What is AI?

Think about this morning. Your email had already moved a few messages into spam. Your phone suggested the next word while you typed. A map picked a route around a slow road. Nobody sat down and made those choices for you. A kind of software did, and it's the kind people mean when they say artificial intelligence (AI).

Put plainly, AI is software that learns patterns from examples, then uses those patterns to make a prediction. Which word comes next. Whether an email is spam. Which road is faster. Every AI you use is doing some version of that: looking at something new and making its best guess, based on everything it saw before.¹²

For most of computing history, a person wrote the rules. If the subject line says "free money," send it to spam. That works until the spammers change the wording. So engineers tried something different. They gave the computer thousands of emails that people had already marked as spam or not spam, and let it find the patterns itself. Nobody wrote the rule for spam. The machine found one in the examples, and when the spammers changed, it could learn again from new ones.³

You'll hear three words again and again.

• Data: the examples it learns from, such as emails, photos, or pages of text.
• Training: the learning itself, done ahead of time, over a lot of data.
• Inference: using what it learned to make a guess about something new, which is what happens each time you use it.

That's the whole cycle. Examples in, patterns learned, guesses out.³⁴

The name is older than most people think. A computer scientist named John McCarthy coined it in 1955, and he described it as the science and engineering of making intelligent machines. For decades it lived mostly in research labs. What changed is that computers got faster, and the world started producing enormous numbers of examples to learn from: photos, messages, searches, and whole libraries of text. With enough examples, the guesses got good enough to put in everyday products.²

Knowing that it's a guessing machine explains a lot. It's very good at patterns it has seen many times, and it's fast. But a guess is still a guess. AI doesn't know when it's wrong, and it can sound just as sure about a mistake as about the truth. There's nothing spooky about that. It's how a guessing machine works, and it's why the most useful habit with any AI tool is to check the things that matter.⁵

Each AI is built for a kind of job. The system that filters your spam can't drive a car, and the feature that recognizes your face on your phone can't write a poem. Even the most impressive chatbots are one kind of tool doing one kind of thing very well: predicting language. So when you hear about an AI that can do everything, it's fair to ask what it was trained on, and what it was trained to do.¹

So the next time a product says it uses AI, you can ask two plain questions. What examples did it learn from? And what is it guessing? Between them, they explain most of what it gets right, and most of what it gets wrong.