What AI Actually Is (Because Almost Nobody Arguing About It Knows)

Everyone has an opinion about AI now. Your uncle thinks it's going to take his job. Your coworker thinks it's going to cure cancer. Someone on your city council and even your friends think it's why the power bill went up. They're all arguing about the same two letters, AI, and most of them could not tell you what those letters actually mean.

I don't say that to be smug. I spent years earning a doctorate in this field — my research was on making AI systems explain themselves, specifically in cybersecurity — and I still catch myself using the word "AI" lazily. The term has been stretched to cover everything from a spam filter to a chatbot to the robot apocalypse. When one word means everything, it means nothing, and you can't have a real argument about nothing.

So this is the first post in a series. Before we fight about datacenters, jobs, regulation, or whether this technology ends us, let's agree on what we're fighting about.

It's pattern recognition. That's the whole trick.

Strip away the marketing and modern AI is this: a program that finds patterns in enormous amounts of data, then uses those patterns to make predictions about new data it hasn't seen.

That's it. That's the core of the thing.

Show a system ten million photos labeled "cat" and "not cat," and it learns which combinations of pixels tend to show up in cat photos. It doesn't know what a cat is. It has never petted one. It has a giant pile of numbers that got adjusted, millions of times, until "cat photo goes in, cat label comes out" happened reliably. We call that adjustment process "training," which sounds like teaching a dog, but it's closer to tuning millions of tiny dials until the machine's guesses stop being wrong.

The same trick works on network traffic (my world — is this login normal, or is someone breaking in?), on medical scans, on your Netflix queue, on what song plays next. Different data, same underlying move: find the pattern, predict the next thing.

"But ChatGPT talks to me"

Fair. Chatbots are the reason your uncle has opinions now, so let's deal with them directly.

Tools like ChatGPT are trained on a mind-boggling amount of text — most of the public internet, roughly — and they learn one skill extremely well: predicting the next word. Given "the capital of France is," the pattern says "Paris." Given your question, the pattern says whatever words tend to follow questions like yours. Do that word after word, very fast, with patterns learned from billions of sentences, and you get something that writes fluent paragraphs and feels like a mind on the other end.

It is genuinely very impressive. I do personally use these tools. But "predicts the next word really well" explains both the magic and the failures. The system has no internal fact-checker, no model of what's true — only what's likely to be said. Which is why it will confidently invent a court case, a citation, or a person, in the same smooth tone it uses for real ones. Researchers call this "hallucination," and it is not a rare glitch. Stanford's 2026 AI Index Report — probably the most rigorous annual snapshot of the field, and worth bookmarking — found hallucination rates on one knowledge benchmark ranging from 22% to 94% across 26 top models, depending on how the question was framed.

A machine that's brilliant at sounding right and structurally indifferent to being right.

Hold onto that, because half the fights about AI trace back to it.

What AI is not

Three things people get wrong, in both directions:

• It's not thinking. There's no understanding in there, no goals, no "wanting" anything. When a chatbot says "I'd be happy to help," nothing is happy. It's pattern-completing what a helpful reply looks like. Whether something mind-like could eventually emerge from scaling this up is a real debate — we'll get there later in the series — but what's running on your phone today is prediction, not thought.
• It's not a database. People treat chatbots like Google with better manners. Wrong model. A search engine retrieves documents that exist; a chatbot generates text that's statistically plausible. Sometimes plausible and true overlap. Sometimes they don't, and it will not warn you which one you got.
• It's not fake, either. This is for my fellow skeptics: dismissing all of it as hype is as lazy as the hype itself. Pattern recognition at this scale does things that genuinely weren't possible ten years ago — from a cybersecurity perspective, catching network intrusions humans miss, spotting tumors in scans, folding proteins. Per that same Stanford report, 88% of organizations now use AI somewhere in their operations. You don't have to like it. You do have to take it seriously tho.

Why the confusion exists

Two industries profit from you misunderstanding AI, and they push in opposite directions.

Tech companies benefit when AI sounds like magic — magic justifies the valuations, the datacenters, the "we can't slow down or China wins" framing. So everything is "intelligent" now. Your toothbrush has AI.

Hollywood and the attention economy benefit when AI sounds like the Terminator — fear gets clicks. So every story is framed as machines waking up.

The boring truth — very powerful pattern-matching with real uses, real limits, and almost no rules — doesn't sell products or ad space. That's the version I'm giving you anyway, because it's the only version you can make good decisions with. Even governments are still working out the vocabulary; NIST's AI Risk Management Framework, the closest thing the US has to official guidance, spends real effort just defining terms before it gets anywhere near managing risk. That tells you something about where we are.

Why any of this matters

You can't regulate what you can't define. You can't protest it coherently, defend it honestly, or vote on it wisely either. Right now, billion-dollar decisions — where datacenters get built, what jobs get automated, what rules exist or don't — are being made while most of the public is arguing with a cartoon version of the technology.

I work in cyber for a living, and I'll be straight with you about where I stand: I think this technology is genuinely useful, badly governed, and moving faster than our ability to manage it. All three of those things are true at once. This series is me trying to earn each of those claims instead of just asserting them.

Next in the series: that datacenter that just showed up outside your town. What's actually inside it, why it needs so much power and water, and why your objections to it are more legitimate than the industry wants to admit.

Sources worth your time: Stanford HAI's AI Index Report 2026 and NIST's AI Risk Management Framework.