Superintelligence refers to an intellect that greatly outperforms the best human minds in nearly every field, from scientific creativity to general wisdom and social skills. It is also called artificial superintelligence (ASI). It does not exist and remains hypothetical.
Where it fits
In the common three-level model, superintelligence comes after narrow AI and general AI. The difference between general AI and superintelligence is not only speed. A superintelligent system could reason in ways humans cannot follow.
Origins of the idea
In 1965, mathematician I. J. Good described an “intelligence explosion”: a machine smart enough to design a better machine, which could design a better one still, and so on. The idea reached a wide audience through philosopher Nick Bostrom’s 2014 book Superintelligence: Paths, Dangers, Strategies.
Possible forms
Bostrom described three broad forms:
- Speed superintelligence: thinks like a human, but much faster.
- Collective superintelligence: many smaller intelligences working together better than any human group.
- Quality superintelligence: thinks in fundamentally smarter ways.
Why people worry about it
The main concern is alignment: making sure a system far smarter than us pursues goals that are good for people. A small mistake in its goals could have large effects, and it might be hard to correct once deployed. This is why AI safety research exists as its own field.
The sceptical view
Not everyone takes the idea seriously as a near-term issue. Critics argue intelligence may not scale without limits, that physical and data constraints will slow progress, and that focusing on superintelligence draws attention away from present problems like bias and misuse. Both views are worth knowing.