The Lesson She Would Not Hand To The Machine

On Wednesday, WKAR Public Media in East Lansing profiled a group of teachers gathered at Michigan State University for the third year of its Holocaust and Genocide Curriculum Fellowship. Among them was Bethany Cordle, who teaches in Lapeer County, a rural stretch of thumb-side Michigan.¹

Cordle is not a skeptic of these tools. She uses AI to write lesson plans. She permits her students to use it. She has built a multi-step system for how they may use it, including limiting the sources a chatbot may draw from and teaching them, directly, how to spot the moment the machine is inventing a fact. What she will not do, she told the reporter, is use AI to teach genocide or the Holocaust. The topic is too serious. The misinformation around it is too dense. She does not want a student walking away with the wrong lesson from a machine that was confident about the wrong thing.¹

Why This Line, And Not Another

The line Cordle drew is not new to anyone who has spent a year with a chatbot. Every teacher who uses these tools eventually finds the topic where she stops. Sometimes it is a family in her class whose story is close to the material. Sometimes it is a discipline where a wrong fact travels for a decade. Sometimes it is a piece of history dense enough with denial that even a small, confident error from the machine, dropped into the middle of a fifteen-year-old's understanding, would take years to walk back.

The Holocaust is one of the clearest cases. In 2024, UNESCO and the World Jewish Congress published a report warning that generative models, trained on the open web, inherit denial and distortion alongside history. The report catalogued cases of chatbots fabricating events that never happened, generating a false quote from Joseph Goebbels claiming he had tried to save Jewish lives, and producing deepfaked audio of a Harry Potter actor reading from Mein Kampf. The report's recommendation was not a ban. It was that education systems teach the next generation to hold a piece of confident-sounding text at arm's length long enough to check it.²

Cordle's classroom rule is a version of that recommendation, arrived at through her own reading of her own students. She lets the machine into the room for most things. When the material is one where a hallucination would compound the harm, the machine stays out. It is a small, specific act of judgment. It is also the thing an AI purchasing policy cannot write down.

The Second Job Of A Teacher

A teacher does two jobs. The first is to deliver the material. This is the one every tool on the market is trying to do faster, cheaper, and more patiently than any adult can. The second is to notice which student, at which minute, needed a different question. This is the job that has never been automated, because it depends on knowing the child in the room, and because knowing is not a task.

The tools that are moving fastest right now, and selling hardest, are the ones that lean on the first job. Draft the plan. Grade the paragraph. Answer the follow-up question at ten at night. All useful. What Cordle's line makes visible is that a teacher who is any good already knows, quietly, that the second job is what she is actually paid for. The AI can help her prepare. It cannot decide what her seventh-period student is not yet ready to be told by a machine that does not know him.

What The Classroom Needs To See

The reason this small story matters, on the second day of school in most of the country, is that it is the shape of the next argument. Not whether AI belongs in classrooms, which has been settled by adoption. Not whether it can help with the first job, which it clearly can. The next argument is about the second job. What does a school need to be able to see so that a teacher who is spending her afternoon on the first job with a tool is still holding onto the second one?

The blue book of last week's essay saw the finished page. It did not see the fifteen minutes the student spent staring at the second paragraph before rewriting the whole thing. A quiz score sees the answer. It does not see the moment the student almost got it and then, at the last second, chose a different one. A conference in October sees a mother across a table. It does not see the six revisions her daughter made to the same sentence in September, or the one she deleted because she thought the teacher would not like it.

At Koan we spend our days building the layer that does see those things. Not because seeing them replaces the teacher. Because it puts back, into her hands, the raw evidence of the second job. Not a probability score. Not a summary. The record of the child's thinking as it happened, so the teacher who has been rescued from an hour of grading can spend that hour asking the question no chatbot could have known to ask.

What Cordle's Rule Suggests

Bethany Cordle's rule is a small one. It will not appear in a state framework. It will not be printed on a purchase order. It is exactly the kind of judgment schools will need more of in the next five years, not less. A teacher who has already decided, on her own, where the tool stops being useful is a teacher who has been paying attention to her students. The schools that thrive will be the ones that make it easier for her to keep paying that kind of attention. Not by adding more tools. By making sure the ones already in the room leave a trail she can read.

Where in your building has a teacher already decided, quietly, that the machine does not belong? What did she see that the policy did not?