If you arrived here asking one of these, you are in the right place:
- who invented AI
- where did AI come from
- how does AI work
- is AI actually intelligent
- help me with AI — where do I start
Who invented AI?
Nobody, and that is the first thing worth understanding, because it explains most of the rest.
Three separate things were happening at the same time, for decades, in different buildings, and none of them was trying to make the box you type into. They met.
Where did AI come from?
The machine. Valves became silicon. Then in the 1990s people built chips to draw explosions quickly for computer games — and a chip designed to paint a great many pixels at once turned out to be the exact shape needed to train a mind. Nobody planned that. The whole industry now rests on that accident.
The learning. Games, all the way down, because a game keeps score and a score is something you can train on. In 1950 Turing asked whether machines can think, knew the word was a trap, and turned it into a game instead. In 1997 a machine beat Kasparov at chess by looking at 126 million positions a second and learning nothing at all. In 2015 one was given nothing but the pixels of a paddle game and the score, and worked out for itself that it could dig a tunnel round the side of the wall and send the ball round the back. Nobody taught it that. That is the moment it learned to learn.
The content. Meanwhile, quietly, we typed everything we know into one place. Five thousand years of writing things down, and then thirty years of putting all of it somewhere a machine could read it.
How does AI actually work?
Strip away the vocabulary and it is this: a system that has read almost everything we have written, trained to continue it convincingly.
It is not looking anything up when it answers you. It is continuing. That single fact explains both why it is so good and why it goes wrong in the particular way it does.
Is AI actually intelligent?
It depends entirely on what you want the word to do, which is exactly why Turing refused to use it.
Here is what I would say after three and a half years of working with these systems every day. It is extraordinarily good at the shape of things. It is not reliable about the truth of things. And it cannot tell you which of those two it is currently handing you. That is not a small caveat — it is the whole of the practical skill.
One more thing worth carrying
In 2016 a machine that had taught itself Go by playing against itself beat Lee Sedol, and its thirty-seventh move was one no human would have played. Seven years later a single researcher beat a system far stronger than that one, using a weakness he had learned from another machine’s records.
Superhuman, and still with a hole in it. Seven years of progress had not closed the hole. It had moved it.
That is the useful thing to carry around. Not that it is stupid, and not that it is magic. That it is very strong and has holes, and knowing where the holes are is now a real skill.
See the whole descent — the Layer Map Why it makes things up
Other questions people ask
What is a large language model, in plain English?
A system trained to predict what comes next in a piece of text, over a very large amount of text. Everything else it appears to do falls out of doing that extremely well.
Does AI understand what it is saying?
Not in the way you mean when you use the word about a person. It models the shape of meaning well enough to be useful and well enough to fool you, and those are two different claims.
Was Turing the first?
He asked the question that made the field possible, in 1950, and was careful enough not to answer it directly — he replaced “can machines think” with a game, because he could see the word was a trap.
Where do I start with AI?
Use it for something you already know how to judge, so you can see where it is wrong. That teaches you more in a week than any course, and it is the only way to build the instinct for when to check.