Where AI Learns

Meet the sorter

This is a tiny sorter. Its only job is to look at a new picture and say one word: cat or dog. It has never seen a single one yet. You are its teacher.

Cats

pointy ears, whiskers

Dogs

floppy ears, big nose

An AI learns from examples, not rules. Nobody types in "a cat has pointy ears." You show it labeled pictures, and it copies the patterns it sees. That means the examples you pick become the whole world it knows.

Teach it (the easy, lazy way)

In a hurry, you grab whatever pictures are closest. They happen to be mostly cats. Feed the sorter, then test it on new animals it has never seen.

🧺 Examples the sorter has seen 0 cats · 0 dogs

🐱 catsdogs 🐶

Tip: add a big pile of cats and only one or two dogs, just like a rushed teacher would.

Fix it: balance the examples

You cannot fix a sorter with a lecture. You fix it with better examples. Add dogs until the basket is even, then test the very same animals again.

🧺 Examples the sorter has seen 0 cats · 0 dogs

🐱 catsdogs 🐶

Keep adding dogs until both sides are close to even.

The big idea

You just saw the most important rule in all of AI, and you saw it for real, not as a slogan.

🧠 Carry these three

1. An AI only learns what its examples show it.
2. If the examples are lopsided or unfair, the AI will be too. That is called bias, and it is real.
3. Garbage in, garbage out. Fair examples in, fairer results out.

This matters in the real world. Real AI sorts job forms, photos, and voices. When its examples leave a group out, it gets that group wrong, the same way your sorter missed dogs. People whose job is to build AI well spend a lot of time checking that the examples are fair.