Steven Byrnes defines AGI as autonomous agents matching human adaptability

Steven Byrnes, in a LessWrong post aimed at a broad audience, redefines what he means by Artificial General Intelligence (AGI). He contrasts current AI—which needs massive specialized R&D to tackle new tasks like driving or robotics—with a future AGI that, like a human, can be given a few hours of practice and then competently perform almost any economic activity. Byrnes emphasizes that a single AGI design, once perfected, can be copied infinitely and autonomously learn everything humans have built from scratch.

Byrnes argues that the real AGI will be not a tool but an agent: it will figure things out, take initiative, make plans, pivot on failures, and even autonomously invent new science and technology. He points out that many people already envision AI as an agent, yet they underestimate the magnitude of the shift. The kind of AGI he foresees could replicate the entire arc of human civilization—from inventing language to building microchips and skyscrapers—without human intervention.

He addresses skepticism by noting that every technology seemed like science fiction before its invention. The critical question is not whether it's sci-fi but whether it's possible—and the existence of human brains proves it is. Just as humans once thought vision was magical but now understand and replicate it with cameras, future engineers will decode the principles behind human general intelligence and build AGI systems that match or exceed our cognitive abilities. Byrnes warns that the impact will be far more disruptive than GDP or employment shifts; it could introduce a new intelligent species that vastly outnumbers and outperforms humans.