Specialty Digest

Global English editionLearners’ Study Dashboard — in development
The Knowledge Library for All

Artificial General Intelligence

nounArtificial Intelligencealso AGI, strong AI, human-level AI

In one line

Artificial general intelligence (AGI) is a hypothetical AI that can understand, learn and perform any intellectual task a human can, across many domains.

Artificial general intelligence (AGI) is a hypothetical form of artificial intelligence that could understand, learn and carry out any intellectual task a human being can, across many different domains, without being redesigned for each one. No such system has been demonstrated. Today’s AI products are narrow, which means they perform well on specific tasks, such as translating text or recognizing images, but cannot freely transfer their skill to unrelated problems. AGI is a goal for some research labs and a topic of intense debate among scientists, ethicists and policymakers.

History of the AGI concept

The aspiration for a machine with general intelligence goes back to the founding of the field. The 1956 Dartmouth workshop proposed that every feature of learning and intelligence could in principle be described precisely enough for a machine to simulate. As early projects hit limits, most researchers turned to narrower goals. The label artificial general intelligence was popularized in the early 2000s by researchers who wanted to revive work on broad, human-level systems, and it appeared in the title of a 2007 book edited by Ben Goertzel and Cassio Pennachin. Rapid progress in large language models after 2020 brought the topic into public discussion, because these models handle many tasks with a single system.

What AGI would be able to do

Definitions vary, but they usually include a set of abilities.

  • Learning new tasks quickly from limited examples.
  • Transferring knowledge from one domain to another.
  • Reasoning and planning in unfamiliar situations.
  • Understanding language and the physical and social world.
  • Setting subgoals and acting autonomously to reach them.

Some definitions are economic, describing systems that can do most work that people are paid to do. Others focus on cognition, asking for human-level performance on a wide range of mental tasks. The lack of an agreed definition is one reason claims about AGI are hard to evaluate.

AGI vs. narrow AI vs. superintelligence

Narrow AI, also called artificial narrow intelligence, is specialized. It can beat people at chess or flag tumors in scans but cannot do both. AGI would match human-level breadth. Artificial superintelligence is a further hypothetical stage in which a system greatly exceeds the best human minds in most domains. These categories are conceptual labels, not points on a measured scale, and real systems often fall in between.

Approaches to building AGI

Researchers pursue several routes. One is scaling: training ever larger neural networks on more data and computing power and observing which abilities emerge. Another combines learning with symbolic reasoning, in an approach called neuro-symbolic AI. A third studies the brain and tries to copy its organization, and a fourth builds agents that learn by acting in rich simulated or real environments. There is no consensus on which will work, or whether current methods are enough.

How would AGI be measured?

Because no one agrees on a definition, no single test settles whether a system is generally intelligent. The Turing test, which asks whether a machine can pass as a human in conversation, is the oldest proposal, but many researchers consider it a weak measure because conversation can be mimicked without deep understanding. Other proposals use broad benchmark suites that cover mathematics, science, coding and common sense, or ask whether a system can learn a brand new task from only a few examples. Some economists propose practical measures, such as the share of human jobs a system can perform at a competent level. Each approach captures one side of the idea, which is why discussions of AGI often start with the question of what the word means.

The debate over AGI

Experts disagree widely. Optimists argue that general intelligence may emerge from scaling and better algorithms within decades. Skeptics argue that current systems lack reliable reasoning, grounding in the physical world and common sense, and that fundamental breakthroughs remain missing. Forecasts about timing range from years to never, and surveys of researchers show a wide spread. Safety researchers argue that if AGI were built, aligning its goals with human values would be a serious challenge. Others consider that framing premature and prefer to focus on present-day harms such as bias, misinformation and privacy.

Risks, safeguards and governance

Because AGI is hypothetical, most governance today targets existing AI. Frameworks from standards bodies and laws such as the EU Artificial Intelligence Act regulate systems by risk level. Some laboratories publish safety policies describing tests they will run before releasing more capable models. Researchers study AI alignment, interpretability and evaluation methods to catch dangerous capabilities early. Whether such measures will suit systems more capable than anything today is an open question.

Why AGI matters economically and socially

If systems with general human-level ability were built, their effects would reach well beyond technology. Supporters expect faster scientific discovery, cheaper expertise in medicine and education, and large gains in productivity. Critics warn of rapid disruption to jobs, concentration of power in the few organizations able to build such systems, and new risks from misuse. Even without AGI, the prospect shapes investment, research funding and policy, since governments and companies plan around how fast capable systems might arrive. The uncertainty is itself a reason for careful monitoring and open evaluation of what current systems can actually do.

Common misconceptions about AGI

One misconception is that AGI is simply a larger version of today’s chatbots. Whether scaling alone is enough is exactly what experts dispute. A second is that AGI would necessarily be conscious. General capability and subjective experience are separate questions, and nothing in the definition requires awareness. A third is that AGI would arrive on a single day. Progress in AI has been gradual and uneven, with systems becoming better at some tasks long before others, so any boundary may be a matter of definition. Finally, AGI is often confused with science-fiction robots, though a general intelligence could exist purely as software.

FAQs

Does AGI exist today? No. Current systems, including advanced chatbots, are generally considered narrow or partially general, and experts disagree on how close they are to AGI.

Is AGI the same as strong AI? The terms are often used interchangeably, though strong AI sometimes refers specifically to a machine with genuine understanding or consciousness, a stricter claim.

Who coined AGI? The phrase was popularized in the early 2000s by researchers including Ben Goertzel and Shane Legg, though earlier writers described similar ideas.

Key facts

  • Artificial general intelligence refers to a hypothetical system with human-level flexibility across many intellectual tasks, and none has been demonstrated.1
  • The term AGI was popularized in the early 2000s, notably in a 2007 book edited by Ben Goertzel and Cassio Pennachin.2
  • Today's deployed systems are narrow AI, built for specific tasks.3

Related concepts

In the news

Quick checkHas artificial general intelligence been achieved?Show answer

No. AGI remains hypothetical, and experts disagree about how it should be defined, measured and when, if ever, it could arrive.

Sources

  1. Wikipedia, Artificial general intelligence, https://en.wikipedia.org/wiki/Artificial_general_intelligence
  2. Goertzel and Pennachin, Artificial General Intelligence, https://link.springer.com/book/10.1007/978-3-540-68677-4
  3. Stanford Encyclopedia of Philosophy, Artificial Intelligence, https://plato.stanford.edu/entries/artificial-intelligence/

Editorially reviewed by Specialty Digest Editorial TeamLast reviewed November 17, 2026Researched and drafted with AI assistanceReport an issue