What Is AGI in Artificial Intelligence?

AGI (artificial general intelligence) is a hypothetical AI system that could match or exceed human cognitive abilities across any task.

At Layer3Labs, we build AI workflow automations for small and mid-sized businesses, so we judge AI by the work it handles today rather than by whether it counts as AGI.

As of October 2026, no lab has shipped a system that researchers agree is AGI, and there is no agreed test that would settle the question.

What AGI Means

Artificial general intelligence (AGI) is a hypothetical AI system that could learn and perform most intellectual tasks a human can, at human level.

According to IBM, AGI represents "a hypothetical stage in the development of machine learning" where software "can match or exceed the cognitive abilities of human beings across any task."

In the OpenAI Charter, OpenAI defines AGI as "highly autonomous systems that outperform humans at most economically valuable work."

Outside the technology sector, the same three-letter acronym carries an unrelated legal meaning.

In United States federal tax filing, AGI stands for adjusted gross income.

The Internal Revenue Service (IRS) defines adjusted gross income as "your total (gross) income from all sources" minus certain adjustments.

Competing Definitions of AGI

Researchers disagree over whether today's most advanced large language models already count as AGI.

As documented on Wikipedia, whether AGI exists "remains a subject of intense debate within the AI community," with no agreed test or consensus that any existing system qualifies.

IBM states that "current LLMs are not considered AGI," noting that while some researchers argue advanced large language models (LLMs) qualify, many other computer scientists disagree.

Our AI vs AGI comparison shows where ChatGPT and other current tools sit in that debate.

The Levels of AGI Framework

The Levels of AGI framework establishes a formal taxonomy to evaluate machine learning progress by separating performance levels from task scope.

In the "Levels of AGI" paper, published on arXiv in November 2023 and revised in September 2025, Morris and co-authors (including Shane Legg) classify systems across six performance tiers.

The authors contrast narrow systems against general systems across these performance tiers.

In the paper's findings, the authors state that Competent AGI "has not been achieved by any public systems at the time of writing."

The authors write that Competent AGI "best corresponds to many prior conceptions of AGI."

The paper's own examples of Level 1 Emerging AGI are ChatGPT, Bard, Llama 2 and Gemini; it gives no rating for models released after it.

• Level 0 (No AI): Narrow systems include simple calculators and compilers; general systems involve human-in-the-loop computing like Mechanical Turk.
• Level 1 (Emerging): Performance is equal to or somewhat better than an unskilled human; narrow examples include classic symbolic AI and SHRDLU, and the paper's general examples at the time of writing were ChatGPT, Bard, Llama 2 and Gemini.
• Level 2 (Competent): Performance reaches at least the 50th percentile of skilled adults; narrow implementations include Siri, Alexa, Watson, and Jigsaw, and the paper said no public system had reached Competent AGI at the time of writing.
• Level 3 (Expert): Performance reaches at least the 90th percentile of skilled adults; narrow examples include Grammarly, Imagen, and DALL-E 2, and the paper listed Expert AGI as not achieved at the time of writing.
• Level 4 (Exceptional): Performance reaches at least the 99th percentile of skilled adults, renamed from Virtuoso in the latest revision; narrow implementations include Deep Blue and AlphaGo, and the paper listed Exceptional AGI as not achieved at the time of writing.
• Level 5 (Superhuman): Outperforms 100% of humans; narrow implementations include AlphaFold, AlphaZero, and Stockfish, while general superhuman capability corresponds to artificial superintelligence (ASI), which has not yet been achieved.

How AGI Would Be Measured

ARC-AGI tests general capability by measuring how efficiently a system learns tasks it has never seen.

The ARC Prize administers the ARC-AGI benchmark, developed from François Chollet's 2019 research on artificial intelligence evaluation.

It tests fluid intelligence rather than stored knowledge, across ARC-AGI-1, ARC-AGI-2 and ARC-AGI-3.

Microsoft and OpenAI added a verification step in a partnership update announced on the Microsoft official blog on October 28, 2025.

The agreement explicitly states: "Once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel."

Origins and Evolution of the Term

According to Wikipedia, Mark Gubrud used the term in 1997.

Researchers Shane Legg and Ben Goertzel subsequently revived and popularized the concept in the early 2000s.

IBM credits Ben Goertzel with popularizing the phrase in 2007 following a suggestion from Shane Legg.

Distinctions Across AGI, ASI, and Federal SI

AGI means human-level ability across tasks. ASI and the federal SI label each mean something different.

Artificial superintelligence (ASI) describes a hypothetical software system that surpasses human cognitive capability across virtually all domains.

Nick Bostrom defines superintelligence in the Wikipedia entry on superintelligence as "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest."

IBM defines ASI as "a hypothetical software-based artificial intelligence (AI) system" possessing "an intellectual scope beyond human intelligence," and calls it "still a hypothetical future state."

In the Levels of AGI matrix, Level 5 Superhuman General AI corresponds directly to ASI, which no organization has developed.

To explore the technical distinctions between human-level versatility and superhuman systems, read our direct comparison of AGI vs ASI alongside our overview of what is superintelligence.

Federal administrative policy introduces a separate linguistic distinction that should not be confused with superintelligence.

On September 29, 2026, the White House issued Executive Order 14434, instructing executive-branch agencies to substitute the term "Super Intelligence" (SI) for "AI" in non-statutory documents.

The order defines SI by the statutory AI definition in 15 U.S.C. 9401(3).

Federal SI therefore means today's AI rather than superhuman capability, and the order never mentions AGI or ASI. Our AI vs SI comparison covers the label.

Timeline Projections and the Current Consensus

No scientific consensus exists regarding when artificial general intelligence will arrive, and no authoritative research institution provides a verified delivery date.

OpenAI's leadership put the nearest thing to a horizon in writing in 2023, and it was about superintelligence rather than AGI.

In their May 2023 post on the governance of superintelligence, OpenAI leadership wrote that "it's conceivable that within the next ten years, AI systems will exceed expert skill level in most domains."

The post gives no year for AGI.

For more on OpenAI's latest claims, see our analysis of whether GPT-6 Astra is AGI.