The word is doing too much work
Artificial intelligence was coined in 1956, at a summer research workshop, by people who wanted funding to study thinking machines. It was a pitch, not a description.
Seventy years later the same two words are attached to a chess program, a photo filter, a bank's fraud alert, a warehouse robot and a chatbot. Inside, those systems have almost nothing in common. A word that covers all of that is not telling you much.
The thing that keeps happening
In 1997 a computer beat the reigning world chess champion. It was front-page news about machine intelligence. Today a free chess app on a cheap phone would beat that computer, and nobody calls it AI. It is a chess engine.
The same story ran for reading handwritten postal codes, for filtering spam, for finding faces in a camera viewfinder, for translating between languages. Each one was intelligence right up until it worked reliably. Then it quietly became software.
Researchers have a name for this: the AI effect. AI is roughly whatever machines have not quite managed yet. So the label does not describe a technology. It marks a frontier, and the frontier keeps moving away from whatever you already have.
What people mean when they say it now
Almost everything called AI today is machine learning: a system whose behaviour came from examples rather than from written instructions. That distinction is the subject of the next lesson, and it is the one that actually matters.
Most of the current noise is about one branch of that: very large models trained on enormous amounts of text and images. Chatbots are the visible face of it. Behind them sit the same techniques doing dull, useful work you never see.
A better question than is it AI
When you meet a system, ask three things instead.
What goes in, and what comes out? A photo goes in, a name and a number come out.
Where did its behaviour come from? Did a person write down the rules, or did the behaviour come from examples?
What happens when it is wrong, and who finds out?
Try it on a real case. A farmer in Kaduna points a phone at a maize leaf and an app says leaf blight, 0.82. In: a photograph. Out: a disease name and a confidence number. Behaviour from: some tens of thousands of labelled leaf photos. When wrong: money is spent on the wrong treatment and a crop is lost, and nobody may ever learn that the app was the reason.
Those four answers tell you everything you need in order to decide how much to trust it. Whether it counts as intelligence has become an uninteresting question, which is the point.
Two things AI is not
It is not one system. There is no such thing as the AI. There are millions of separate models, trained by different organisations, for different jobs, and they share nothing with each other. The model that reads your bank's transactions has no connection to the one that writes your emails.
It is not a being you are talking to. When a chatbot writes I think, that is a style of writing, learned from text written by humans and kept because people find it easier to read. Whether anything like experience is going on inside these systems is a real open question that serious people disagree about. But the word I in the output is not evidence either way, because that word would appear whether or not anything was going on.
Keep the question live if you find it interesting. Just do not let a pronoun answer it for you.
Before you move on