What children should learn when AI already knows everything
Knowledge is getting cheap. Thinking is not. Why maths and physics matter more right now, and what that looks like on an ordinary afternoon.
The question comes up in almost every conversation between parents, usually near the end: what is my child actually supposed to learn? If a machine hands over every date, every word and every formula in two seconds, why is the child still sitting over homework?
The answer starts with a distinction. Knowledge is getting cheap. Thinking is not. What an AI has in abundance is retrievable information. What it does not hand over is a mind that notices when the answer is wrong.
Why maths and physics show up in the answer
Andrej Karpathy, formerly at OpenAI and Tesla and now running his education company Eureka Labs, set out a clear priority on the No Priors podcast. Subjects where you have to transform and reason through something matter more than subjects where you mainly have to retain something. At retaining, machines already beat us. Maths, physics and computer science he calls the best foundation for thinking:
it's just like the best thinking skill core
The point is not the subject, it is what the subject does to a mind. Studying physics means translating a piece of reality into a model, testing that model, and throwing it away when it does not fit. Anyone working with an AI needs exactly the same move: form a hunch, hold the result up against reality, ask again, discard.
What this looks like at home
None of this adds up to a tutoring plan. It adds up to a habit, and you can practise it on an ordinary afternoon.
- Estimate first, ask second. How many litres fit in the bath? Let them guess, then work it out, then ask the AI. The order is the whole trick.
- Take the answer apart. Not just: is this right? But: where does the AI claim to know this from? Children find the gaps faster than you expect.
- Build something instead of looking something up. A game, a comic, a small program. Building shows you immediately what you had not understood.
- Leave the mistake in and hunt for it. The moment something fails is the moment learning happens. The AI may help, but only after your own attempt.
- Collect questions rather than answers. Write questions down for a week and it becomes obvious that good questions are the scarce thing.
You are a patient physics teacher for a ten year old. I am about to ask you an everyday question. Do NOT answer straight away. Instead, ask me three questions that lead me to make my own estimate. Only once I have estimated do we work it out together, and you tell me where my estimate was good and where it was off. My question is:Sources
FAQ
So should children stop learning history and languages?
No. This is about weighting, not banning. Subjects built mainly on retention lose their edge to AI; subjects that require transforming and checking gain ground. A language, in any case, is not learned for the vocabulary alone.
Do I need to be good at maths myself?
No. The five points above work without any technical knowledge. Estimate first, then work it out, then question it: you can guide that even when you do not have the answer yourself.
From what age is this worth doing?
As soon as a child can estimate and give reasons, so roughly from eight. Younger works too, just without the arithmetic: guess, look, be surprised. The move is the same.