The term "artificial general intelligence" entered the literature in the 1990s. Mark Gubrud used it in a 1997 discussion of military technology, and it was taken up and popularised in the 2000s by researchers including Ben Goertzel and Shane Legg, who used it to separate the goal of general intelligence from the narrower systems that dominated the field.
Why a new term was needed
By the 2000s, "AI" had been captured by specialist systems: chess engines, recommender systems, vision classifiers. Each was a triumph within its lane and none was aimed at the original ambition of the field. AGI was a way of naming that ambition again without pretending the specialist systems were it.
The older lineage
The idea predates the phrase. Turing's 1950 paper asked whether machines could think, and the Dartmouth proposal of 1955 assumed that every aspect of learning or intelligence could be described precisely enough to be simulated. AGI is a modern label on a question that is older than the computer.
What the term is doing now
Since 2023 the phrase has moved from research papers into product marketing, which has stretched it. The original technical meaning, breadth plus transfer, is still the useful one.