Ninety-Six, Then Forty-Eight

A Brown economist gave his class a take-home midterm as an act of mercy — the room had just lived through a shooting. Forty of eighty-six students came back with a perfect 100, on an exam harder than usual, and the class averaged 96. Weeks later he sat the same students down in a room, took the machine away, and gave them the same kind of test. The average fell to 48. Nineteen failed. Nothing about the students changed between those two numbers. The only thing that moved was whether the cord was in the room.

Roberto Serrano has taught economics at Brown for decades — he holds an endowed chair there. Last December a gunman opened fire on his campus. Nine people were wounded. One of the dead was Ella Cook, a student who, just days before, had asked Professor Serrano to be her academic advisor. So when the spring midterm came around, he did what a decent person does for a room full of kids who had just watched that happen. He made the exam take-home. Do it on your own time. Breathe. I trust you.

That is the part I want you to hold onto, because everything else follows from it. This did not start as a story about cheating. It started as an act of mercy.

The exam went out March 5th. It was, by his own account, harder than the ones he'd given in years past. Eighty-six students took it. Forty of them came back with a perfect score. The class averaged a 96 — on a test that in normal years landed somewhere between 65 and 80. On paper, the most traumatized class he'd ever taught had just turned in the best performance of his career.

It fell apart in the grading. The same convoluted line of reasoning kept showing up on exam after exam — the same detour to the same answer, when the actual solution was shorter and cleaner. His graders ran the questions through ChatGPT and got back the detour. Not the elegant path a person finds. The machine's path, copied out by dozens of hands that never noticed it was the long way around.

Here is the number that matters, and it is not the 96. Serrano confronted the class. Twenty-seven students dropped the course — twenty-two of them had scored a perfect 100. Then he gave the final in a room, in person, with no machine to reach for. Fifty-nine students showed up. Nineteen of them failed. The class average collapsed from 96 to 48.

“Ninety-six with the machine. Forty-eight without it. Same students, weeks apart — the only thing that moved was whether the cord was in the room. The gap between those two numbers is the size of the tether, printed as a grade.”

Same students. Same material. Same professor. A few weeks apart. The average cut in half. Run the experiment yourself and you cannot escape what the second number is measuring. It is not measuring how much they cheated. It is measuring how much was actually in there when you took the tool away. Ninety-six with the cord. Forty-eight without it. The gap between those two numbers is the size of the tether, printed as a grade.

This is the cleanest data I have seen on the thing this whole site is about, and it arrived by accident, in the worst possible way. We spend most of our time here arguing about dependence in the abstract — is it real, is it a big deal, is it just old people afraid of a new thing. Serrano ran the controlled trial without meaning to. Give people the machine and they look like geniuses. Take it away and you find out how much of the genius ever belonged to them. For half his class, the honest answer was: about half.

And notice what the tool did to their judgment on the way in. Forty students turned in a 96-level performance on a harder-than-usual exam, in the same semester their campus buried a classmate, and apparently not one of them looked at that result and thought this is not me. The cord doesn't just do the work. It quiets the small internal voice that would otherwise say — you did not actually earn this. That voice is the whole thing. That voice is what an education is supposed to build. It is the first thing the tether takes.

Serrano did not stay quiet, and that is the other reason this story matters. "The empirical evidence of fraud is overwhelming," he said. "Silence is the worst treatment for this problem." He is right, and the temptation to stay silent is enormous — because to say it out loud is to admit the mercy got used against him. He extended trust to a grieving room and a chunk of the room fed his questions to a machine. A smaller man would have quietly gone back to blue books and never said a word. He put a number on it instead. Ninety-six, then forty-eight. Say it out loud.

His hardest line — "humanity has chosen to become idiots" — reads like despair, and people are quoting it that way. I don't think that's what it is. It's the word chosen doing the work. Nobody forced a Brown junior to paste an exam question into a chatbot. There was no proctor to beat, no clock to outrun, no gun to their head — the opposite; the whole point of the take-home was that the pressure had been lifted. They had the room, the time, the trust, and the ability. And still, given the choice between doing the work and having it done, forty of them let the cord decide. That is the part that should scare you. Not that the machine is strong. That when we are handed mercy and told do this yourself, we increasingly answer: I'd rather not.

The 48 is who they were the whole time. The 96 was the cord talking. And the only reason we know the difference is that one professor, in the middle of a semester nobody should have to teach, put them both on the record and refused to be quiet about the gap.

We argue about dependence like it can't be measured. It just was. A Brown class averaged 96 on a take-home and 48 on the in-person final — same students, same material, weeks apart, and the only variable removed was the machine. That 48-point drop is not a cheating statistic. It's a portrait of how much of the competence was ever theirs to begin with. Roberto Serrano gave that class a take-home out of mercy, after a shooting took one of his advisees, and he watched the mercy get handed to a chatbot. Then he did the one thing that keeps this from happening quietly forever: he said the number out loud. Silence is the worst treatment for this problem. The 96 was the cord. The 48 was the kid. Learn to tell them apart before the only test that ever removes the cord is the one that decides your life.