Society, Environment & Big Picture
AGI (Artificial General Intelligence)
The hypothetical AI that can do any mental work at human level — a matter of definition, an article of faith and a marketing instrument at once. Whether five years or fifty: for today’s decisions AGI is usually the wrong reference point. The relevant question is more modest and harder: what can this year’s systems do reliably — and what damage do they already cause?
An example
The term has no agreed definition, and that is not a detail. For some, AGI is reached when a system performs most economically valuable tasks better than a human. For others, when it learns like a human — from a few examples, with transfer to new things. For others still it is a contractual term: in the agreement between OpenAI and Microsoft, AGI was defined via a profit threshold.
A closer look
There is no recognized definition and no recognized test. Proposals range from economic thresholds — a certain share of human labor — to lists of capabilities. Without a definition, timeline forecasts for it cannot be checked.
Why it matters
For business decisions the term is unusable, for a precise reason: it does not describe a verifiable state. Whoever justifies an investment with “AGI is coming soon” justifies it with nothing. What remains useful is the observation that capabilities arrive in leaps and unevenly — so plan against what demonstrably works today, and set up the architecture so that a better model can be dropped in without anything having to be rebuilt. That is the only preparation that is right in every scenario.
Commonly misunderstood
The debate ties up attention that more concrete questions need. Whether a system is reliable in your workflow does not depend on what it is called.