artificial general intelligence
A level of machine capability, or a system that reaches it, matching or exceeding people across most intellectual tasks rather than in one specialty.
- GPT-4, called an early, incomplete AGI by Microsoft researchers in 2023
- ChatGPT and Bard, rated emerging AGI by Google DeepMind researchers
- OpenAI’s Astra, which its president personally counted as AGI in 2026
- Earliest AI use
- 1989
- Merriam-Webster’s first known use. The dictionary does not name the source.
- Also called
- Strong AI
- Ray Kurzweil’s usage. Philosophers use it for a Searle hypothesis on minds.
- Opposite
- Narrow AI
- AI built for one task or domain. Kurzweil paired it with strong AI.
In plain words
Artificial general intelligence (AGI) is AI that would equal or surpass people across most intellectual tasks; how to define and recognize it is disputed.
Most AI systems are built for one job, such as playing chess or flagging fraud. Artificial general intelligence, or AGI, describes a system that could match or beat people across most kinds of intellectual work. Merriam-Webster defines it as that potential capability and as a system that has it.1
No definition is agreed, and each sets a different bar. OpenAI’s Charter uses an economic test: autonomous systems that beat people at most economically valuable work. Google DeepMind researchers proposed five levels, from emerging to superhuman, and in 2023 put chatbots such as ChatGPT at the lowest.2 François Chollet, who created the ARC benchmark, asks how efficiently a system learns new skills.3
Companies have made the term a goal and a contract term. OpenAI states its mission as ensuring that AGI benefits all of humanity. Its deals with Microsoft tied payments to AGI.4 In October 2025 the two agreed that an independent expert panel would verify any OpenAI declaration of it. Terms announced in April 2026 set fixed dates instead and do not mention AGI.
Claims and forecasts differ by who makes them. In 2023 Microsoft researchers called GPT-4, a large language model, an early and incomplete version of AGI. In September 2026 OpenAI’s president, Greg Brockman, said he personally thought OpenAI had reached it. Meta’s AI chief, Yann LeCun, told TIME in 2024 that he disliked the name because human intelligence is not general. In a survey published in January 2024, AI researchers gave even odds that machines would outperform humans at every task by 2047.5
AGI is often confused with generative AI, which names a kind of output, and with superintelligence, ability far beyond the best humans. Ray Kurzweil called it strong AI, a name philosophers use for a different question about machine minds.6
How it works
No agreed test decides whether a system counts as AGI. The proposals measure different things and draw the line in different places.
Capability scales rate how well a system performs and how many kinds of task it handles. Google DeepMind’s scale runs from emerging, the only level its authors judged reached in 2023, to superhuman.2
Economic tests count work or money. OpenAI’s Charter looks at economically valuable work. In December 2024, TechCrunch relayed a report by The Information on the Microsoft and OpenAI agreement. It said the deal defined AGI as systems able to generate at least $100 billion in profits.
Learning tests ask how quickly a system picks up something new. Chollet argues that skill at a familiar task can be bought with more training data.3 His benchmark relies on knowledge close to what humans are born with, so machines can be compared fairly with people.
Declaration rules settle who decides. Under the October 2025 terms, OpenAI would declare AGI and an independent expert panel would verify the claim.4 In September 2026, OpenAI’s president said no contractual AGI trigger remained.
Origin and first known use
The phrase adds general, meaning not limited to one field, to artificial intelligence. Merriam-Webster dates its first known use to 1989 but does not name the source. The earliest use with a named author is Mark Gubrud’s 1997 article on future technology and its risks. Around 2002 Shane Legg proposed the phrase to Ben Goertzel and Cassio Pennachin as a book title. They preferred it to strong AI, which philosophers already used for a hypothesis of John Searle’s. Legg has said he meant a field of study, not an artifact. OpenAI’s 2018 Charter treated AGI as systems that could be built.
- 1989 Merriam-Webster dates the first known use of the phrase, in its current dictionary sense, to 1989. The entry does not name the source.Merriam-Webster, “artificial general intelligence” merriam-webster.com →: source for 1989
- 1997 Mark Gubrud uses the phrase in an article on future technology and its risks. Ben Goertzel later cites it as the earliest use he knows of.Ben Goertzel, “Who coined the term ‘AGI’?”, 28 August 2011, on Mark Gubrud’s 1997 article goertzel.org →: source for 1997
- 2007 Springer publishes Artificial General Intelligence, a volume edited by Ben Goertzel and Cassio Pennachin that names a research field. Shane Legg suggested the title around 2002.Goertzel and Pennachin (eds.), “Artificial General Intelligence”, Springer, Cognitive Technologies series, 2007 springer.com →: source for 2007
- 2018 OpenAI’s Charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work”.OpenAI, “OpenAI Charter”, April 2018 openai.com →: source for 2018
- 2023 Google DeepMind researchers, Shane Legg among them, review existing definitions of AGI and propose levels for its performance, generality and autonomy.Morris et al., “Levels of AGI for Operationalizing Progress on the Path to AGI”, arXiv, 4 November 2023 arXiv 2311.02462 →: source for 2023
- 2024 TechCrunch relays a report by The Information that the Microsoft and OpenAI agreement defines AGI as AI systems able to generate at least $100 billion in profits.Maxwell Zeff, TechCrunch, 26 December 2024, reporting The Information techcrunch.com →: source for 2024
- 2026 New Microsoft and OpenAI terms set fixed dates: OpenAI pays Microsoft a revenue share through 2030, independent of technical progress. The announcement does not mention AGI.OpenAI, “The next phase of the Microsoft OpenAI partnership”, 27 April 2026 openai.com →: source for 2026
- 2026 At the launch of OpenAI’s Astra model, president Greg Brockman says there is no longer a contractual AGI trigger, and that personally he thinks OpenAI has reached AGI.Lucas Ropek, “OpenAI launches Astra, its powerful (and controversial) new model”, TechCrunch, 3 September 2026 techcrunch.com →: source for 2026
In the wild
-
“Shane Legg, an AI researcher who had worked for me previously, came up with Artificial General Intelligence.”
-
“Our mission is to ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity.”
-
“…it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.”
Bubeck et al., Microsoft Research, arXiv preprint arXiv 2303.12712 →: Bubeck et al., Microsoft Research, arXiv preprint, 22 Mar 2023Archived →: copy of Bubeck et al., Microsoft Research, arXiv preprint, 22 Mar 2023 -
“We propose a framework for classifying the capabilities and behavior of Artificial General Intelligence (AGI) models and their precursors.”
Morris et al., Google DeepMind, arXiv preprint arXiv 2311.02462 →: Morris et al., Google DeepMind, arXiv preprint, 4 Nov 2023Archived →: copy of Morris et al., Google DeepMind, arXiv preprint, 4 Nov 2023 -
“…I don’t like to call it AGI because human intelligence is not general at all.”
Yann LeCun, interview with TIME time.com →: Yann LeCun, interview with TIME, 13 Feb 2024Archived →: copy of Yann LeCun, interview with TIME, 13 Feb 2024 -
“Microsoft and OpenAI have a financial definition of AGI: Report”
-
“Artificial general intelligence (AGI), AI that’s at least as capable as humans at most cognitive tasks, could be here within the coming years.”
Anca Dragan, Rohin Shah, Four Flynn and Shane Legg, Google DeepMind blog deepmind.google →: Anca Dragan, Rohin Shah, Four Flynn and Shane Legg, Google DeepMind blog, 2 Apr 2025Archived →: copy of Anca Dragan, Rohin Shah, Four Flynn and Shane Legg, Google DeepMind blog, 2 Apr 2025 -
“Once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel.”
OpenAI, announcement of revised Microsoft terms openai.com →: OpenAI, announcement of revised Microsoft terms, 28 Oct 2025Archived →: copy of OpenAI, announcement of revised Microsoft terms, 28 Oct 2025 -
“RSI is the new AGI — and it’s just as hard to pin down”
Not to be confused with
-
Generative AI
Generative AI names a kind of output: new text, images or code made on request. AGI names a level of ability across most intellectual tasks. A system can generate fluent content without that breadth.
-
Superintelligence
Superintelligence usually means ability well beyond the best humans in nearly every field. AGI usually sets the bar near human level. In Google DeepMind’s 2023 levels, superhuman performance is the top level of AGI.
-
Strong AI
In philosophy, strong AI names the hypothesis, associated with John Searle, that a machine that behaves as if it has a mind really has one. AGI concerns ability, not minds. Ray Kurzweil used strong AI for roughly AGI.
Questions
What is artificial general intelligence (AGI)?
Artificial general intelligence, or AGI, is AI that would match or exceed people across most intellectual tasks, instead of excelling at one. No definition is agreed. OpenAI’s Charter uses an economic test: highly autonomous systems that do most economically valuable work better than people. A 2025 Google DeepMind post describes AI at least as capable as humans at most cognitive tasks. François Chollet looks at how efficiently a system learns new skills.
Has AGI been achieved yet?
There is no agreement. In 2023 Microsoft researchers wrote that GPT-4 could be viewed as an early, incomplete AGI. Google DeepMind researchers judged the same year that chatbots such as ChatGPT had reached only emerging AGI, the lowest of five levels. In September 2026 OpenAI’s president, Greg Brockman, said he personally thought OpenAI had reached AGI. Yann LeCun argued in 2024 that human intelligence itself is not general.
When will AGI arrive?
Forecasts depend on who makes them and on the definition used. In a survey published in January 2024, 2,778 AI researchers gave even odds that unaided machines would beat humans at every task by 2047. That was 13 years sooner than in the same team’s survey a year earlier. A 2025 Google DeepMind post said AGI could be here within the coming years.
Who coined the term artificial general intelligence?
Merriam-Webster dates the first known use to 1989 but does not name the source. The earliest use with a named author is a 1997 article by Mark Gubrud on future technology and its risks. Around 2002 Shane Legg proposed the phrase to Ben Goertzel and Cassio Pennachin as a book title. Springer published their edited volume, Artificial General Intelligence, in 2007.
What does AGI mean in the Microsoft and OpenAI deal?
In December 2024, TechCrunch relayed a report by The Information on the agreement. It said AGI was defined as systems able to generate at least $100 billion in profits. Under terms announced in October 2025, an independent expert panel would verify any AGI declaration by OpenAI. Terms announced in April 2026 set fixed dates and do not mention AGI. In September 2026 OpenAI’s president said no contractual AGI trigger remained.
Watch
Cite this entry
“Artificial general intelligence.” SUPERR_dictionary, Superrduperr, 24 Sept. 2026, superrduperr.com/
Superrduperr. (2026, September 24). Artificial general intelligence. In SUPERR_dictionary. https://superrduperr.com/
Superrduperr. “Artificial general intelligence.” SUPERR_dictionary. September 24, 2026. https://superrduperr.com/