Kids will grow up alongside AI the way earlier generations grew up alongside the internet. The question isn’t whether they learn it, but how. Most material is either too abstract, too text-heavy, or not safe for young learners.
We set out to build a learning platform that teaches children the real fundamentals of AI, not “here’s a chatbot,” but the actual ideas: that AI learns from examples, that more good examples make it smarter, and that it can be wrong. And we had to do it in a way that’s playful, age-appropriate, and something parents and teachers can trust.
The core idea, in one game: AI learns from examples
The platform is built around a hands-on moment: a game where kids give the AI examples (pictures of fruits and vehicles) and watch it get better at telling them apart. The more examples it sees, the better its guesses get. Then you flip it: biased or unfair examples make an unfair AI.
That one interaction teaches what most adults don’t understand about machine learning.
Three ideas, taught by doing
- AI learns from examples: not by magic. Kids are the teacher.
- Better examples, better AI: and biased examples make a biased AI. A powerful, age-appropriate lesson.
- AI can be wrong: so we should check it. Healthy skepticism, learned young.
Safety isn’t a feature, it’s the foundation
A platform for children lives or dies on trust. Ours is built so that the AI a child interacts with runs inside a guardrailed sandbox: age-appropriate content only, no open-ended chat with the open internet, no data collection beyond what a teacher needs to see progress. The fun is the surface; the safety is the structure underneath.
Why it matters: the generation in school today will manage, question, and build AI systems their whole lives. Teaching them early, that AI learns from examples, and that those examples can be biased or wrong, is one of the most useful things we can do.
What mattered most
- Teach the real ideas, learning from examples, bias, fallibility, through play, not lectures.
- Let kids be the teacher: labeling examples makes “training data” tangible.
- For children, a guardrailed sandbox and minimal data collection are non-negotiable.