AGI: the word that confuses more than it explains

“We’ll reach AGI by 2027.” “By 2030.” “We’re already there.” “We’ll never get there with this technology.” If you follow AI you’ll have heard all of these, sometimes from the same person a few months apart.

The problem isn’t who’s right. It’s that they’re not talking about the same thing.

One word, many definitions

AGI stands for Artificial General Intelligence. But everyone defines it their own way:

a system that can do any intellectual task a person can;
a system that can do most economically valuable work better than people;
a system able to learn new things by itself, like a human, rather than being trained;
a system able to do scientific research, including research on AI itself, autonomously.

By the second definition, some argue we’re nearly there. By the third, we’re still far away. And definitions shift for non-scientific reasons too: contracts, fundraising, media attention.

Why the label isn’t very useful

Arguing about “when AGI arrives” leads to two opposite mistakes.

The first is underestimating the present: since “it’s not AGI yet”, people ignore that today’s models already write code for hours, analyse contracts, research dozens of sources. Many jobs are changing now, with no magic threshold needed.

The second is overestimating the future: imagining a precise day when everything changes. A gradual, uneven transition is more likely, with systems brilliant at some things and surprisingly fragile at others (models capable of advanced mathematical proofs can still invent a court ruling).

Look at tasks, not labels

A more useful way to think about the future is to ask, task by task:

Does AI already do it well? (summarising, translating, drafting, searching)
Does it do it well with supervision? (code, analysis, research)
Does it do it autonomously for hours? (agents completing projects)
Does it do it better than top experts? (for now, a few very specific cases)

Every year many tasks climb a rung. That ladder, not the AGI label, tells you what will change in your work.

And superintelligence?

Some researchers fear systems far more capable than us and hard to control. It’s a serious topic (I discuss it in the opinion on how worried to be), and it deserves research and rules. But here too the word risks hiding the concrete questions: who controls these systems, with which checks, within which limits.

In one line

AGI is a useful word for headlines and an awkward one for reasoning. To understand AI’s future, ask what it can do today, what it will do tomorrow, and who decides.

Sources

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.