What is AI?

If you have ever asked your phone for directions, watched a streaming service recommend a show, or had your email hide a spam message before you saw it, you have already used AI. It just did not announce itself.

So let's start with the simplest honest definition.

AI (artificial intelligence) is software that has learned to do a task by studying examples, instead of being told step by step how to do it.

That is the whole idea. Everything else is detail.

The old way: writing rules

Traditional software is a list of instructions written by a person. A calculator app works because someone wrote the rules of arithmetic into it. A spreadsheet works because someone wrote the rule "when the user types a formula, compute it." The computer does exactly what it is told, and nothing else.

That approach is great for tasks where the rules are clear. It falls apart for tasks where nobody can write the rules down. Try writing step-by-step instructions for "recognise a cat in a photo." Where do you even begin? Pointy ears? Some cats have folded ears. Fur? Some cats are hairless. A person can spot a cat in a fraction of a second but cannot explain the rule they used.

The new way: learning from examples

AI flips the approach. Instead of writing the rules, you show the computer a very large number of examples and let it work out the patterns itself.

Show it a million photos labelled "cat" and a million labelled "not cat," and a training process slowly adjusts the internal settings of the software until it gets good at telling the two apart. Nobody wrote a cat rule. The software found one. This process is called machine learning, and it is the engine behind almost everything people call AI today.

The same trick works for many kinds of examples:

• Show it lots of text, and it learns to predict what words come next. That is how chatbots like ChatGPT, Claude and Gemini work.
• Show it lots of speech with transcripts, and it learns to turn your voice into text.
• Show it lots of translated documents, and it learns to translate.
• Show it lots of medical scans with diagnoses, and it learns to flag suspicious ones for a doctor.

What AI is not

Some myths are worth clearing up early, because they cause a lot of unnecessary worry.

AI is not a robot. Robots are machines with bodies. Some robots use AI, most AI has no body at all. The AI you will meet is an app or a website.

AI is not conscious. Today's AI does not have feelings, wants or awareness. A chatbot can write "I'm happy to help!" in the same way a greeting card can say "Happy Birthday": the words are there, the feeling is not. It produces text that sounds like a person because it learned from text written by people.

AI is not always right. Because it learned from patterns rather than from a rulebook, it can be confidently wrong. Chatbots sometimes invent facts, dates, quotes and citations that look completely convincing. People in the field call this "hallucination." You should treat AI output the way you would treat advice from a well-read but slightly overconfident friend: useful, often correct, worth double-checking when it matters.

AI is not one thing. "AI" is a family name, like "vehicle." A spam filter, a chess program, a self-driving car and a chatbot are all AI, but they are as different from each other as a bicycle is from a cargo ship.

The words you will keep hearing

You do not need to memorise these, but it helps to have seen them once.

• Machine learning: the general method of learning from examples. Most modern AI is this.
• Model: the finished product of training. When people say "the model got better," they mean a new version of the trained software.
• Large language model (LLM): a model trained on enormous amounts of text so that it can read and write. ChatGPT, Claude and Gemini are all built on LLMs.
• Generative AI: AI that produces something new: text, images, music, video, code. Chatbots and image generators are generative AI.
• Prompt: whatever you type to an AI. A question, an instruction, a document to summarise.
• Training: the process of showing the model examples so it learns. This happens before you use it. When you chat with an AI, it is not learning from you in real time.

Why now?

The ideas behind AI are decades old. Three things came together in the last few years to make it suddenly useful:

1. Data. The internet produced a gigantic pile of text, images and video to learn from.
2. Computing power. Specialised chips became fast enough to process that pile.
3. A better recipe. In 2017 researchers published a design called the "transformer" that turned out to be extremely good at learning from text. Nearly every chatbot you have heard of is a descendant of it.

Put together, those three things produced models that could hold a conversation, and in late 2022 one of them was released to the public as ChatGPT. That is the moment AI stopped being a research topic and became a household word.

The one-sentence version

If someone asks you what AI is, you can say this:

"It's software that learned to do things by studying lots of examples, so it can handle tasks nobody could write exact rules for, like understanding language or recognising images. It's very useful and it makes mistakes, so you check its work."

That is a better answer than most people can give. Next, let's look at why any of this matters to you.