AGI (artificial general intelligence)

The goal is as old as the field. The phrase for it is not, and it did not come from inside AI: the first documented use of "artificial general intelligence" is Mark Gubrud's, in a November 1997 abstract on nanotechnology and international security written for the Fifth Foresight Conference on Molecular Nanotechnology. The label turned up in an abstract about something else, decades after people had begun building toward the thing it names.

Then it turned up a second time, and that arrival is the one the research community kept. Shane Legg suggested the phrase again around 2002, as the title for an edited volume by Ben Goertzel and Cassio Pennachin, and that is how it entered the research vocabulary.

The work in the phrase is done by "general", which makes it a claim about range. A system that plays one game better than any person, or writes code better than most people, is not what the phrase denotes, however good it is at that. The test the words set is breadth: most intellectual work a person can do, across domains. Strength at a single task, however extreme, leaves the claim untouched.

The p(doom) song, annotated dates the phrase to 1997 and glosses it as the thing the song is nervous about, what everyone is waiting for and afraid of at once.

The category is also not the same thing as a document about it. Sparks of AGI is one paper claiming to have seen early signs of general ability in a model, which makes it evidence offered for the category rather than a definition of it.