Artificial General Intelligence (AGI) is a hypothetical form of 📝artificial intelligence able to match or exceed human performance across most cognitive tasks, rather than excelling only within narrow, predefined domains.
AGI is usually defined in contrast to narrow AI: chess engines, image classifiers, and recommendation models that perform one task well but cannot carry that competence into unrelated problems. An AGI would generalize, learning new skills, reasoning across domains, and adapting to unfamiliar situations without being rebuilt for each one. No agreed test marks when that threshold is crossed, and definitions vary between cognitive, economic, and benchmark-based framings.
The term appeared in Mark Gubrud's writing in 1997 and was popularized in the early 2000s by researchers including Shane Legg and Ben Goertzel. It has since become a stated mission for frontier labs: 📝OpenAI's charter defines AGI as highly autonomous systems that outperform humans at most economically valuable work, and 📝DeepMind researchers proposed a "Levels of AGI" framework in 2023 that grades systems by both performance and generality.
AGI is commonly treated as the step before 📝Artificial Super Intelligence (ASI), a system that would far exceed human ability. Timelines are contested: 📝Situational Awareness: The Decade Ahead argues AGI by 2027 is plausible, while skeptics question whether scaling current large language models leads there at all. The pursuit also carries an ideological dimension, which critics examine under the 📝TESCREAL label.