How Does AI Learn?

Imagine you're teaching a dog a new trick. Sit! Down! Paw! The dog tries over and over — and when it gets it right, it gets a treat.

At first, the dog is just guessing. But after many tries, it realizes: "Aha, when I sit down, I get a reward!" It has recognized a pattern.

AI Learns in a Similar Way

An AI that's supposed to recognize cats is shown millions of cat photos. For each picture, it's told: "This is a cat" or "This is not a cat."

After lots and lots of pictures, the AI recognizes the pattern: pointy ears, whiskers, fur, a certain eye shape — that's probably a cat!

The Big Difference

But there's an important difference between the dog and the AI:

• The dog learns through real experience — it feels the treat, sees your smile, hears your "Good boy!"
• The AI only learns from data — from numbers and patterns. It has no experience and no feelings.

The dog eventually understands what "sit" means. The AI only recognizes that certain patterns go together.

▶What is training data?

The images, texts, or sounds that an AI learns from are called training data. The more and the better the training data, the better the AI becomes.

Imagine you want to learn to cook. If you only know three recipes, you can't do much. But if you've read a thousand recipes, you can even invent new dishes!

Large AI models like ChatGPT were trained with huge amounts of text from the internet — almost as if they had read half the internet.

Test Your Knowledge