Artificial general intelligence
In short
Artificial general intelligence (AGI) is a hypothetical AI that could match or beat skilled humans across most thinking tasks, not just one. There is no agreed definition or test for it, and experts disagree on whether or when it will arrive.
- Also called
- AGI, General AI
- Topic
- The basics
- Level
- Some background helps
- First described
- According to Morris and colleagues, the term was first used in a 1997 article on military technology by Mark Gubrud, and was popularised among computer scientists from 2001 by Shane Legg and Ben Goertzel1.
Most AI today is good at particular jobs. AGI describes a system with the breadth of a capable human. Definitions vary widely: a 2023 Google DeepMind paper reviewed nine prominent ones1.
Some focus on thinking tasks. The paper credits Shane Legg and Ben Goertzel with popularising the term among computer scientists in 2001, describing a machine able to do the cognitive tasks people typically can1. OpenAI's charter takes an economic angle, defining AGI as "highly autonomous systems that outperform humans at most economically valuable work"1.
The same paper proposed "Levels of AGI", rated by both performance and generality, from Level 1 "Emerging" to Level 5 "Superhuman", which outperforms 100% of humans1. It judged that frontier chatbots of September 2023, such as ChatGPT, counted only as "Emerging AGI", and that "Competent AGI" had not yet been achieved1.
AGI also carries commercial weight. In October 2025 Microsoft said that once OpenAI declares AGI, the declaration will be verified by an independent expert panel, and some of Microsoft's rights depend on that verification2.
AGI is not the same as consciousness. The DeepMind paper notes there is no scientific consensus on how to tell whether machines are conscious, so it defines AGI by what a system can do instead1.
Example
The scientist Gary Marcus proposed five tests, including understanding a novel, working as a competent cook in an unfamiliar kitchen, and writing a bug-free 10,000-line program1. A chatbot that writes well but cannot cook fails his bar for general intelligence.
Why it matters
AGI is the stated goal of several big AI labs and features in billion-dollar contracts2, yet nobody agrees how to tell when it has arrived.
Questions people ask
Has AGI been achieved?
There is no agreed answer because there is no agreed definition or test. In late 2023, Google DeepMind researchers rated chatbots like ChatGPT as "Emerging AGI" and said "Competent AGI" had not been reached1.
Who coined the term AGI?
Mark Gubrud used it in a 1997 article about military technologies. Shane Legg and Ben Goertzel popularised it among computer scientists from 20011.
Is AGI the same as superintelligence?
No. AGI usually means human-level breadth across most tasks. In the Levels of AGI framework, "Superhuman" is a separate top level for systems that outperform 100% of humans1.