Superintelligence, explained.
What artificial superintelligence means, how it differs from AGI, and how to read the news without confusing predictions with capabilities.
What is superintelligence?
Artificial superintelligence generally describes AI that would exceed human capabilities across a broad range of intellectually demanding work. It is a concept about capability, not a synonym for a fast chatbot or a product with an impressive demo. OpenAI's governance discussion considers what such systems could mean for oversight.
AI, AGI, and superintelligence
AI is the broad field. Artificial general intelligence (AGI) refers to broadly capable systems, though organizations use different definitions. Superintelligence points beyond human-level capability. A claim about one benchmark does not establish performance across every important task.
Our name reflects the questions surrounding increasingly capable intelligence. We still use precise technical terms in individual stories rather than treating every release as superintelligence.
What is the singularity?
The singularity is a proposed period of accelerating technological change that becomes difficult to predict, often associated with AI improving the process of creating future AI. It is a forecast or interpretation rather than a single agreed test. Read our breakdown of Jack Clark's argument for a concrete example of how the idea enters policy discussions.
Speed and capability answer different questions
Generation speed tells you how quickly text appears under particular conditions. Capability concerns what the system can do. Reliability concerns how consistently it does it correctly. These dimensions can improve independently. Mercury 3 illustrates why speed alone does not measure an entire workflow.
Why governance matters
More capable systems raise questions about access, accountability, and who can require changes. An oversight proposal is different from an agreement that has been adopted. Our AI safety explainer examines how different institutional designs distribute authority.
How to follow the field
Ask four questions: what changed, what supports it, what is the important limitation, and what does it mean for you? Look for primary sources, identifiable task boundaries, realistic access information, and a clear distinction between a reported result and a prediction.
Continue with AI research, industrial applications, or how companies propose sharing the upside. For our approach to sources, see how we report.