How should we define AI?
Imagine you're at a dinner table with four people, and someone asks: "So… what actually is AI?"
The first person says it's the robot from the movies, the one that becomes self-aware and takes over. The second says it's just clever software, the thing that finishes your sentences when you type. The third says it's the future of every job. The fourth shrugs and says, "Honestly? I think it's a buzzword companies put on everything now."
Here's the surprising part: all four are partly right, and that's exactly the problem. AI has become a word so stretched that it means something different to everyone who says it. Before we can do anything useful with it, we need to agree on what we're talking about. So let's build that understanding the way you'd build trust with a new colleague, slowly, with real examples, and no pretending.
Let's not start with a definition. Let's start with three things AI already does.
Definitions are slippery. Examples are sticky. So before we define AI, let me show you three places it's already woven into your day, even if you've never noticed.
1. The vehicle that drives itself
Picture a delivery van that needs to get a shipment across the city. To do that on its own, it has to juggle several skills at once: figure out the best route, see the road and spot obstacles, and make split-second decisions when a cyclist swerves or the weather turns. Each of those skills is a different flavour of AI, working together. The same mix powers delivery robots, drones, and even autonomous ships.
The promise is real, fewer accidents, smoother logistics, but so is the lesson: what looks like one clever machine is actually several specialised abilities stitched together.
2. The invisible hand that picks what you see
Open your phone. The songs suggested to you, the next video that autoplays, the ads in your feed, even the order of the news headlines, almost none of that is the same for the person sitting next to you. It's chosen, quietly, by AI that has learned your habits.
This one matters because it's so easy to miss. The front page of a printed newspaper is identical for every reader. The online version is tailored to you specifically. That personalisation is helpful, and it has a shadow side we'll come back to: filter bubbles, where you slowly stop seeing anything that disagrees with you.
3. The machine that recognises, and invents, images
Your photo app already groups pictures by face. Passport gates compare your face to your document. The same kind of technology helps a self-driving van tell a person from a lamppost. And it now runs in reverse too: AI can generate faces, scenes, and videos of things that never happened.
That last part flips an old assumption on its head. We grew up believing "seeing is believing." With AI-generated images and video, that's no longer a safe rule, a point worth keeping in your back pocket for the rest of this course.
So why is AI so hard to define?
Now that you've seen what AI does, here's why nobody can give you a tidy one-line definition. There are three honest reasons.
Reason one: even the experts haven't agreed. The field keeps moving its own goalposts. There's an old joke that AI is "whatever computers can't do yet." It sounds cynical, but there's truth in it: once a trick becomes routine, like finding the fastest route on a map, we stop calling it AI and start calling it ordinary software. The frontier keeps shifting.
Reason two: the movies got there first. Decades of films gave us walking, talking robots with feelings and grudges. Those stories are wonderful, but they're really about us, our hopes and fears, not about how the technology actually works. They set expectations that real AI never promised to meet.
Reason three, and this is the big one: easy and hard are backwards. Pause and pick up a cup near you. Effortless, right? Yet behind that simple act is a staggering amount of coordination, your eyes locating the cup, your brain planning the motion, your hand applying just enough grip. Teaching a robot to do that reliably is one of the hardest problems in the field. Meanwhile, chess, which feels like the summit of human intelligence, turned out to suit computers beautifully; a machine beat the world champion back in 1997. The things that feel hard to us are often easy for machines, and the things that feel effortless are often brutally hard. Hold on to that idea, it explains a lot of what comes later.
A definition you can actually use
If "what computers can't do yet" is too slippery, here's a more practical handle. Instead of one rigid sentence, look for two qualities. When a system has a meaningful amount of both, it's fair to call it AI:
- Autonomy, it can handle tasks in messy, real-world conditions without someone guiding its every step.
- Adaptivity, it can improve with experience, learning from data rather than waiting for a human to rewrite its instructions.
A thermostat is automatic, but it doesn't learn. A system that gets better at predicting demand every month as more orders flow through it, that's showing adaptivity. Keep these two words handy; they're the most reliable test you'll have.
A word of caution about words
Here's a trap that catches smart people every day. The words we borrow to describe AI, learning, understanding, intelligent, carry a lot of human baggage.
When we say a system is "intelligent" because it reads medical scans well, it's tempting to assume it could also hold a conversation, plan a holiday, or notice you're upset. It can't. When we say a vision system "understands" an image, we don't mean it understands the way you do.
A pioneer of the field, Marvin Minsky, called these "suitcase words", words you pack full of meanings, so the listener unpacks the one you never intended. Two things follow from this, and they're worth remembering:
First, intelligence isn't a single dial. You can't line up a spam filter, a chess engine, and a self-driving van and rank them from "least to most intelligent." The question doesn't even make sense, because today's AI is narrow, being brilliant at one task tells you nothing about another.
Second, a small habit that marks out people who really get it, "AI" isn't a thing you can count. AI is a field, like biology or mathematics. You wouldn't say "we need more biologies." So rather than "an AI" or "two AIs," say "an AI method." It's a tiny phrasing change, but it quietly signals that you understand what AI actually is.
If you remember nothing else from this first section, remember this: AI isn't a single magical mind. It's a collection of specialised methods that show autonomy and adaptivity, each narrow, each good at one thing. The movies oversold the robots; the real story is quieter, stranger, and far more useful, and it's already running in the van on the road, the feed on your phone, and the camera at the border.
In the next section we'll place AI on the map alongside its close relatives, machine learning, deep learning, data science, so the jargon stops being a fog and starts being a toolkit.
Did it stick?
Three quick questions, tap an answer to check yourself.
