Play Twenty Questions Against an AI, Then Beat It

Twenty Questions is a lesson in narrowing a search, dressed up as a car game. Playing it against a machine and then against a person shows a child exactly what each is good at.

How it goes

1. Your child thinks of something and writes it on a folded slip so nobody can accuse anyone of cheating.
2. The AI asks the questions. Play twenty questions. I am thinking of an object. Ask me one yes-or-no question at a time and keep count.
3. Watch the shape of the questions. The good ones split the possibilities in half; the bad ones guess at random. Your child will notice this within six questions.
4. Swap roles. Your child asks; the machine holds something in mind. Ask them to plan a first question that rules out half of everything.
5. Now play a person, with no computer involved, and use the same halving strategy.
6. Compare. Which was harder to beat, and why?

Who does what

AI plays both roles patiently and never gets bored, which no sibling will do for six straight rounds.

Your child invents the strategy, notices what makes a good question, and applies it against a human.

You be the human in round five, and lose to the halving strategy.

Make it easier or harder

For a five-year-old: restrict the category out loud first, to animals or to things in this kitchen. Twenty questions across everything in the universe is unbounded, and unbounded is where young children lose the thread.

For a twelve-year-old: cap it at ten questions rather than twenty. Halving the budget forces genuine planning, and it makes the difference between a good question and a lazy one impossible to ignore. Ask them to write their first three questions down before starting.

Take it off the screen

Round five onwards needs nothing. This becomes a car game, a queue game, a waiting-room game, and the halving idea sticks.

Play a version on paper with twenty animals written on a sheet, crossing off what each answer rules out. Seeing the list shrink makes the strategy visible for younger children.

When it goes sideways

The model will sometimes contradict an answer it gave four questions ago, or announce a solution that fits none of them. Do not let it slide past. Scroll back, find the contradiction, and read both lines aloud. This is the clearest demonstration of how these systems work that you will get for free, and it lands harder than any explanation.

If your child cannot see the halving idea, make it physical: write twenty animals on a sheet and cross off everything each answer eliminates. Watching the list shrink by half twice makes the strategy obvious.

One thing to watch

The AI holding something “in mind” is not really doing that between messages, and it may change its answer or contradict itself. When that happens, do not smooth it over; it is one of the clearest demonstrations available of how these systems work. Our guide on teaching kids that AI can be wrong covers what to say.

Related: can AI guess your drawing is the same idea with pictures instead of words.