Teaching artificial intelligence to curious kids.

One person is the "programmer" and gives exact commands. The other is the "robot" and follows them to the letter — even when it goes wrong. That's how computers work.

1. Draw a grid on paper or use floor tiles.
2. Put a toy at the start and a "treasure" at the finish.
3. Give commands: "forward 2, turn left, forward 1".
4. The robot obeys exactly. Went wrong? Talk about what was missing.

Simulate how AI learns to classify: you show the rule, the child applies it. Then the rule changes — and the "AI" has to learn again.

1. Gather about 20 mixed objects: buttons, blocks, bottle caps.
2. You're the "trainer": sort by colour and explain the rule.
3. The child is the "AI": sorts new objects with the same rule.
4. Change the rule (by size) and talk: the AI needed new training!

Write each step on cards, then follow only what's written, to the letter. Where did it break down? Finding and fixing the mistake is called debugging.

1. The child dictates each step while you write it on a card.
2. Follow only what's on the cards, literally.
3. Find the gaps: "pick up the bread" — which hand? From where?
4. Rewrite until the algorithm works without errors.

Show cards from one category as "training examples". The child, playing the AI, sorts new cards. Then come the tricky cases.

1. Show 6 training cards from one category (e.g. animals).
2. The child sorts 10 new cards into "yes" or "no".
3. Add ambiguous items and talk about disagreements.
4. Ask: who decided what counts? Why does that matter?

In 30 minutes, the child trains a model to tell fish from rubbish in the sea, watches it get things right — and discovers its limits.

1. Go to code.org/oceans (no account needed).
2. Label the examples as "fish" or "not fish".
3. Test the trained model on new images.
4. Talk: where did it go wrong? Why? How could you fix it?

Spend a day noting every place AI shows up: at home, at school, at the shops, on devices. Realising it's everywhere changes how a child sees things.

1. Make a chart: "Where did we find AI today?"
2. Hunt all day: assistants, feeds, autocorrect, face unlock.
3. At dinner, count and sort your finds.
4. Talk: what surprised you? What helps? What worries you?

An experiment comparing how different prompts change the answer — recorded the way a scientist would.

1. Pick a task: "explain photosynthesis" or "write a poem".
2. Write 3 versions of the prompt: vague, specific and creative.
3. Note the 3 answers and compare their quality.
4. Together, write a "good prompt guide" for that kind of task.

A collection of real and AI-generated images and texts. The child is the detective: which clues give the AI away?

1. Gather 10 examples: 5 real and 5 AI-generated.
2. The child examines each one and gives a verdict with reasons.
3. Reveal the answers and talk about which clues helped or misled.
4. Build a "how to spot AI" checklist together.

Build a gesture or sound recogniser. Train it, test it and try to fool it on purpose — that's how you find its limits.

1. Open teachablemachine.withgoogle.com and choose "Image Project".
2. Record 30 or more examples of 3 gestures with the webcam.
3. Train the model (about 2 minutes) and test it live.
4. Try to fool it: lighting, background, another hand. Where does it fail?
5. Use "Export Model" to get a shareable link to show the family.

A structured debate on a real case: facial recognition at school, essays written by AI, who's liable when a self-driving car crashes.

1. Pick a case (AI4K12.org has suggestions).
2. Research both sides for 20 minutes.
3. Debate: 3 minutes per side, 2 for rebuttal — then swap sides.
4. Decide together: what rule would you recommend?