It's the I in AI That Does It

There was a song on children's television when I was young where the whole point was that a single letter decided how we heard a word. It's the vowels in the words that do it. Swap one letter and the meaning shifts under your feet.

I've come to think it works roughly the same way with AI. It's the I that does it.

It's the word intelligence that makes us attribute to today's technology properties it doesn't have. When someone says artificial intelligence, most people hear understanding, reasoning, insight, maybe intent, maybe something resembling consciousness. The word drags all of that along for free, before the technology has demonstrated a single thing.

A word that promises more than it keeps

Today's generative models are, at bottom, statistics. They predict the next word, the next pixel, the next step from enormous amounts of training data. The result can be impressive without the system understanding anything in the sense you and I put into the word understand.

That isn't a put-down. It's a description. A model that guesses the right word often enough becomes an enormously useful tool. But it doesn't know what it's saying, and above all it doesn't know what you forgot to ask. The model solves what you ask it to. It has no notion of what you left out. Call it intelligent, and you stop thinking about exactly that boundary.

The superlatives do the rest

It isn't only one word. The marketing supplies the rest.

Every new model is revolutionary, groundbreaking, human, agentic, reasoning, world-leading. The superlatives move the gaze from what the technology actually does to what we hope it will be. After enough of them you're no longer judging a product, you're relating to a promise.

Douglas Adams wrote about this before the technology existed. In his world the Sirius Cybernetics Corporation sells its robots and machines with Genuine People Personalities, and the story is so convincing that it becomes part of the product. You're no longer buying a door, you're buying a door that is genuinely happy to open for you. The language is the merchandise.

We're heading the same way. The words shape the expectations before the technology has demonstrated its actual properties, and then we measure the technology against the expectations instead of the other way around.

We read it in from the output

It shows most clearly in how we judge. Good answer, then it's intelligent. Bad answer, then it's stupid.

In both cases we do the same thing. We treat a prediction system as a thinking actor. We anthropomorphise it, upward when it impresses and downward when it misses, and we place a judgement on the machine that belongs with us. It was neither wise nor stupid. It guessed, and we read the guess as a character.

Swap the word, and the expectation collapses

Run the thought experiment. Suppose the technology had been called something else from the start. Artificial prediction. Advanced statistical model. Probabilistic language engine.

Considerably fewer people would have expected understanding or consciousness. Nobody asks whether their weather forecast is conscious, even though it does essentially the same thing: processes data and predicts. It was never the technology that carried the human expectations. It was the word.

The nuance that mustn't become an excuse

There's no settled definition of intelligence, and that's why the debate gets muddy. Define it functionally, as the ability to solve problems and reach goals, and you can say today's AI displays certain forms of intelligence. Demand understanding, consciousness, a subjective experience, and it isn't intelligent in the human sense. Same word, different things, and two competent people can argue for an hour without realising they were never talking about the same thing.

But that nuance can't become an escape. Because for how you're supposed to work, it doesn't matter which definition wins. Whichever you pick, you land in the same place: the system produces an output, and someone has to decide whether the output holds. That someone is you.

Why this is a question of accountability

This is where I don't stop at a note about language, because the word does something to how we take responsibility.

Believe something is intelligent, and you trust it. You lean back. You let the answer pass because it sounded wise. Know that it's inference, prediction, statistical generalisation, and you validate. You read every line, because you know nobody understood it before you did.

So the word intelligence isn't only a marketing problem. It's an invitation to hand over exactly the part you cannot delegate. The illusion that the machine understands is precisely what makes a person stop carrying their own judgement. And the judgement was the whole job.

It's the I in AI that does it. And maybe the superlatives too. They make us see intelligence where it's usually inference, prediction and statistical generalisation. Don't let them make you see a consciousness where there's only a tool you're still accountable for.

See also: Same Word, Different Things (series 30) and I'm Smarter Than the AI (series 1).