A Microsoft and Carnegie Mellon study published earlier this year surveyed 319 knowledge workers on AI’s effect at work. One finding cuts deeper than the efficiency numbers: the more workers trusted their AI tool, the less critical thinking they applied. Confidence in the system and delegation of thought moved together.
Those are adults — people who built reasoning skills before the tools arrived. For students growing up with these tools now, the stakes are different. EEG research on students using ChatGPT found the weakest neural engagement of any group tested. When the AI was removed, they couldn’t recall their own earlier work. Researchers named the pattern “cognitive debt.”
A piece in Psychology Today this spring made the asymmetry plain: adults lose skills they’ve already built. Children may never build them at all.
The Choice China Already Made
I teach at an international school in Shanghai, which means I watch two different educational systems handle this question at the same time.
China made its choice early. The Ministry of Education’s 2024 guidance required AI in primary and secondary schools — not as elective enrichment, but as curriculum. By fifth grade, students are learning about intelligent agents and algorithms. The national goal is producing AI-literate citizens who can contribute to China’s strategic position in the technology race.
That debate deserves its own piece. The point is the question of what kind of generation to build got asked — and answered.
In the U.S., states are working through the same question more slowly. Legislators are tracking 134 bills across 31 states this session. Alabama now requires an AI component in computer science for graduation. New York’s state university system is embedding AI literacy into general education for all incoming undergraduates this fall. The policy machinery is moving.
Most of those policies focus on what students are and aren’t permitted to do — allowed to submit, allowed to use, required to disclose. The question of what cognitive habits are being built in the process shows up in almost none of the bills.
The Problem With Good Tools
Highly effective tools have a specific effect on cognitive habits — and that’s where the classroom problem lives.
When AI tools are occasionally wrong, users stay engaged. They check, verify, apply judgment. When a tool is nearly always right, and users know it, confidence in the system becomes a reason to stop thinking. The Microsoft finding makes more sense in this light.
In classrooms, the version of this problem that matters most isn’t the student who submits an AI-generated essay — that’s a compliance issue, and schools have always had compliance issues. It’s the student who uses AI as the first move rather than the last one. Who drafts by prompting instead of thinking. Who outsources the difficult early work — the part where understanding gets built — because the tool is faster.
The Long Game
Teaching through this transition in real time, the question that drives the work isn’t whether students should use AI — they will, in every career and context, for the rest of their lives. The question is what kind of relationship with thinking we help them build before that happens.
Two futures are visible from a classroom in 2026. One is a generation that knows how to use AI tools fluently — to prompt, generate, edit — but hasn’t built the capacity to work through hard problems without them. The other uses AI the way good writers use research: as material to engage with, not a shorthand for the work itself.
The first group will be more efficient in the short term. The second will be better at the things AI can’t do: original judgment, genuine synthesis, reasoning that requires having thought something through.
We’re choosing right now which one we’re building. Not primarily through policy or legislation, but through how we design classrooms and what we ask students to struggle with.
Where Teachers Come In
For teachers, the most direct change is asking students to do their thinking first.
Before prompting, before the AI: five or ten minutes writing what you think. A draft. A list. A position. Something that came from your head before any of the work got outsourced.
A small change in sequence. But the sequence is where cognitive habits get formed. Students who learn to reach for AI before they’ve formed a thought build one kind of habit. Students who use it after they’ve formed a thought — to test it, extend it, push back on it — build something different.
Cognitive debt accumulates when mental effort gets delegated without being redirected. The classroom is where we decide whether that happens.
David Jacobson teaches AP World History and AP US History at Shanghai American School. His book, The AI Doesn’t Know Your Students, is forthcoming from Amazon KDP.
