AI (Artificial Intelligence)

AI (Artificial Intelligence) is the umbrella term for systems that solve tasks usually associated with perceiving, language, planning, or decision-making. Not every AI system learns from examples — a trained model is just one (currently especially successful) subset of that broader category.

An example: The spam filter in your mailbox decides which emails are junk, even though you never explained to it what junk is. Today, decisions like that usually live in a model whose numbers come from lots of labeled example emails. A filter with a few hand-written if-then rules, by contrast, is an ordinary program; rule-based AI means large rule collections that, like expert systems, capture the knowledge of specialists.

Not to be confused with a giant rulebook: Many people picture AI as a collection of rules that experts wrote down line by line. That picture did exist: early expert systems consisted mainly of if-then rules. The trained models people usually mean today contain no readable rules, though, only tuned numbers. How those numbers come about is described in the entry on machine learning.

Where you’ll come across it: In news and debates about technology, in chat assistants like ChatGPT, on product pages for phones and software, and in features that translate text or sort your photos. What’s meant is almost always a trained model, even though the term itself is much broader. “AI” is also used quite generously as a marketing word, so it’s worth looking at what actually sits behind a feature.

Introduced briefly in Program, Algorithm, Model Compared.