What Should Children Learn to Do Without AI?

On AI-assisted schoolwork, Chinese pupils are running rings around the rest of us. Their marks have climbed 18% over the past two and a half years, and the time they spend preparing for lessons has fallen by a third.

The researchers then set those same pupils an exam with no access to AI, and yes, you guessed right, their scores fell by 20%. Metacognitive laziness, as it is now called, has arrived. Technology charges its usual fee: dependence on one side, and expectations, mine and everyone else's, of easy wins on the other.

This was a huge study, covering nearly 27,000 pupils. Worrying? Possibly. But what does it actually tell us?

The Casio calculator watch was the obligatory accessory for every nerdy boy in my year, and I was thrilled the day I finally got one.

I spent entire nights multiplying and dividing on those pinhead buttons. Nobody imagined it would one day be plated in gold so that tattooed men in vests could feel fashionable. Only Ilan in my class insisted on doing the sums in his head, to "sharpen the mind". Today Professor Gronau is deputy dean of computer science at Reichman University, where I am barely a rank-and-file lecturer. It seems to have worked out for him.

But is long multiplication really a skill worth preserving? Summarising a long document? Translating between languages? Walking up to Jerusalem, which my ancestors presumably did (even 100 years ago), is no longer required either. Nor is memorising every phone number and every street name in Tel Aviv.

So which skills are we willing to surrender in order to free up time and attention for higher things? What has to be kept, and what will slip away whether we like it or not?

When change moves this fast there is no real time to understand or measure the human impact of using these tools, or of giving them up. New York’s answer, at the start of the September 2026 school year: game over for now.

In September 2026, New York’s pupils went back to class, only without access to artificial intelligence.

The city has decided to freeze the use of these tools for one school year, from age two right through to eighth grade, which is roughly Year 9 here. That covers 600,000 children.

High schools are heading the other way. Every pupil from ninth to twelfth grade will take AI lessons this year and may use the tools, though they too are barred from chatbots that simulate friendship or emotional support. A further 50,000 pupils will join pilots testing how to fold AI in sensibly.

Teachers face restrictions of their own. Approved tools are fine for lesson planning. AI may not be used for marking, for assessing pupils, for behavioural monitoring or for drafting special educational needs plans.

Why?

The official reasoning rests on protecting human contact, hands-on learning and independent thought, alongside an admission that the research on younger children is still thin. A paper in the Journal of Medical Internet Research notes that little high-quality work exists on AI in education below eighth grade, and that outcomes depend heavily on age and context. It stops short of saying harm has been demonstrated, yet it points at precisely the worry the Chinese study surfaced: cognitive offloading, among teachers as much as pupils.

"The tech industry wants us to believe that AI-based early education is inevitable and even essential," said Mayor Zohran Mamdani. "We do not see it that way."

New York is not alone. Norway has blocked AI access for the first seven years of school since August. Los Angeles has frozen district-owned devices across every age group.

Elsewhere the instinct runs the other way entirely, towards training as many children as possible.

In the United Arab Emirates, which has had a minister for artificial intelligence for about a decade, AI is now a formal subject from kindergarten to twelfth grade. China requires at least eight hours of AI teaching a year, and Estonia gives sixteen and seventeen year olds access to the tools with an emphasis on critical thinking.

And the rest of us? I would happily take both. Children who grow up genuinely understanding what the tool can do, and who reach for it when it makes sense, with a sceptical eye.

An evidence-based set of recommendations is due in April 2027 in New York. The largest school system in America has bought itself a year in which to say "we don't know". The real question gets settled when that report lands. Or perhaps it doesn't.