Somewhere between a kindergarten classroom in Dubai and a fifth-grade classroom in Brooklyn, the world’s education systems have quietly split into two camps. One believes children should grow up fluent in artificial intelligence the same way they grow up fluent in a keyboard. The other believes some kinds of struggle – sounding out a word, wrestling with long division, staring at a blank page – are the actual curriculum, and AI is simply too good at skipping them.
This isn’t a niche policy dispute. It’s a live, uncontrolled experiment being run simultaneously on tens of millions of children, and right now nobody running it agrees on what they’re testing for.
The UAE bet: fluency first
Start with the most aggressive wager on the table. The UAE has made artificial intelligence a mandatory subject from kindergarten through Grade 12, trained tens of thousands of teachers to use it, and is now exporting its curriculum to other countries’ ministries of education. The logic is unapologetically economic: a country building its identity around being a global AI hub cannot afford a generation that treats the technology as a novelty. Five-year-olds meet AI through stories and play; by the time they’re teenagers, they’re expected to probe its biases and question its outputs rather than simply prompt it.
What’s easy to miss in the headlines is that the UAE isn’t actually reckless about this. Ministry rules bar the youngest students from hands-on generative AI use even as the curriculum teaches them about it, and the educators actually implementing the mandate keep returning to the same worry: that the hard part was never getting AI into classrooms, it’s making sure students don’t mistake fluency with a tool for actual thinking. One school has built a matrix forcing students to disclose whether a piece of work was human-only, AI-assisted, or AI-driven, an attempt to keep the “how” of AI use visible even as the “whether” stops being a question.
The Norwegian bet: not yet
Norway looked at the same technology and drew the opposite conclusion. Citing concerns over declining educational performance and the risk of cognitive offloading – the tendency to let a tool do work your brain would otherwise be building the capacity to do – the government has introduced a near-ban on generative AI for children in grades one through seven, roughly ages six to thirteen. Supervised use returns in lower secondary school; broader use comes later. The government is also putting more money into physical books, not fewer.
This isn’t technophobia. Norway had already moved to restrict phones in schools, and research behind that policy found that smartphone bans were associated with less bullying and better academic outcomes, particularly among girls. The government has since doubled down on the idea that schools should be places where children can concentrate, learn and interact without constant digital interruption. The AI restrictions read less like fear of the future and more like a bet that certain cognitive foundations, reading, arithmetic, the willingness to sit with a hard problem before reaching for help, have to be poured in concrete before you build anything on top of them.
New York and Los Angeles: caught in between
Then there’s the American response, which looks less like a philosophy and more like a scramble. New York City’s new one-year moratorium on student-facing generative AI, covering everyone from pre-K through eighth grade, was announced days before the school year began. It bans companion chatbots outright at every grade level, permits only a small, tightly supervised high-school pilot, and pairs the whole thing with new screen-time caps. Mayor Zohran Mamdani framed it as pedagogically consistent if schools exist to teach children to question and interrogate, AI shouldn’t be exempt from that same scrutiny. Critics, including the city’s own teachers’ union, note that a year is a strange amount of time to evaluate a technology that’s still changing month to month, and that the policy says little about how the city will vet the safety of what it does allow to reach classrooms.
Los Angeles arrived at nearly the same place by accident. Its AI restriction wasn’t announced as a standalone policy at all – it surfaced almost as a footnote during a committee meeting, the byproduct of a broader screen-time crackdown that began with a 2024 phone ban. Some school board members admitted publicly they hadn’t realized the district’s own policy had already changed. That’s not a coherent theory of childhood cognition. That’s an institution flinching.
The real fault line
Strip away the geography and the four approaches split along one honest disagreement: does AI competence belong in the category of skills you build early, like reading, or in the category of tools you’re granted access to once you’ve already built the underlying skill, like a calculator handed out only after arithmetic is secure?
The UAE is betting on the first. Norway is betting hard on the second. New York and Los Angeles haven’t really placed a bet at all – they’ve reached for the brake because nobody can prove yet which model is right, and pressing the brake is the only move available to a system that isn’t ready to gamble with 600,000 children’s foundational skills.
That caution is defensible. But it isn’t a philosophy, and dressing it up as one risks the worst outcome of all: kids who lose a year of AI literacy without gaining the deliberate, evidence-based skill-building Norway is actually trying to protect.
If there’s a lesson sitting underneath all four experiments, it’s this: the question was never really “for or against AI in schools.” It’s whether a child should learn to think with a tool that can think for them, or learn to think well enough, first, that the tool becomes something they use rather than something they lean on. Every country currently running this experiment is really just guessing at the order of operations. The only mistake would be pretending any of them already knows the answer.
Sources: Dawn Images, Gulf Business, Rus Tourism News, Khaleej Times, Euro News, Sun Journal
