There's a version of AI that can do anything a person can do. It can pick up a new skill it's never been trained on, reason through a problem it's never seen, and apply what it learned in one domain to a completely different one. That version doesn't exist. Everything you've used, everything in this learning center, every AI product on the market today — all of it is something categorically different.
It's all narrow AI.
The word "narrow" isn't a criticism. A narrow AI system can be astonishingly capable within its domain. A model that writes code doesn't also diagnose diseases. A model that recognizes faces doesn't also translate languages. Each one is trained for specific tasks and operates within specific boundaries it can't cross on its own. That's not a limitation to apologize for; it's just an accurate description of what these systems are.
Artificial general intelligence (AGI) is the hypothetical counterpart: a system that can match or exceed human cognitive ability across virtually any task, transfer knowledge between domains, and solve novel problems without task-specific training. The term was popularized in 2007 by AI researcher Ben Goertzel, partly to give a name to the long-standing goal, and partly to clarify that what the field was actually building in the meantime was something else entirely.
AGI is a stated goal of several major AI labs. It's also an unsolved problem — and there isn't even consensus on what would count as achieving it. A 2023 Google DeepMind paper proposed a five-level framework for AGI capability and placed current large language models at the lowest rung. Impressive as those models are, they remain narrow: trained on language, operating within language, unable to autonomously acquire genuinely new capabilities the way a person can.
This distinction matters more than it might seem. When people assume current AI is approaching general intelligence, two things tend to go wrong. They over-trust it, treating its outputs as more reasoned and more aware than they are. And they fear it in
