artificial general intelligence (AGI)

basicoutlook

Artificial intelligence that is not yet achieved, capable of solving almost any intellectual task at least as well as a trained human.

In plain terms

Earlier AI systems were narrow specialists: one played chess, another recognized faces, a third translated. Today's large models are much more versatile, but still not reliable generalists. Artificial general intelligence would be a system that learns any new task, just as an intelligent human does.

The point is not whether it is better than a human in a specific area, as such machines have existed for a long time (a calculator multiplies faster than you). The point is generality: the same system must be able to plan, learn, navigate unknown situations, and transfer knowledge from one area to another.

AGI today is more of a goal and a topic of debate than a product. Major AI companies explicitly aim to build it, but researchers disagree on when we can say it has been achieved and when we might reach it.

How it works

There is no ready-made recipe: no one knows for sure which technique can build AGI. The following steps therefore show where the search for the path is heading.
The most widespread path is the further development of today's large language models: more data, greater computing capacity, better training methods, and the combined handling of text, image, audio, and video.
Models are increasingly used as agents: the system not only answers but independently plans multi-step tasks, uses tools, checks results, and corrects errors.
Many researchers consider this insufficient, noting missing capabilities: long-term memory, continuous learning from experience, reliable reasoning, and understanding of the physical world.
Progress is measured through tests and levels, for example by the proportion of tasks where the machine reaches the performance of a trained human. There is no single, universally accepted standard.

An everyday example

A technology company executive states in the news that "we will reach AGI within a few years," while experts below the article argue about what this means. In such cases, it is worth checking how each defines the term, because often very different things are discussed with the same word.

Who is affected

everyoneIf realized, work, science, the economy, and power relations could change simultaneously, making it a public matter who develops it and under what rules.

employeeThe intermediate results on the path to AGI are already transforming intellectual work, even if AGI itself is not yet here. → Work & HR

investor, saverThe high valuation of AI companies is partly based on promises regarding AGI, which may not be fulfilled on time, and this can affect small investors through pension and investment funds.

lawyer, citizenRegulation must already decide what safety and liability frameworks to impose on the most powerful models. The EU AI Act, for instance, has imposed specific obligations on the largest general-purpose models since August 2025. → Law

A common misconception

MythChatGPT and its peers are already general artificial intelligences, as they can converse about almost anything.

In realityToday's models are convincing in many areas, but they fail unpredictably, do not learn permanently from their experiences, and often get stuck on long, independent tasks. There is no professional consensus that this could be called AGI.

Limits and risks

There is no unified definition: every company and researcher sets the bar differently, so the claim 'we have achieved AGI' means little on its own.
Predictions regarding timelines are unreliable; estimates range from a few years to several decades, and the predictors are often commercially interested.
The raised risks are serious: massive workplace restructuring, concentration of power in a few companies or states, and the question of whether the goals of such a system can be aligned with those of people.
The debate on AGI easily diverts attention from the real, already present problems of today's AI, such as errors, bias, and data protection.

Similar, but not the same

Narrow artificial intelligence – Narrow AI is designed for a well-defined task (such as face recognition or translation), whereas AGI would be capable of any intellectual task.
superintelligence – AGI refers to roughly human-level general capability, while superintelligence denotes one that far surpasses even the best humans in every field.
generative AI – Generative AI is an existing technology that produces text, images, or audio, while AGI is a capability level not yet reached.

Related concepts

Broader

artificial intelligence

Narrower

–

On the edge

superintelligence, Narrow artificial intelligence, Technological singularity, AI safety, AI-agent

A brief history

The idea of human-level machine intelligence has been present since the beginning of artificial intelligence research in the 1950s. The term "artificial general intelligence" itself was used by Mark Gubrud in 1997, and was popularised in the 2000s by Ben Goertzel and Shane Legg.

Where things stand

As of late 2026, there is no system that broad professional consensus would call AGI. Leading AI companies have openly set this as their goal, and models are developing rapidly, but debate remains lively regarding both the definition and the timeline.

Status: 2026 November

In the news

Appears in 5 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Hassabis says AGI could arrive within two to three yearsOct 21, 2026 · Agent Age
Google and DeepMind launch institute to study the impact of AGISep 16, 2026 · Agent Age
ChatGPT, Grok, Claude and Gemini go down at the same timeSep 3, 2026 · Agent Age
OpenAI releases GPT-6 Astra and says the "AGI era" has begunSep 1, 2026 · Agent Age
Altman admits he overestimated how fast AI would disrupt the economyAug 24, 2026 · Agent Age

Also known as

AGI · general AI · artificial general intelligence · general AI · strong AI · human-level artificial intelligence · Artificial General Intelligence · human-level AI

Sources

Morris et al. (Google DeepMind): Levels of AGI – Operationalizing Progress on the Path to AGI (arXiv, 2023)

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.