The term "artificial intelligence" arrives carrying fifty years of cinema — thinking machines, robot minds, red camera eyes. Then you meet the real thing and it's… a spam filter. A photo app that finds your dog. A chatbot. The gap between the name and the reality confuses everyone, so let's close it properly.
The old way vs the new way
Traditional software works by explicit rules: a programmer writes if the email contains this phrase, mark it as spam. Every behaviour was typed by a human. This breaks down fast — nobody can write rules for "recognise a cat in any photo ever taken."
Machine learning — the engine behind essentially all modern AI — flips the approach: instead of writing rules, you show the system millions of examples and let it find the patterns itself. Feed it a million labelled photos and it works out, statistically, what combinations of shapes, textures and colours tend to mean "cat." Nobody wrote a cat rule. The rules emerged from the examples — and they're stored as millions of numbers no human directly authored.
What this means in practice
- AI is brilliant at patterns it has seen many times: recognising faces, transcribing speech, spotting fraud-shaped transactions, predicting the next word in a sentence.
- AI has no understanding underneath: the cat-recogniser doesn't know what a cat is — that it's alive, that it purrs. It knows what cats statistically look like. This distinction explains most AI failures you'll ever see.
- AI inherits its training data: patterns learned from biased or outdated examples reproduce those flaws at scale. The system isn't malicious; it's a mirror.
A useful deflation for the hype and the fear alike: today's AI is narrow. The system that plays chess can't drive a car; the one that writes essays can't fold laundry. The movie version — one general intelligence that does everything a human can — remains a research goal, not a product you can buy.
Why IT people specifically should care
Because AI is becoming infrastructure. It's already inside the tools you'll support — email filtering, endpoint security, helpdesk ticket routing, monitoring alerts — and increasingly it's a system you'll be asked to deploy, secure, and troubleshoot. You don't need to build models. You do need the mental model above, because "it's pattern-matching from examples, without understanding" correctly predicts both what these tools will nail and how they'll fail. The rest of this AI trail builds on exactly that foundation.