AI, for real

If you've felt lost in all the talk about AI, you're not alone. A Microsoft report on AI in education found that 52% of US students have received no training on the subject, despite growing up surrounded by the technology. The noise around AI has drowned out the reality.

Here's what AI actually is.

AI has read everything, lived nothing

Two ways of learning what the ocean is. A kid who's swum in the waves knows saltwater stings your eyes, that currents can pull you out, that wet sand sticks to your feet. They know because they've lived it.

An AI has read millions of texts about the ocean. It can tell you seawater holds roughly 35 grams of salt per liter, quote poetry about the sea, explain how tides work. Its answers are accurate. But it has never set foot in the water. It learned all of it by reading, without experiencing any of it.

That's the difference. An AI like ChatGPT has ingested billions of texts and learned which words tend to follow which. When it answers you, it assembles the most likely sequence of words for your question. Statistics at a massive scale, nothing more.

What AI actually does well

Specific, well-defined tasks. Spotting patterns: fraudulent transactions in banking data, conditions on an X-ray, what you're asking in plain language. Making predictions: what you might buy next, the weather, a delivery window. Automating the repetitive: answering common questions, sorting emails, transcribing audio to text.

Same principle every time: find patterns in data, apply rules learned from it. AI is a pattern-matching tool, a very sophisticated one, but that's the whole story.

What AI isn't

Not magic. Mathematics, a lot of it, running fast on powerful computers. Every decision an AI makes comes down to a calculation.

Not sentient either, whatever the movies suggest. It doesn't think, doesn't feel, doesn't understand the way a person does. When ChatGPT writes you a poem, it isn't feeling creative. It's calculating which words are likely to follow each other, based on millions of examples.

It gets things wrong too, sometimes with total confidence. It can hallucinate facts, miss context a human would catch instantly, fail at anything it wasn't specifically trained for.

And the word "AI" itself is misleading. It's an umbrella term for very different technologies, from simple rule-based systems to complex neural networks.

Where things actually stand

What works: AI saves real time on repetitive tasks. It spots patterns humans miss in large datasets. It makes some services more accessible, real-time translation for one. It keeps getting cheaper for small teams and individuals.

The limits: important decisions still need human oversight. AI struggles with anything genuinely new or that requires real understanding. It needs real resources, compute and training data. It can carry forward the biases baked into its training data.

The honest truth? Most useful AI applications today are narrow and specific. We're nowhere near a "general" AI that can do anything a human can do. What exists are powerful tools for precise tasks.

Why this understanding matters

Once you know what AI can and can't do, you make better calls on when to use it. You stop asking it for things it isn't built for. You spot the people selling magic solutions. And used for what it's actually good at, it becomes a genuinely useful tool instead of a letdown.

A practical way to think about it

Think of AI like a calculator. A calculator is excellent at arithmetic, faster and more precise than any human. But you wouldn't ask it to write a poem or understand your feelings.

Same with AI. Exceptional at certain tasks, useless at others. The whole game is knowing which is which.

Sarah Connor?

So no, AI isn't taking over the world. We're a long way from that.

Terminator - I'll be back

Understanding how this tool works changes how you use it. To see what AI can actually do in practice, my article on AI agents is a good place to start.