Artificial General Intelligence (AGI) refers to a hypothetical machine able to understand, learn and perform any intellectual task that a human being can. Whereas today's AI is a collection of specialised tools — one plays chess, another translates, a third generates images — AGI would be a single versatile mind, able to move from one domain to another the way we naturally do.
Narrow, general, super-intelligence
Three commonly confused levels are worth distinguishing:
| Type | Capability | Example |
|---|---|---|
| Narrow AI (ANI) | A single task | ChatGPT, AlphaGo, spam filters |
| AGI | Any human task | Does not yet exist |
| Super-intelligence (ASI) | Surpasses humans everywhere | Theoretical hypothesis |
Today, all the AI deployed in the world is narrow — even the most impressive large language models.
Why is it so hard?
AGI assumes several abilities that current systems handle poorly:
- abstract reasoning and long-term planning;
- transfer of knowledge from one domain to a brand-new one;
- efficient learning from few examples (like a child);
- common sense and a causal understanding of the world.
A contested horizon
There is no scientific consensus on when AGI might arrive, nor even on whether it is achievable. Researcher estimates range from a few years to several decades — or never. The debate blends technical, economic and safety concerns: how could the goals of such a system be aligned with human values?
AGI remains a conceptual compass as much as a technical target: it guides research far more than it describes any existing reality.