doing the work either. The first is durable. The second is solvable.

If the actual ratio in your building is six in ten and not nine, would anyone there know it?