What is artificial general intelligence (AGI)?

In plain English

AGI stands for artificial general intelligence. It refers to a system that could take on more or less any mental task an average adult can — learning a new skill from scratch, reasoning across unfamiliar situations, planning over long horizons, transferring knowledge from one domain to another — rather than excelling at specific jobs it was built for.

Today’s AI, however impressive, is narrow in this sense. A language model that writes a polished essay cannot reliably manage a week-long project, learn from its mistakes between sessions, or notice when it is out of its depth.

Crucially, there is no agreed definition or test for AGI. Some people mean “matches humans at most economically valuable work”, others mean something closer to human-like flexible learning. The vagueness is part of why experts disagree so sharply about how near it is.

Why it matters

AGI is not only a technical question; it drives real decisions now. Companies justify vast investment by it, governments frame safety regulation around it, and forecasts about jobs and the economy depend on it. Claims about AGI being imminent — or impossible — should be read as positions in an active argument, not settled facts.

An example

A capable assistant today can draft your quarterly report if you give it the figures and a clear brief. An AGI, by the definitions usually offered, could be told “sort out our quarterly reporting”, work out what data it needs, ask the right colleagues, spot an anomaly in the numbers, and flag it — without being shown how.

What to watch out for

Beware of anyone declaring AGI achieved or imminent without saying what they mean by it. Impressive benchmark scores can reflect material similar to the training data rather than genuine general ability, and fluent language is easy to mistake for understanding.