A few things caught my attention this week, and they sit surprisingly well together: the very different ways governments across Asia are introducing AI into schools, some predictions about how AI hiring will change over the next year, a mathematical result produced by a machine, and a move to give young people a seat at the AI policy table. Taken together, they say something about how carefully, or how quickly, we are choosing to move.
Different bets on AI in schools
An ABC piece on how Asia is bringing AI into schools is really a study in contrasting strategies. China has made AI education compulsory, with eight hours a year in Beijing across primary and secondary. Singapore is aiming for a population that is bilingual in a different sense: fluent in a profession, and fluent in applying AI to it. South Korea went furthest in its language, calling AI talent a national survival strategy, then abandoned its AI textbook rollout after it moved too fast for teachers to keep up. Japan, by contrast, is staying neutral and waiting on the evidence.
I find the range instructive. There is no settled consensus here, only different bets placed with different levels of confidence. The South Korean reversal is the one worth sitting with: ambition that outruns the capacity of teachers to absorb it does not accelerate anything. It stalls.
The shape of AI hiring
Some predictions on AI hiring over the next twelve months. The picture splits hard by geography: countries making AI keep hiring deep into the infrastructure and frontier-research stack, while countries implementing AI, Australia included, hire a very different shape of role.
- Head of AI and Chief AI Officer are diverging into two roles: the CAIO sits at the board table and owns risk appetite and prioritisation, while the Head of AI runs the operating layer, closer to product and engineering.
- AI sys admins: the first wave of enterprise AI lands on existing IT teams, covering licensing, provisioning, identity management, data loss prevention and acceptable-use policies.
- AI engineers: misnamed, in my view; this is a science role wearing an engineering title, empirical rather than deterministic, and you cannot unit-test your way to a working agent.
- Forward-deployed engineers: technical enough to ship and human enough to listen, with the bar high on both axes at once.
- AI governance, compliance and red team: EU AI Act high-risk obligations drag every multinational into compliance hiring, and the scarce skill is people who can read legislation, a model card and a deployment architecture and connect all three.
- Change managers and process redesigners: the shortage is people who understand what the technology can do, how teams adopt new tools, and the documentation work to release tribal knowledge.
The next twelve months will not be engineering followed by change management. It will be both at once, and the organisations that get the human side right will pull decisively ahead.
A machine cracks an old problem
An AI model produced a counterexample to a 1946 Erdős conjecture that mathematicians have worked on for 80 years. The detail I keep returning to comes from the Fields Medallist Timothy Gowers, who said he would have recommended publication without hesitation if a human had submitted it. You can read the account in The Conversation.
Research mathematics depends on three things: deep expertise, sustained effort across many speculative lines, and the occasional conceptual leap that reframes a problem. AI might be genuinely strong at the first two. Whether it can do the third is still open, and there is a quieter question underneath it: did the model succeed here by exhaustively exploring a space that a human would have reached more directly? Impressive either way, but the two are not the same thing, and the distinction matters for how we describe what these systems are actually doing.
Young people in the room
The Children's Commissioner for England is recruiting a youth advisory board for the Department for Education, giving 16 to 18 year olds a way to share their views on AI in education. It is a small thing, but the right instinct.
AI can help teachers, support personalised learning and improve outcomes. It also has to be used safely, and it must not replace the human relationships at the heart of education. The people best placed to notice when that line is being crossed are the students living inside these systems every day. Building them into the policy process, rather than consulting them after the decisions are made, is how you keep the technology in service of the people it is meant to help.