What is AGI, and how far are we from it?

What AGI means

Most AI systems we know are built for specific tasks. A chess program plays chess, a translation tool translates. Experts call this narrow AI.

AGI, short for artificial general intelligence, describes the opposite: a general AI that can find its way in a very wide range of areas, learn new tasks and perform at least at human level. The term describes a goal, not a finished product.

How high the bar is set depends on who you ask. OpenAI, the company behind ChatGPT, defines AGI in its charter as highly autonomous systems that outperform humans at most economically valuable work. Other definitions set the bar lower and only require an AI to handle many tasks roughly as well as an average person.

Why there is no clear line

The Association for the Advancement of Artificial Intelligence (AAAI), a major professional body in AI research, stated in a 2025 report that AGI is not a formally defined concept and that there is no agreed test for its achievement.

One reason: human intelligence has many dimensions. According to the report, machines will likely far exceed humans in some areas while remaining weaker in others. It is also unclear which humans to compare against, laypeople or experts.

A group of AI researchers led by Meredith Ringel Morris and Shane Legg therefore proposed, in a paper titled "Levels of AGI", thinking in levels rather than a single finish line, similar to the levels used for self-driving cars. They separate how good an AI is (performance) from how many different tasks it can handle (generality). The levels run from emerging through competent, expert and exceptional to superhuman.

Where today's AI stands

In this level model, language models such as ChatGPT were classed in 2023 as emerging AGI. That meant: for some tasks, such as short essays or simple coding, they reached a solid level, while for most others they were roughly on par with an unskilled person. The authors considered the competent AGI level, which matches many earlier ideas of AGI, not yet reached at the time.

Much has changed since then. The International AI Safety Report 2026 describes clear progress, especially in mathematics, coding and carrying out tasks autonomously. At the same time, it calls these abilities jagged: leading systems can solve some hard problems yet fail at seemingly simple ones, such as counting objects in an image, reasoning about physical space or recovering from basic errors in longer workflows.

Our assessment: whether today's systems have reached the next level is disputed. There is no official body that decides this.

How experts see the road ahead

Opinions differ widely. Roughly three positions can be distinguished.

First, some major AI companies, including OpenAI, have made AGI their explicit goal. Setting such a goal implies they consider it achievable.

Second, many researchers are sceptical. In the AAAI survey of 475 experts, 76 percent said it was unlikely or very unlikely that AGI can be reached simply by scaling up current approaches. They expect fundamentally new ideas will be needed.

Third, some consider AGI a poor goal because the term is too vague, and prefer to measure progress by concrete capabilities.

On timing: in a large survey of 2,778 AI researchers published in 2024, the aggregate forecast for a 50 percent chance that unaided machines outperform humans at every possible task was the year 2047. Full automation of all occupations was given a 50 percent chance only by around 2116. Surveys like this show expectations, not certainties.

What remains open

The International AI Safety Report 2026 considers several paths plausible up to 2030: progress could slow or plateau, for example because of bottlenecks in data or energy. It could continue at the current pace. Or it could accelerate sharply, for example if AI systems begin to speed up AI research itself.

It is also unclear how AGI would be measured at all. The authors of the level model call for new tests that reflect real tasks people value, not just metrics that are easy to count.

Finally, this is not only about technology. The more independently AI systems act, the more important questions of safety, control and effects on work and society become. These are being debated intensively in research and politics.