What is artificial intelligence?

"AI" is everywhere: in the news, in advertising, at work, on your phone. Yet the word is used for very different things, from the filter that keeps spam out of your inbox to a chat assistant that can write a letter with you.

This page gives you a simple picture of what AI is today, what kinds there are, and what the word does not mean. You don't need any technical background.

A short answer

Most artificial intelligence (AI) today is software that learns patterns from examples and uses them to produce something useful: a prediction, a recommendation, a decision, or new content such as text or images.

That is close to how official bodies describe it. The OECD, an international organization of governments, defines an AI system as a machine-based system that, for objectives set by people, "infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments" (updated definition, November 2023). The European Union's AI Act uses almost the same wording.

The key word is infers: the system works out an answer from what it receives, instead of simply following a fixed list of instructions.

How is that different from ordinary software?

Ordinary software follows rules that a programmer wrote down step by step: "if the customer clicks this button, show that page." Everything it does was spelled out in advance.

A lot of modern AI works differently. Instead of writing all the rules, people show the system a very large number of examples, and it learns the patterns itself. A spam filter is not given a complete list of spam messages; it has seen many examples of spam and normal mail, and it has learned what tends to set them apart. That is why it can also recognize a spam message it has never seen before.

How that learning works, and how a language model becomes a chat assistant, is explained on the page How AI works.

The name "artificial intelligence" itself is older than most people think. It appears in the title of a 1955 proposal by four researchers for a summer workshop at Dartmouth College in 1956, a workshop Dartmouth College calls "the birth of this field of research." You can read more in the short history of AI.

Two broad kinds of AI

"AI" is a very broad term. It helps to separate two kinds.

Narrow AI: one system, one task

Most AI you have used for years was built for a single job. It can be very good at that job, but it can do nothing else. Some everyday examples:

the spam filter in your email;
the route planner in a navigation app, which predicts traffic and suggests the fastest way;
face unlock on a phone;
product or film recommendations in a web shop or streaming service ("you might also like…");
the automatic translation of a single sentence, or speech recognition that turns your voice into text.

A spam filter cannot plan a route, and a route planner cannot recognize your face. Each one is a specialist.

General-purpose AI: one system, many tasks

More recently, a different kind of AI has become widely available: systems that can handle many different tasks, mostly through language. The best-known are large language models, or LLMs. A language model is an AI system trained on enormous amounts of text, so that it has learned how language works and a great deal about what people have written. "Large" refers to the size of the model and of the amount of text it learned from.

One and the same LLM can, for example:

answer questions and explain things;
summarize a long document;
translate between languages;
help you write or rewrite an email, a report or a speech;
help with computer code.

You usually meet an LLM through a chat assistant: a program in which you type (or say) something and get an answer. As of September 2026, well-known examples include ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google). These are named as examples, not recommendations.

The International AI Safety Report, an international scientific review of what general-purpose AI can do and what risks it brings, describes "general-purpose AI" simply as AI models and systems "that can perform a wide variety of tasks" (February 2026).

"General-purpose" is not "science fiction"

Be careful with one possible confusion. "General-purpose" here only means usable for many tasks. It does not mean what is sometimes called artificial general intelligence (AGI): the idea of a machine that equals or surpasses humans in practically everything, the way it appears in films and novels. Today's systems are not that. Whether and when anything like it could ever be built is an open question. Even the near future is uncertain: the same report says AI progress through 2030 could slow down, continue at the current pace or speed up sharply.

In fact, today's general-purpose AI is surprisingly uneven. The same International AI Safety Report notes that these systems do well at complex things such as writing code or answering expert-level questions in science, yet "struggle with some tasks that seem more straightforward, such as counting objects in an image, reasoning about physical space, and recovering from basic errors in longer workflows."

What AI is not

Many images of AI come from science fiction. They make good stories, but they are a poor guide to the systems we actually have. A few gentle corrections:

AI is not a robot. Most AI is software on a phone, a computer or a server somewhere. Robots can use AI, but the two are not the same.
AI is not all-knowing. A chat assistant can sound very sure of itself and still be wrong. Fluent language is not the same as being right. The page Why AI makes mistakes explains why, and Checking answers shows how to deal with it.
AI is not the same as a search engine. An LLM does not simply look up a stored answer; it produces a new text each time. Some assistants can also search the web, but that is an extra function.
AI is not magic, and it is not one single thing. "AI" is a collective name for many different systems, made by many different companies and researchers, each with its own strengths and weak spots.
Whether AI has a mind of its own is an open question. Whether systems like LLMs have anything like understanding, experience or feelings is debated. Some think there is nothing "inside" at all; others think the question deserves careful study. Nobody can settle it today, and this website tries not to pretend otherwise, in either direction.

More common misunderstandings are discussed on the page AI myths.

A picture you can pass on

If someone asks you "what is AI, really?", you could say something like this:

AI is software that has learned patterns from lots of examples, instead of following only rules that someone typed in. Some AI does just one job, like filtering spam or planning a route. Newer AI, like the chat assistants, has learned from huge amounts of text and can help with many language tasks. It can be very useful, but it can also be confidently wrong, and it is not the thinking robot from the movies.

What this website focuses on

This website is mostly about large language models and what is built on them: chat assistants, and increasingly AI agents, systems that use an LLM not only to answer but to carry out tasks for you, such as looking things up, filling in forms or working with files. Those are the systems that are changing daily life, work and school most visibly right now, and they raise the most questions about how humans and AI can live and work well together.

Where to go next

How AI works: how these systems are made, without mathematics.
When AI is wrong: why AI makes mistakes.
Strengths and limits: what AI does well, and what it can't do yet.
AI agents: AI that carries out tasks, not only answers.
Using AI: practical guidance for everyday use.

Sources

Updates to the OECD's definition of an AI system explained, Stuart Russell, Karine Perset and Marko Grobelnik, OECD.AI (OECD AI Policy Observatory), 2023-11-29 (consulted 2026-09-28)
Article 3: Definitions, EU Artificial Intelligence Act (Regulation (EU) 2024/1689), text as published at artificialintelligenceact.eu (consulted 2026-09-28)
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, August 31, 1955, John McCarthy, Marvin L. Minsky, Nathaniel Rochester, Claude E. Shannon, reprinted in AI Magazine 27(4), 2006-12-15 (consulted 2026-09-28)
The Research Conference Where AI Began, Dartmouth College, undated (consulted 2026-09-28)
2026 Report: Executive Summary, International AI Safety Report, 2026-02-03 (consulted 2026-09-28)