For school leaders
AI in Schools: Which Struggle to Protect
Every school I talk to is asking a detection question. How do we catch it, how do we prove it, which tool flags it. I understand the instinct, and it is the wrong question, or at least a question that will never get you where you actually want to go.
The problem is not that students have a new way to cheat. The problem is that the signal a school is built on has quietly stopped meaning what it meant. A good essay used to be evidence of a student who had learned to think, because you could not produce the essay without doing the thinking. AI severs that. The essay improves while the thinking may not, and every grade, every rubric, every promotion to the next level is still reading the artifact as proof of the person.
Under the artifact is productive struggle, the effortful, uncomfortable, slightly-too-hard work where learning actually happens. It is not a nice-to-have. It is the mechanism. Learning scientists have argued for years that some difficulty is generative, that the effort of working through a hard problem is part of what encodes it, an idea captured in research on productive struggle and productive failure. In one widely discussed MIT Media Lab preprint on AI-assisted essay writing, students who used ChatGPT reported lower ownership of their work and struggled to accurately quote their own essays afterward. The output was there. The encoding was not.
Which turns the policy question inside out. The task is not to build a wall against the tool. It is to decide, deliberately and subject by subject, which struggles are load-bearing and must be protected, and which are just friction a student is right to hand off.
A better AI policy starts with four questions
- What capability is this assignment meant to develop?
- Which part of the struggle builds that capability?
- Where can AI support the work without replacing the learning?
- How will we verify that the learning transferred, without the tool?
A school that gets this right will not be the one with the best AI-catching software. It will be the one that got clearest about what it was trying to develop in a student in the first place.
Research behind this
Common questions
- How do we know if students are actually learning or just using AI?
- Not from the finished work, which AI has made an unreliable signal. You see learning in the student under conditions the tool cannot cover: explain it live, apply it somewhere new, defend it without the assistant open.
- Should schools ban AI or teach with it?
- The ban-or-adopt framing misses the real decision, which is where AI sits in the learning process, task by task: protect the tool where output is the point, and protect the struggle where development is the point.
- What is productive struggle and why does AI threaten it?
- Productive struggle is the effortful, slightly-too-hard work where learning is actually encoded. Learning scientists have long argued that some difficulty is generative, not an obstacle. AI threatens it by offering to remove exactly that effort, so the result arrives while the learning it was meant to produce does not.
- What should a school AI policy protect?
- The load-bearing struggles, the ones where the difficulty is the point. Start from what you are trying to develop in a student, then decide where the tool serves that and where it hollows it out.