AI projects for kids: train your own mini AI

Once your child understands that AI learns from examples, the next step is to do it. Here are beginner AI projects at three levels. All can be done with a grown-up and free tools.

Level 1: no computer needed

The human classifier. Make cards with pictures of two things, such as fruit and vegetables. Show your child ten, tell them the answers, then show five new ones and let them guess. Talk about which were hard.

Training day. Do the same with deliberately poor examples (only red apples) and see what happens with a green one.

Level 2: train a model in the browser

Several free educational tools let a child train a small image or sound classifier using a webcam or pictures, with no code. The usual steps are:

1. Choose two or three categories, for example a thumbs-up and a thumbs-down.
2. Collect a handful of examples of each.
3. Train the model with one click.
4. Test it with new examples, and see when it gets confused.
5. Add more examples and see it improve.

Check any tool's privacy policy and use a parent account, because webcams and microphones are involved.

Level 3: a little code

Block-based coding tools such as Scratch have extensions for machine learning, so kids can use their trained model in a game. Older kids can try short Python projects with a grown-up.

What to talk about

• What examples did you give it?
• When did it get things wrong, and why?
• What would make it better?
• Is it fair to everyone it might meet?

Safety

Keep projects on a shared device, use no real names or faces unless you have permission, and delete the data when you finish.

Next steps

For more ideas see 12 AI activities for kids, and for building games game design for kids.