Read the story
Sprout tries to teach Moogi what a cookie is, and finds out an AI learns from examples, not explanations.
Sprout: Moogi! I baked cookies! Do you want one?
Moogi: A cookie? What's a cookie?
Sprout: Oh! Um... it's round. And brown. And it has little chocolate spots.
Moogi: Round... brown... spots. Hmm. Words are tricky for me.
Moogi: I'm an AI! I don't learn from rules. I learn from examples. Can you show me some?
Sprout: Let's do it! You can help too. Tell Moogi which ones are cookies.
Teach Moogi: play it in the world above!
Sprout: Wait... Moogi thought the button was a cookie!
Moogi: It was round, brown and spotty! Every cookie you showed me looked like that.
Moogi: And the heart cookie didn't look like any of my examples, so I said no.
Sprout: So you weren't really learning what a cookie is. You were learning what my examples had in common.
An AI learns by finding patterns in examples. It doesn't understand things the way you do. It spots what the examples have in common.
Moogi: Exactly! And if all my examples look the same, I think everything has to look like that.
Sprout: So we should show you lots of different cookies, and lots of things that aren't cookies?
Moogi: Yes! More examples, and more different examples, make me smarter.
An AI is only as good as its examples. Gaps in the examples turn into mistakes, and the AI can be very sure of itself and still be wrong.
Moogi: Real AIs learn from millions of examples. Pictures, words, sounds...
Sprout: Then we'd better bake MORE cookies. For science!
Moogi: For science! ...What does science taste like?