Strip away the branding and AI is this: software that learns its behavior from examples instead of being told what to do, line by line.
Traditional software is a recipe. A programmer writes explicit rules — if the password matches, log them in — and the computer follows them exactly. This works beautifully for tasks you can fully describe. It fails for tasks you can't. Nobody can write down the rules for recognizing a cat. You know one when you see one, but the knowing isn't made of rules you could list.
Machine learning — the engine behind nearly everything called AI today — flips the approach. Instead of writing rules, you collect examples: thousands of photos labeled cat or not cat. The system finds the statistical patterns that separate one pile from the other, on its own. The rules end up written by the data.
Why the word "AI" confuses everyone
"AI" gets applied to at least three different things:
- The aspiration — machines with general, human-like intelligence. This does not exist. It's a research goal, not a product.
- The field — decades of academic work, from 1950s logic machines to today's neural networks.
- The products — chatbots, image generators, recommendation feeds. These are narrow systems, each good at one kind of task.
Marketing collapses all three into one word, which is how a spreadsheet autocomplete and a sci-fi robot end up sharing a name. When you read a headline, ask which of the three is actually being discussed. Most of the time it's the third wearing the costume of the first.
What to read next
Everything a model learns, it learns from data — so start where the models start: data and datasets, the raw material everything else is made of.