A hypothetical system matching human ability across essentially all cognitive tasks, rather than being narrowly good at particular ones. Worth knowing mainly because the term is used very loosely and means different things to different people, from ‘economically valuable work at human level’ to ‘matches the best human experts at everything’. Nothing today meets any strict definition, and predictions about timing vary from a few years to never, which should tell you how weak the evidence behind any of them is.
Technical view
There is no agreed definition or test, which makes claims about progress towards it difficult to evaluate and easy to make. Debate divides broadly between those who expect continued scaling of current approaches to suffice and those who argue that fundamental capabilities, such as robust causal reasoning, reliable long-horizon planning or continual learning, require different methods altogether. The term carries substantial commercial and political weight, so it is worth noting who is using it and to what end.
Related
Part of AI Terminology Index, section: Problems, Risks and Ambitions.