Black October, explainedAppsGuidesAI DictionarySmall FixesOn RecordAboutFollow the lab

Artificial general intelligence

Also called AGI, general AI, human-level AI

AGI, or artificial general intelligence, is the idea of an AI that can learn and do almost any thinking task a person can, not just one kind.

The basicsEveryday

In plain words

Today's AI tools are impressive but uneven. A chess program cannot hold a conversation. A chatbot can write a cover letter in seconds and still trip over a puzzle a child would solve.

AGI describes something broader: one system that could learn new skills, reason through unfamiliar problems and handle most mental work at the level of a capable adult. General, as opposed to narrow.

It does not exist yet, as of October 2026. There is also no agreed definition and no agreed test. Some companies define it by economic value, some researchers by a list of abilities, and others say the term is too vague to be useful.

That fuzziness matters, because claims that AGI is near, or far, often depend on which definition the speaker has in mind.

Everyday example

A billing machine vs a new hire

A pharmacy billing machine is brilliant at one job and useless at anything else. Now imagine a new office hire who can do the accounts on Monday, draft a legal letter on Tuesday and fix the Wi-Fi on Wednesday, learning each from scratch within a day. Narrow AI is the machine. AGI would be that hire: one mind that adapts to whatever the week brings.

Where the example breaks: People argue whether AGI would need to be humanlike at all. It might be general in a very different way from a person: strong at things no human can do and odd at others.

Go deeper

How it actually works

Because there is no single definition, researchers have tried to make the idea measurable.

  1. Where the term comes from

    Researcher Shane Legg suggested "artificial general intelligence" around 2002 as the title of a book edited by Ben Goertzel and Cassio Pennachin, published by Springer in 2007. Mark Gubrud had used the phrase in a 1997 article.

  2. Definitions in use

    OpenAI's charter speaks of "highly autonomous systems that outperform humans at most economically valuable work." Others define AGI as human-level performance across a broad set of thinking tasks, or as the ability to learn new tasks as efficiently as people do.

  3. Levels instead of a finish line

    A 2023 Google DeepMind paper proposed "levels of AGI", grading systems on two axes: how well they perform and how many kinds of task they handle. In that view, progress is gradual rather than one moment.

  4. How it might be measured

    Benchmarks test reasoning, coding, science and real work tasks. Models now score highly on many of them, but critics note that tests get used up and that passing a test is not the same as general ability.

Term spread by
Shane Legg, Ben Goertzel and Cassio Pennachin, around 2002 to 2007
Status
Not achieved and no agreed test, as of October 2026
Survey of 2,778 AI researchers (2023)
The combined forecast gave a 50% chance by 2047 that machines outperform humans at every task
AAAI community survey (2025)
76% of respondents (475 took part) said scaling up current approaches is unlikely to produce AGI

How close is AGI?

One view

It could arrive within years

Some lab leaders and researchers point to fast, steady gains from scaling up models, data and computing power, and to systems that already match experts on many tests. They argue that general ability may come from pushing current methods further.

The other view

It needs new ideas, or is far off

Many researchers argue today's models still lack reliable reasoning, lasting memory, learning from a few examples and grounding in the physical world. In a 2025 AAAI survey, 76% of respondents said scaling current approaches alone is unlikely to produce AGI.

Most likely: There is no consensus. Forecasts range from a few years to many decades and depend heavily on which definition is used. Treat any confident date, in either direction, as an opinion rather than a fact.

Questions people ask

Does AGI exist yet?

No, not by any widely accepted definition, as of October 2026. Some people argue that today's most capable models show early or partial general intelligence. Others say they are still narrow systems with broad reach. Much of the disagreement is about definitions.

What is the difference between AI and AGI?

AI is the whole field and includes every system in use today, nearly all of which handle specific kinds of task. AGI is a hypothetical kind of AI that could learn and perform almost any intellectual task a person can, at a similar level.

What is the difference between AGI and superintelligence?

AGI means roughly human-level ability across most thinking tasks. Superintelligence means ability far beyond the best humans in practically every field. Some think one could quickly lead to the other; others doubt either is close.

When will AGI be achieved?

Nobody knows. A large survey of AI researchers run in 2023 gave a 50% chance that machines outperform humans at every task by 2047, but answers ranged widely and such forecasts have shifted a lot between surveys. Some lab leaders predict much sooner; many researchers expect later.

Is ChatGPT AGI?

Most researchers would say no. ChatGPT runs on large language models that handle a wide range of tasks but remain unreliable in ways a generally capable person is not, such as making up facts and failing some simple reasoning.

Further reading

Prepared by Mundhra Labs. Reviewed by Aditya Mundhra. Checked October 2026.