artificial general intelligence
/ ar-tuh-FISH-ul JEN-ur-ul in-TEL-uh-juns /
Artificial general intelligence is the hypothetical machine that could do roughly anything a capable adult human can do with their mind: learn a brand-new skill from a few examples, reason through an unfamiliar problem, plan, use common sense, and carry knowledge from one area into another. Where narrow AI is a specialist locked in one lane, AGI would be a generalist — able to switch lanes, invent new lanes, and figure out tasks nobody trained it for.
The key word is general. A human child who learns to ride a bike can later transfer balance and intuition to a skateboard; a person who masters cooking can apply patience and planning to a completely different craft. AGI names that flexible, transferable competence applied across the full breadth of human activity. It does not require consciousness or feelings — what defines it is the breadth and adaptability of the capability, not any inner experience.
Be candid about its status: AGI does not exist, and there is no agreed recipe for building it or even a precise, measurable definition of when we'd have crossed the line. Serious researchers disagree wildly about whether it is years, decades, or centuries away, or whether scaling up today's methods leads there at all. Much public talk treats AGI as imminent or inevitable; the honest answer is that it remains an open, unresolved question. Today's most powerful systems are still narrow, however broad they may feel.
A useful litmus test people imagine for AGI: drop the system into an ordinary job it was never trained for — say, running a small cafe for a week — and see if it can learn the ropes, handle surprises, and improvise like a new human hire would. No system today comes close to that kind of open-ended adaptability.
AGI is defined by open-ended, transferable competence — not yet achieved by anything that exists.
AGI is a goal and a hypothesis, not a product you can buy. Beware claims that it has 'arrived' or is one breakthrough away — there isn't even a single agreed definition or test, and impressive narrow performance is regularly mistaken for general intelligence.