There's a wasp that does something almost no one notices, but once you see it, you can't unsee it. The biologists who study Sphex ichneumoneus describe the routine in the entry on artificial intelligence in Encyclopædia Britannica. The female wasp comes home with food. She sets it down right at the edge of her burrow. Then she goes inside to check that no intruders are lurking. If the coast is clear, she comes back out and drags the food in. Tidy. Careful. Looks smart.
So here's the experiment. While she's inside checking, you nudge the food a few inches away from the entrance. When she comes out, she doesn't just grab it and go in. She drags it back to the threshold, sets it down, and goes inside to check for intruders all over again. Move it again, and she does the whole thing again. And again. She'll repeat that loop as many times as you move the food. She is not adapting. She is running a script.
That wasp is the cleanest illustration of the question this whole course is built around — what we actually mean when we call a machine "intelligent." Because the thing that's conspicuously missing in the wasp is the thing we keep assuming is present in AI: the ability to genuinely adapt, to understand, to know what it's doing. And the central claim here, the one idea everything else hangs on, is that today's AI is much closer to the wasp than to the mind you imagine behind it. It's a pattern-matching machine running an extraordinarily sophisticated script. It predicts the most likely next thing based on the data it learned from. That's it. That's the engine.
Now, that probably rubs against everything you've absorbed about AI, so let's slow down and build it properly. Start with the word "intelligence" itself, because it's doing a lot of quiet work. Psychologists, as Britannica lays out, don't define human intelligence by one trait. It's a bundle — learning, reasoning, problem-solving, perception, and using language. When researchers built AI, they went after those pieces one at a time. And here's the part that trips most people up: a machine can imitate any single one of those pieces convincingly without having any of the others, and without understanding a thing.
Take learning, the simplest piece. Britannica describes the most basic kind as rote learning — a chess program that tries random moves until it stumbles into checkmate, then just memorizes "this position goes with this move." Useful, but brittle. It only knows exactly what it's seen. The harder, more human trick is generalization — taking past experience and stretching it to a new situation you've never faced. Their example is beautifully small. A program that learns English past tenses by rote can't handle the word "jump" unless someone already showed it "jumped." But a program that generalizes figures out the rule — add "-ed" — and handles "jump" on its own, just from seeing similar verbs. That leap, from memorizing examples to spotting the pattern underneath them, is the whole game. Hold onto it, because it's the engine of every system we'll meet.
So if intelligence is a bundle of separate abilities, what is AI, plainly? Google Cloud puts it about as simply as anyone: AI is a set of technologies that lets computers learn, reason, and do tasks that used to need a human — understanding language, analyzing data, making suggestions. IBM frames it as machines simulating human learning, problem-solving, and decision-making. Notice the verb both of them lean on. Simulating. Imitating. Not possessing. Both Google and IBM are clear that underneath the simulation, these are, in Google's exact phrase, "complex pattern-matching machines." They find patterns in mountains of data, patterns a person would miss, and they use those patterns to predict.
Here's the kitchen-table version, and it's worth sitting with. Think of someone who's watched ten thousand hours of cooking shows but has never once tasted food, never smelled an onion hit hot oil, never felt hungry. Ask them what goes next in a recipe and they'll nail it — garlic after the onions, salt before the simmer — because they've seen the pattern that many times. They predict the next step beautifully. But they don't know what any of it tastes like. That gap — flawless prediction sitting right next to zero actual understanding — is the gap at the heart of every AI system you'll ever use.
That reframe clears up the misconceptions, and there are three big ones worth naming out loud. The first: AI is sentient — it has feelings, it's aware, there's somebody home in there. Google Cloud is blunt about this. AI systems can process and even simulate emotions, but they don't have consciousness, self-awareness, or genuine feelings. When a chatbot writes "I'm so sorry to hear that," it is predicting that those words are the likely continuation of a sad message. It feels nothing. There's no wasp in there feeling worried about intruders — there's just the script that produces worried-sounding words.
The second misconception is the dangerous one: AI is always right. This is where the prediction idea starts paying off, so stay with it for one step. If a system's whole job is to produce the most likely next thing, then "likely" and "true" are not the same target. It can generate something that sounds completely right and is completely wrong, with total confidence, because confidence isn't part of the calculation — likelihood is. And it's only ever as good as what it learned from. Both Google and IBM hammer the same point: AI is only as good as the data it's trained on, and if that data carries human bias, the AI learns the bias right along with everything else. A later episode opens on a lawyer who filed a legal brief citing six court cases — all six invented by AI, all six sounding perfectly real. That's not a glitch. That's a prediction machine doing exactly what it was built to do.
The third misconception is subtler, and it's the one even careful people get wrong. They talk about "AI" as if it's one thing — a single technology, maybe a single product. It isn't. Google Cloud describes AI as a broad field, not a single technology, and it pulls from computer science, statistics, linguistics, neuroscience, even philosophy. IBM describes it as nested layers built up over more than seventy years. AI is the big outer circle. Inside it sits machine learning — training a system on data instead of hand-writing every rule. Inside that sits deep learning, which uses layered neural networks. And the chatbots everyone's talking about now sit deeper still. So when a headline says "AI did X," that's about as specific as saying "a vehicle did X" — could be a bicycle, could be a cargo ship. Worth knowing whenever someone makes a sweeping claim about what "AI" can or can't do.
Here's a genuine disagreement worth flagging, because it's alive right now and serious people land on opposite sides. Britannica's classic definition insists that real intelligence "must include the ability to adapt to new circumstances" — the exact thing the wasp lacks. By that strict standard, a system that only predicts likely patterns isn't truly intelligent at all, no matter how fluent it sounds. But the vendor view, the one you'll hear from IBM, leans the other way — it lists "creativity and autonomy" right in its definition of what AI simulates, treating fluent imitation as a meaningful form of the real thing. The honest read, and the one this course takes, sides closer to Britannica. Imitating a piece of intelligence is not the same as having it. The wasp's routine looks intelligent until you move the food. A chatbot's essay looks intelligent until you check the citations. Keep that test in your pocket: poke it, move the food, see if it adapts or just reruns the script.
So if someone stopped you right here and asked what AI actually is, in one breath — what would you say? … It's a pattern-matching machine that predicts the most likely next thing, based on the data it learned from. Not a mind. A predictor.
And that one idea is a master key, which is the whole reason this course is built around it. Why can ChatGPT write a passable essay? Because it's predicting likely sequences of words. Why does it confidently make things up? Because likely and true aren't the same target. Why does it sometimes echo ugly stereotypes? Because it learned from human data, and our data is full of them. Why are governments in 2026 scrambling to regulate it? Because a prediction machine that sounds authoritative, deployed at massive scale, can do real damage when it's wrong. Every one of those — the powers and the failures — falls out of the same simple fact.
Strip it all back, and three things are doing the real work here. Intelligence is a bundle of separate abilities, and AI imitates the pieces without the understanding. Prediction is the engine — not thinking, not feeling, just the most likely next thing. And "AI" isn't one technology but a stack of nested ideas, which is exactly the stack we'll climb from here. The next stop is the part everyone gets wrong about how we got here — the decades when the dream of thinking machines mostly failed, and the strange reason it suddenly started working.