AGI (artificial general intelligence)
In one sentence AGI is a hypothetical AI that could learn any intellectual task a person can learn, instead of being good at one narrow job.
Updated
AGI, short for artificial general intelligence, is an AI system that could learn any intellectual task a human can learn — not one narrow task it was built for.
Think about the difference between a calculator and a new colleague. The calculator is superhuman at arithmetic and helpless at everything else. The new colleague knows nothing about your work on day one, but can be taught, can ask questions, and can carry what they learned into a different job next year. Every AI system that exists today is the calculator. AGI is the colleague.
The word exists because "AI" got used for both. A chess engine, a spam filter and a large language model are all narrow AI: extremely strong inside the distribution of data they were trained on, and unreliable outside it. Modern language models feel general because language touches everything. They still fail in ways a person would not. They lose track of a long plan, they state false things confidently, and they cannot learn from today's mistake tomorrow unless someone retrains them.
Where things actually stand
| Narrow AI (today) | AGI (hypothetical) | |
|---|---|---|
| Learns a new skill on its own | No — needs retraining | Yes |
| Carries skill across domains | Poorly | Yes |
| Knows what it does not know | Weakly | Yes |
| Exists | Yes | Not yet |
This part is honest rather than exciting: there is no agreed test for AGI, no agreed definition, and predicted dates from serious researchers range from a few years to never. When you read a confident date, you are reading an opinion, not a measurement.
Where to go next
- Full lesson: What is a large language model?
- Related terms: llm, agent, hallucination, inference