What is AGI?

You've probably seen the term artificial general intelligence (AGI) in a headline, often next to a prediction. Most people take it to mean an artificial intelligence (AI) that could do almost anything a person can do, across many kinds of tasks rather than one. It's an idea that excites some people and worries others, so it helps to know what it means, and what it doesn't.

Every AI in use today was built and trained for particular kinds of work. A chatbot predicts language, a navigation app finds routes, and a spam filter sorts mail. Even the most capable chatbots, which can write, summarize, and solve math problems, are one kind of system doing one kind of thing. General intelligence would mean learning and handling new kinds of problems the way people do, without being trained for each one.¹²

Most headlines leave out one thing: there's no agreed definition of AGI, and the people who use the word mean different things by it.

• OpenAI's charter describes it as highly autonomous systems that outperform humans at most economically valuable work.
• Researchers at Google DeepMind proposed levels of AGI instead of a single finish line, judged on how capable a system is and how general.
• Many researchers point out that there's no consensus on what would count, or how to test for it.

So when someone says AGI is a few years away, the first thing to ask is: by whose definition?¹³⁴

Today's best models are impressive on some tests. Stanford's AI Index reports that leading models now meet or beat human baselines on some very hard science and math benchmarks. But a high score on a test isn't the same as general ability. The same systems can still make basic mistakes, struggle with tasks they weren't trained for, and need people to check their work. Being very good at many tests is a long way from being good at everything.⁵

Forecasts of when something like AGI might arrive range from a few years to many decades, and some experts doubt it will arrive in the form people imagine. Part of the spread comes from the missing definition. If people mean different things, they'll predict different dates. And some of the loudest predictions come from people who are building, funding, or selling AI. In 2026, OpenAI's chief executive, Sam Altman, said he expected the company to have a system he would call AGI before the end of the year, by OpenAI's own definition. That doesn't make the prediction wrong. It's one more reason to ask who's talking.⁶⁷

Four questions cut through most AGI headlines.

• By whose definition?
• Measured how, and on what test?
• Who is making the claim, and what do they have riding on it?
• What can the system do today, and what is still a prediction?

You don't need to be an expert to ask these. They're the same questions you'd ask about any bold forecast.

AGI is worth following, because it shapes how companies and governments plan. But the tools in your hands today are specific and checkable, and understanding them is a better use of your attention than worrying about a word nobody has pinned down.