Grandkids teaching grandparents AI: the tutoring that runs the other way

The scene repeats every Sunday: grandma asks what this artificial intelligence thing actually is, somebody answers in two sentences, and the conversation dies. The grandchild can use it but cannot explain it; grandma wants to understand but does not want to be a nuisance. Both stay exactly where they were.

Flip it round and make the child the teacher. It works in both directions: whoever explains discovers what they had not understood, and whoever listens loses the fear, because the person beside them has endless patience. From nine or ten a child can run the lesson alone; younger than that, they do the typing while the grandparents ask. There is a head start too: grandparents watched the telephone, the television and the cash machine arrive, so this is not their first new machine.

Exercise one: let the machine send you into the kitchen

My grandmother and I are cooking together. In her cupboard she has: [INGREDIENTS]. She likes [FAVOURITE DISH] and cannot eat [WHAT DOES NOT WORK]. Suggest three dishes we can make with that, each under 45 minutes, with a short ingredient list and the steps written in full sentences. Use plain language and no abbreviations. At the end, ask us which one we pick, and give us a shopping list for the missing bits of that one.

• Exercise two, the birthday poem: grandad names three true things about the birthday person, the AI offers three versions, and he rewrites the lines that do not sound like him. What gets read out at the table is then his, not the machine's.
• Exercise three, hunt the mistake: ask the AI about something the grandparents know better than anyone, their village, their trade, the year that factory closed. An invented detail almost always turns up. Finding it together is the whole exercise: the machine always answers, even when it knows nothing.
• Nothing can break. If something looks odd, close the chat and start again. Saying that out loud at the beginning relaxes half the afternoon.

What does not go into the chat

Before you start, one short rule for all three of you: no full name with address, no medical details, nothing to do with the bank. Not because the AI is evil, but because whatever gets typed leaves the house and nobody can say exactly where it ends up. And the third exercise delivers the most important lesson of the day: the machine does not tell you when it is wrong, so anything that matters gets checked.