Definition
Artificial superintelligence (ASI) refers to a hypothetical AI system whose intelligence comprehensively and substantially exceeds that of the smartest humans, outperforming them in virtually every domain. “Super” points past matching humans to clearly surpassing them.
How it works
There is no accepted path to ASI. It is usually discussed as a hypothetical stage after AGI: a system able to improve itself might grow beyond human understanding and control. Such arguments rest on assumptions about how intelligence scales and remain thought experiments.
Usage
ASI comes up mainly in discussions of long-term AI risk, safety research, and alignment research, where it names an extreme scenario to guard against — not any existing technology.
Related and confusable terms
The difference between ASI and artificial general intelligence is the capability ceiling: AGI is benchmarked against human level, ASI assumes beyond it. Both are hypothetical stages, distinct from artificial intelligence as a description of real-world technology.