5 October 2026 · 2 min read
AGI and superintelligence are two different claims
One says a machine can match us. The other says it can leave us behind. They need different evidence, and they are often confused.
The two terms travel together, and that causes trouble. Artificial general intelligence, AGI, is a claim about parity: a system that can do most intellectual work as well as a capable person. Superintelligence is a claim about a gap: a system that greatly outperforms the best people in virtually every domain. A forecast, a product plan or a law that treats them as one thing will be wrong about at least one of them.
AGI has no agreed definition
Ask five researchers and you get five tests. Some define AGI by breadth: the system handles new kinds of task without being redesigned. Some define it by economics: it can do most work that people are paid to do at a desk. Others ask for the ability to learn a new skill from a handful of examples, the way people do.
The oldest test is Alan Turing's imitation game from 1950, in which a machine passes if a judge cannot tell it from a person in conversation. It measures how well a system talks. It does not measure whether the system can run a laboratory or fix a bridge design. Fluent conversation turned out to be reachable long before general competence.
The evidence is different
A claim of AGI can be checked against people. Give the system and a group of professionals the same work and compare. The method is slow and expensive, but it exists.
A claim of superintelligence cannot be graded the same way, because nobody is qualified to mark the answers. The evidence has to be indirect: results that can be verified after the fact, such as a proof that checks, a prediction that comes true, or a design that works when built.
How wide is the gap?
There are two honest answers. The gap may be short, because a human-level system has advantages no person has. It can run faster, run as many copies, and share everything it learns between them. Speed and scale alone already meet two of the three definitions set out in our essay on what superintelligence means.
The gap may also be long. Matching people in tasks they invented says little about doing better than them at inventing tasks. Reasoning of a higher quality is an open scientific problem with no schedule.
A habit worth keeping
When someone gives a date for AGI, ask which definition they mean. When someone says superintelligence, ask which of the three forms. Most disagreements about timelines turn out to be disagreements about words.