AI, machine learning, deep learning: what's the difference?

You've probably seen a box or an ad that promises artificial intelligence (AI), machine learning, and deep learning, as if they were three separate features. They read like a list. They're closer to three circles, drawn one inside the other. Once you can see the circles, labels like that get much easier to read, and so do the headlines that use them.

The biggest circle is AI. It's the whole field: any software that does a task we'd normally call smart, like understanding speech, recognizing a face, or choosing a move in a game. That includes older systems built entirely from rules a person wrote. A chess program that follows a long list of if-this-then-that steps can count as AI, even though it never learned a thing.¹²

Inside that sits a smaller circle: machine learning. It's the part of AI that learns from examples instead of following rules a person wrote. Show it enough past examples and it finds the patterns itself, then uses them to make a guess about something new. The shows a streaming app suggests to you, and the text your bank sends about an odd purchase, are likely examples. Every machine learning system is AI. But plenty of AI is not machine learning, which is why the circles are different sizes.¹²

Inside machine learning is the smallest circle: deep learning. Deep learning is machine learning built with a structure called a neural network, made of layers of simple connected parts that pass information from one layer to the next. The word "deep" refers to how many layers there are. Small networks have one or two in the middle. Deep ones have many more, and some have hundreds. You don't need to picture the layers to follow the idea. More layers let a network handle harder and messier material, and that turned out to matter a great deal.¹³⁴

Deep learning is behind much of the AI people notice today, from voice assistants to photo search to chatbots. The reason is messiness. A spreadsheet of numbers is tidy. A photo, a recording of someone talking, or a page of writing is not. With enough layers and enough examples, a network can find useful patterns in that kind of material by itself, without a person first spelling out what to look for. It needs a lot of examples, though, and a lot of computing power.³

So the next time a product or a headline uses all three words, try placing the claim. Is it AI in the broad sense, which might only mean rules? Is it machine learning, learning from examples? Or is it deep learning, with many layers, trained on a mountain of data? Often it's one system that sits in the smallest circle, and so in all three at once. If a box lists them as three features, you can smile: it's likely one feature, counted three times.