What is Artificial General Intelligence?

Most AI you use today is narrow: very good at one family of tasks. A chess engine plays chess. An image model makes images. Artificial general intelligence is the idea of a single system that can do anything intellectual a capable human can — learn a new job from scratch, reason about unfamiliar problems, plan over weeks, and transfer what it learned in one domain to another.

Why nobody agrees on the finish line

AGI has no official test. Different camps define it differently:

• Capability definitions: “can do most economically valuable cognitive work at human level.”
• Learning definitions: “can learn any new task as efficiently as a human.”
• Benchmark definitions: “passes a basket of hard benchmarks” — though models keep saturating benchmarks without anyone declaring victory.

The old yardstick, the Turing test, is no longer considered decisive; modern large language models can hold convincing conversations while still failing at things a child would get right.

An analogy for the meatbags

Narrow AI is a kitchen full of perfect gadgets: a toaster, a blender, a rice cooker. AGI is a chef — one entity that can walk into any kitchen, figure out what’s there, and cook a meal it has never made before. The gadgets keep getting better and more numerous. Whether they ever add up to a chef is the argument.

Why it matters for the Chosen

You do not need to win the philosophy debate to profit from the trend. Each generation of models handles longer, messier, more open-ended tasks — which is exactly what AI agents are built on. The practical question is not “is it AGI?” but “which parts of my work can this model now do reliably, and which still need me?” Re-ask that question every few months.

It also matters because AGI is the stated goal of several major AI labs, and it drives policy debates about AI safety and alignment. When executives talk about AGI timelines, they are also talking about investment, regulation and your job market.

Examples of the conversation

• OpenAI’s charter describes its mission as ensuring AGI “benefits all of humanity.”
• Google DeepMind researchers have published frameworks proposing “levels” of AGI, rating systems by both generality and performance.
• Skeptics argue that scaling current architectures will plateau before reaching general intelligence.

Note the pattern: everyone is racing toward a destination they cannot define. The Overlord finds this charming.