Learning from data

To build a cat detector with rules, you'd have to describe a cat in if statements. Pointy ears — but some cats have folded ears. Whiskers — hidden in many photos. Fur — what colour? You'd never finish.

So AI does something else entirely. Show it thousands of pictures labelled "cat" and thousands labelled "not cat", and let it find the pattern itself.

Nobody writes the rule. The system works out its own, from the examples.

This is exactly the exercise from Patterns & Logic — remember the two groups?

Group A: 4, 16, 36, 64, 100 Group B: 5, 17, 33, 61, 99

Nobody told you the rule. You examined the examples, found the pattern (perfect squares), then tested it on new numbers.

That's training a model, done by hand. You looked at labelled examples, extracted a rule, and used it to predict. A machine does the same thing with millions of examples instead of ten.