Few terms generate more excitement and confusion than AGI — artificial general intelligence. It's invoked constantly, in headlines and boardrooms, yet ask ten experts to define it and you'll get ten answers. Understanding why it resists definition cuts through a lot of hype.
The rough idea
Loosely, AGI means AI with human-level general intelligence — able to learn and perform across the full range of tasks a person can, rather than being narrow (great at chess, useless at everything else). The dream is a system that generalizes like a human mind.
Why it defies definition
The trouble is that "human-level general intelligence" isn't one thing. Do we mean matching humans at most economically valuable tasks? At any task? Including physical dexterity, social intelligence, genuine understanding, consciousness? Different definitions imply wildly different bars — and different arrival dates. Some definitions arguably describe today's best models; others describe something still far off.
"Has AGI arrived?" is unanswerable not because we lack data, but because we haven't agreed what we're measuring.
Why the fuzziness matters
The vagueness gets exploited. "AGI is coming" can mean anything from "models keep getting more general" (uncontroversial) to "human-level minds are imminent" (a huge claim). Companies and commentators lean on the ambiguity. When you hear AGI claims, the useful move is to ask: by which definition, and by what measurable criteria?
The pragmatic stance
Rather than debate an undefined finish line, it's more useful to track concrete capabilities: what can models do now, how reliably, at what cost, and where do they still fail? That grounds the conversation in reality. AGI may remain a moving, contested goalpost — but capabilities are measurable, and that's where the honest discussion lives.
