The Open Thing

One year I had a kid who would not, or could not, start working until I told her exactly what I wanted. Hand the class a prompt with any room in it, anything where the right answer was not already spelled out, and she was at my desk inside a minute. How long did it have to be? Could it be about anything? Was this what I wanted? She needed the whole thing narrowed to a single safe target before she could write a word, because an open thing is a place you can be wrong, and she had learned a long time ago not to be wrong where an adult could see it. So I would narrow it for her, and she would hand back exactly what I narrowed it to, clean and on time and empty, and I would mark it complete.

I thought about her when I read the new piece in The Atlantic, “AI Can’t Fix the Student-Motivation Problem.” Jenny Anderson and Mike Goldstein walk through the promise first. In 2023 Sal Khan stood on a TED stage and called AI the biggest positive transformation education has ever seen, with Khanmigo, his chatbot tutor, pitched as a personal tutor for every student on the planet. A year later Sam Altman was promising the same thing, a tutor in any subject, any language, at whatever pace they need. By this spring Khan was saying his tutor had been, for a lot of students, a non-event. Access went from forty thousand students to almost a million. Usage did not move.

The authors land on the reason, and it is the right one. A tool that answers questions and asks good follow-ups cannot help a student who will not engage, or who does not know what to ask. It waits to be used. The ones who need it most are exactly the ones who will not open it.

The article is sometimes right, and I want to stay with the part it leaves unexplained. It says a reactive tutor cannot reach a student who will not engage. The question still sitting there is why so many will not. We talk about engagement as if it were a setting, switched on in some students and off in others. It is not a setting. It is what students do when someone engages them, and much of school does not engage them. It manages them.

It has to. Managing thirty twelve-year-olds is how one adult survives the day, and managing thirty of them costs less than meeting one. And I do not mean that as an indictment of teachers. Management is not a moral failure. Some days it is the only thing standing between a room and chaos. The trouble is that a system can get very good at what keeps the room quiet and still be bad at what helps a student become braver.

So we manage, we ask them to produce, and we reward the one who hands over the cleanest work with the least trouble. Twelve years of that teaches a lesson underneath all the other lessons: comply. Find out exactly what they want, give it to them in that shape, and you will be left alone. Compliance stops being a choice and turns into an instinct.

That is not nothing. Some days compliance is the doorway. A kid does the shape first and finds the thought later. But a doorway is not a destination, and school has a way of mistaking the quiet finished thing for the whole trip.

Engagement is the opposite response, and it is expensive in the exact place compliance is free. It asks her to sit with the open thing a little longer than is comfortable, to risk being wrong, to want more than the assignment asks, to act like she has a say. That is agency, and it is what the managing wears down, a little more every year. I wrote here once that compliance is taught and agency is trained out of you. Now AI is being handed to a system already built around that bargain.

Mostly, a room teaches what it rewards.

Which means the girl at my desk was not failing to learn. She had learned the real lesson cold. She found out what I wanted and gave it back to me with the least exposure she could manage, and I rewarded it with a good grade, the way the system trained us both. She was not checked out. She was complying, expertly, because complying is what we had taught her to do.

This is also why the tutor was a non-event, and why the next reactive tutor probably will be too. AI only multiplies what is already there. It cannot put agency back where school has spent years wearing it down. Point it at her and it does not wake her up. It makes the cheaper deal cheaper.

Point a model at the compliance instinct and you make that deal free. A finished essay in nine seconds, clean, sourced, in her own approximate voice, run through a humanizer so it reads as hers. Compliance now costs nothing at all. The distance between handing in the work and doing the thinking gets wider than a teacher can see, and you graduate a student who was managed her whole life and never once asked to have a say. That is the real nightmare. Not a lazy student. A perfectly compliant one, with a machine standing by to narrow the world to a single safe target for her before she ever has to choose one.

Aim it at the part of the day that was eating me, and the promise is smaller. It does not teach for me. It gives me a better chance to notice who needs me next.

For a long time the math just sat in front of me. Thirty students, one of me, and the part I most wanted to be there for, the one willing to stay in something unfinished, was the part I had the least of myself left for. Managing ate the front of my attention. Engaging got whatever was still standing by sixth period, which was not much.

So I built something for my own room. Not a tutor that waits to be asked. It reads the work as it comes in and gets a first response back before the waiting has gone cold, and it shows me what I can never see from the front of the room, which is who is actually wrestling with the work and who is handing it back to me exactly the way I assigned it.

It misses things I would catch. It cannot hear the way a student says “I don’t know.” It cannot tell whether a quiet desk is concentration or surrender. I do not trust it with the relationship. I trust it with the first pass, the record, the pattern I would otherwise miss until too late.

That is the narrower promise. It takes the parts I can safely hand off, the first-pass feedback, the tracking, the reminders, the paper trail, and gives me back the attention that was going into them. Most days, attention is what I run out of first.

So the article is right, and right about more than it claims. No bot is going to fix motivation, and the reason is not that we have not built the right one yet. Motivation was never a software problem. It starts after students have been treated, over and over, like their answers are worth waiting for. If we cannot give them that room again, no tutor is going to save us, and we will keep acting surprised every time the next one does not. AI can buy a teacher the time to do the human part, or it can automate the managing and call it school. Which future we get was never up to the tool.

The girl who needed everything narrowed is easy. She will be fine, in the way compliant students often look fine and never quite get what they paid for. It is the other one I think about, the student still working at the bell with one sentence in twenty minutes, the rest of the room already lined up at the door, because the prompt had stayed open and they had stayed with it. That is where learning starts. For years I waved that student along with everyone else. Now, sometimes, I get to pull a chair over and sit down.