AI for kids, without the fear or the hype

12 of the 22 schools listed in this directory state that they teach AI. That is counted from their own programme listings, not estimated. 5 of them state that they take children aged four or under, and the youngest stated starting age across the whole group is 3.

So the market is real and it is being sold young. What follows is what the label covers, what it does not, and what to ask before you pay for a term.

What "AI for kids" usually means in practice

Almost none of it is building AI. For children it nearly always means one of three things: understanding in plain terms how a machine learns from examples, using AI tools to make images, music or stories, or judging what the AI produced. Those are three different lessons and they are frequently sold under one word.

The third one is the valuable one, and it is the one most often left out.

The skill worth paying for is checking, not prompting

Writing a good prompt is a shallow skill with a short shelf life. The tools get better at understanding vague requests every few months, which means the thing being taught is being actively designed out of existence.

Checking does not go away. An AI will tell your child it fixed something when it did not, and it will do so in a confident, well-organised paragraph. A child who has been taught to run it, look at the result and ask whether that is really what I asked for has something that holds up. A child who has been taught to phrase requests nicely has a habit that will be obsolete before secondary school.

You can watch that happen in about twenty minutes. We built a short project where AI changes a working program and reports success when it has not, and the child finds it. It costs nothing and it needs no class.

Does a child need to code first?

For using AI and questioning it, no, and the common advice that coding must come first is half wrong. A nine year old can spot that an answer contradicts itself without writing a line of code.

For anything past that, yes. A child who has written a loop can look at generated code and tell when it is nonsense. A child who has not can only look at whether it ran. That is the real dividing line, and it is why the better AI classes for older children still spend time in Python rather than only in a chat window.

What suits which age

Four questions to ask a school

• Does my child build anything, or only prompt something? Both have a place. They are not the same class and they are often sold at the same price. A term of typing requests into a chat window is a term of typing requests into a chat window, however good the results look on the last day.
• What does my child do when the AI is wrong? This is the question that separates the programmes. In a good one, being wrong is the lesson: the child notices, works out which part failed, and fixes that part. If nobody in the room ever says the machine got this wrong, the most useful half of the subject is missing.
• Which tool, and what is its minimum age? Most general AI tools set their own minimum at 13. A class for eight year olds is therefore using either a child-specific tool or an adult account operated by the teacher. Either can be fine. Not knowing which is not.
• What happens to what my child types in? Anything a child types into a general AI tool may be stored and may be used to train it. Ask what the class uses, and tell your child the same rule you would for any other website: no full name, no school, no address, no photographs of other children.

Keeping it safe, briefly

Check the tool's own minimum age rather than the school's. Keep younger children in the room with an adult. Tell your child that anything typed in may be stored, and that the machine can be confidently, fluently wrong. That last one is a safety rule as much as a learning one, because the failure mode is not a rude answer, it is a plausible answer that happens to be false.

In short

AI for children is worth doing and easy to overpay for. The useful version teaches a child to test what the machine says. The weaker version teaches them to ask it nicely and admire the output. You can tell which one you are buying with a single question: what does my child do when the AI is wrong.