Two directions, same calendar
Look around the world and you mostly see two reactions.
One is pullback. Sweden spent years pushing screens into schools, then noticed reading scores slipping and started returning to paper, handwriting, printed books. France and the UK keep tightening phone bans. AI mostly shows up there as a cheating problem, so the answer is supervised exams again and more control.
The other reaction is integration. South Korea and Singapore are building adaptive learning around AI. In the US it depends on the state, but more places are asking how to teach students to use AI well, not how to pretend it does not exist.
Both sides run into the same practical questions. What is homework for if a chatbot can finish it in three seconds? Do students lose thinking skills, or move faster? And who can afford the good models, and who cannot?
Same curve, different starting point
These countries look like opposites. They may just be at different points on the same curve. I have seen this pattern before: the first wave of enthusiasm, the correction, then the people who lagged behind suddenly pushing hard.
Where schools overdid screens, there is correction now. Where they underinvested, there is catch-up. Pendulum, hype cycle, call it what you want.
Tablets and smartboards are one thing. AI doing the student’s thinking is another: writing the essay, generating the code, working the proof. That is why this round feels different from the last one.
In much of Asia the story is not “we just discovered laptops.” The systems were already intense and test-heavy. AI there is less a classroom novelty than a way to keep up. Different lane entirely.
Memorization school vs competency school
It’s not just digitalization that has two sides. The way we teach divides into two camps too.
In a memorization school you drill facts, recall dates, solve the standard problem. Lexical knowledge. Frontal teaching. Tests that reward what you can repeat. AI breaks that model fast. Essay on a historical battle? ChatGPT has it in seconds. The school responds with bans, plagiarism tools, proctored paper exams. Everyone chases everyone else.
We are still grading students on the kind of work machines are already better at. We test facts on command, template essays, and mechanical tasks.
Competency school looks different. More argument, more projects, more “show me you understand this.” Finland gets cited a lot here, and the idea is spreading elsewhere. In that world AI can sit in the classroom without automatically being the enemy. Generate a draft, then tear it apart. Argue with the bot and see who wins. Use it to organize research, but present the conclusion yourself.
You do not move toward that model because it reads better in a brochure. You move because the old way of measuring lexical knowledge stopped working the day the chatbot went mainstream. People in education research have said this for years. The curriculum just had no reason to hurry until now.
Who is supposed to fix this?
Teachers and students are often left to improvise. Kids are already using something. Teachers too, sometimes openly, sometimes not. The big AI companies are mostly building for business: code, analytics, support tickets. Fixing school is not what they are building for.
Maybe they will self-regulate for children. I would not bet on it.
That pushes the problem toward lawmakers, who tend to arrive late. Social media is the obvious parallel: in 2026 we are getting serious about platforms that showed up in the early 2000s. The EU AI Act is real. Italy’s early ChatGPT ban happened. But countries want different things, and everyone is afraid that being too strict means falling behind.
Even the “ethical AI” story mostly converges on the same pressure. Building a kid-safe foundation model is expensive. Child privacy law is a minefield. The interesting school tools are more likely to come from education companies wrapping existing models, not from the model labs themselves.
Give them a school AI instead of a ban
The idea I find most workable is boring on purpose: an education-specific wrapper around a general model. Not raw ChatGPT in the browser. Something tuned for school, with limits built in. Khan Academy’s Khanmigo is the example people point to. It nudges with questions instead of dumping answers, and it can flag a conversation that went off track.
Three reasons that matter.
If you ban ChatGPT on the school Wi-Fi, the student opens it on mobile data at home. Give them an official tool for assignments and the use becomes visible instead of hidden.
Raw models want to be helpful fast. A school wrapper can be built so the thinking stays with the student, instead of making the plagiarism checker the whole defense.
And if a district buys one system for everyone, it matters less whose parents pay for the premium subscription. Same tutor, same guardrails.
Maybe then teachers could spend less time playing AI police.
Bad news for memorization school. Good news for competency school. The tech is not waiting for the syllabus to catch up.
Banning is easy to announce and a free-for-all does not work, so the middle path is intentionally boring: one bounded school tool for everyone, and honest teaching about what it is good for and what it is not.
What does it look like where you are? Is there an official school AI students are actually allowed to use?