"AI" Means Everything, Which Means It Means Nothing!

There's a thermostat in my hallway that the manufacturer proudly calls "AI-powered." It turns the heating off when nobody's home. My nan did the same thing in 1974 with a sharp word and a cardigan, and nobody gave her a press release (I think?).

This is the problem with the term "AI" It has been stretched so far, over so many wildly different things, that it has very nearly stopped meaning anything at all. It's become the marketing equivalent of the word "natural" on a packet of crisps — technically defensible, practically useless.

So let's pull the term apart and look at the bits, because once you can see the scope of what "AI" actually covers, you stop being dazzled by it and start asking the only question that matters: what does this thing actually do?

The umbrella is enormous

When someone says "AI" they could be talking about any of the following, all of which are genuinely, legitimately called AI by serious people:

• The villain in a video game that decides whether to flank you or run away
• The spam filter quietly binning the seventeenth "URGENT: claim your refund" email of the day
• Netflix deciding you'll probably like another grim Scandinavian murder series
• A self-driving car interpreting a roundabout in the rain
• ChatGPT writing a sonnet about your cat
• A medical model spotting a tumour a radiologist might miss
  These are not variations on a theme. They are radically different technologies built on different principles. The video game enemy is mostly a set of hand-written rules — if player is close, do this; if health is low, do that. The tumour-spotter is a neural network trained on hundreds of thousands of labelled scans. Calling both "AI" is a bit like using the word "vehicle" for both a skateboard and a cargo ship. Accurate! Spectacularly unhelpful.

A quick tour of the family tree

Most of what gets called AI fits into a rough Russian doll structure, and it genuinely helps to know which doll you're holding.

Artificial Intelligence is the big outer doll — the broad idea of getting machines to do things that, when a human does them, we'd call "intelligent." That's it. That's the whole definition, and you'll notice it's so loose you could drive a self-driving lorry through it.

Machine Learning is the doll inside. Instead of a programmer writing every rule by hand, you show the system thousands of examples and let it work out the patterns itself. Your email spam filter learned what spam looks like; nobody sat down and wrote "if it mentions a Nigerian prince, bin it."

Deep Learning is the doll inside that — machine learning using big multi-layered neural networks. This is the engine behind image recognition, voice transcription, and the thing that finally made your phone stop hearing "wreck a nice beach" when you said "recognise speech."

Generative AI (the large language models, the image generators) is the bit everyone's lost their minds over since 2022. It's a particular flavour of deep learning that produces new content rather than just classifying existing stuff.

So when a company says "we use AI," they've told you almost nothing. They might mean a cutting-edge language model, or they might mean a spreadsheet with ambitions.

The AI Effect: the goalposts have wheels

Here's a genuinely funny quirk of the field. There's a long-observed pattern, sometimes called the "AI effect," where the moment an AI technique actually works and becomes useful, we stop calling it AI.

Optical character recognition — reading text from an image — was a hard AI problem for decades. Now it's just "the scanner app." Chess-playing computers were the absolute frontier of artificial intelligence right up until one beat the world champion in 1997, after which everyone shrugged and decided it was "just brute-force search, not real intelligence." Sat-nav route planning, autocomplete, fraud detection — all were AI until they worked, at which point they graduated to being "just software."

As one researcher dryly put it, AI is whatever hasn't been done yet. Which means the term is permanently pointed at the horizon, never at the ground under your feet. Handy for hype. Terrible for clarity.

Narrow vs. general: the gap nobody mentions in the ads

Every single AI system in commercial use today — without exception — is what's called narrow AI. It does one category of thing. A model that's a world-beating chess player cannot make you a cup of tea, recommend a film, or understand why the chess is even fun. It has no idea it's playing chess. It's an extraordinarily sophisticated pattern-matcher pointed at one task.

Artificial General Intelligence — a system that can flexibly handle any intellectual task a human can, transferring knowledge from one domain to another — does not exist. It might one day. People argue ferociously about whether that day is five years or fifty years or never. But the chatbot that wrote your marketing email is not a baby version of it, any more than a very good calculator is a baby mathematician.

The trouble is the word "AI" smuggles the sci-fi version — the conscious, scheming, red-eyed robot — into a conversation that's actually about a tool for sorting customer support tickets. The film 2001 did the branding so well that we're all still paying for it.

"AI-washing" or how to slap a label on anything

Because "AI" sounds expensive and futuristic, a small industry has emerged around calling things AI that simply... aren't. Regulators have started using the term AI-washing for it, in the same spirit as greenwashing.

A few tells that something's been dressed up:

• It's described as "AI-powered" but the company can't tell you what it learns from or how
• It does something with a fixed, predictable output every time — that's a rule, not a model
• The "AI" feature is a chatbot bolted onto a FAQ page that was already there
• The pitch leans entirely on the word "AI" and goes quiet the moment you ask "how?"
  None of this means the underlying product is bad. My hallway thermostat works fine. It's just a thermostat, and the honest version of that sentence sells fewer thermostats.

So what do you actually do with all this?

You don't need to become a machine learning engineer. You need exactly one habit: when something is called "AI," ask what it does and how.

• What's the actual task? (Sorting emails? Generating text? Spotting fraud?)
• Does it learn from data, or follow fixed rules? Both are fine — but they're different products with different strengths and failure modes.
• What happens when it's wrong? A film recommendation being wrong costs you a dull evening. A medical model or a self-driving car being wrong is a different conversation entirely.
  Get those three answers and the marketing fog clears instantly. You stop buying "AI" and start buying a thing that does a job — which is the only thing you ever actually wanted.

The term "AI" is a doorway with a very impressive sign over it. Useful for getting people to walk through. Tells you nothing about which room you've ended up in. So always, always look at the room.

This post was co-authored by Jason Etheridge & Opus 4.8, the hero image was created from a single prompt to GPT-5.5