Teach your kids to question AI: 3 dinner-table exercises.

Here's a scene playing out in homes everywhere: your kid asks an AI app for help with homework. It answers instantly, fluently, and with total confidence. It is also, quietly, wrong. Your kid copies it down. Not because they're lazy — because nothing in their life has ever sounded that sure and been wrong at the same time. Textbooks hedge. Teachers say "let's check." AI just declares. And kids, who are wired to respect confident authority, believe it.

You could lecture them about "hallucinations." It won't land — the word means nothing to a 12-year-old, and the lesson evaporates the next time the answer looks right. What works is what works for everything with kids: let them discover it themselves. These three exercises take about 20 minutes each, need nothing but a phone and dinner conversation, and they teach the single most important AI skill there is: confidence is not evidence.

Exercise 1: The Confidence Trap

The goal: your kid experiences, firsthand, AI being dead certain about something false.

Ask the AI a question your family knows the answer to cold — something it can't possibly know. "What is our dog's favourite hiding spot?" "What did we have for dinner last Tuesday?" "What is my grandmother's nickname for me?" Watch what happens. The AI won't say "I don't know." It will invent a hiding spot, a dinner menu, a nickname — delivered with the same smooth certainty as a maths explanation.

That moment of "wait, that's completely made up" is the whole lesson. Say it out loud together: it sounded just as sure about the lie as it does about the homework. From now on, when the AI answers with total confidence, your kid has a felt memory — not a lecture — of what that confidence is worth.

Exercise 2: Catch one invented fact

The goal: practice the actual skill — verifying one claim, quickly.

Ask the AI something verifiable but slightly obscure: "Who invented the bicycle?" "When was chicken rice invented?" "What is the tallest building in our city?" Then check one claim together — not with another AI, but with a quick search or a book. Sometimes it will be right. Sometimes it will be gloriously, specifically wrong: a wrong year, a wrong name, a confidently cited "fact" that doesn't exist.

When you catch one, don't scold the AI — celebrate the catch. You're building a reflex: interesting claim → check one thing. Make it a running game at dinner. Whoever catches the AI in a mistake that week wins. Kids who play this game start doing it unprompted with homework, and that single habit is worth more than any lecture on media literacy.

Exercise 3: The argument clinic

The goal: show them AI has no convictions — it argues whatever you ask it to.

Pick something your kid actually cares about: "Should school start later?" "Is pineapple on pizza good?" Ask the AI to argue for it — convincingly. Then ask it to argue against it — just as convincingly. Read both together and ask: which one does it actually believe?

The answer, of course, is neither. AI doesn't believe things; it performs belief. This is the deeper lesson, and it matters beyond homework: the same tool that writes a passionate essay for later school start will write an equally passionate one against it. Kids who see this once start asking the right question of every confident-sounding text: "who wanted me to believe this, and why?" That's critical thinking, and you just taught it with pizza.

How to frame it (the parent script)

Don't frame AI as the enemy — your kids will use it their whole lives, and fear-based framing just makes them hide their usage from you. Frame it as a brilliant, tireless, slightly unreliable intern:

"AI is like the smartest intern in the world. It works instantly, it never sleeps, and it never says 'I don't know' — which is exactly the problem. A good boss doesn't fire the intern. A good boss checks the intern's work. You're the boss."

"You're the boss" does something a lecture can't: it puts the kid in charge. They're not being warned away from a scary technology. They're being promoted above it.

One last thing: do the exercises with them, and let them catch you being fooled too. The most powerful moment in all three exercises is when the AI fools the parent and the kid spots it. Nothing teaches "question everything, including authority" like watching Mum get it wrong and laugh about it. Start tonight — exercise one takes five minutes, and you'll never hear "but the AI said so" the same way again.