Definition
In AI research, superintelligence means intelligence substantially exceeding the strongest human cognitive capabilities across a broad range of domains. Artificial superintelligence, often abbreviated ASI, refers to an artificial system with that capability.
The comparison requires a baseline. Surpassing one person on a task, outperforming a benchmark, and exceeding the capability of a large human organization are different claims. A technical definition should state the domains, comparison group, tools, time, learning conditions, and reliability requirements.
Origin and research usage
Nick Bostrom's essay first published in 1998 defines superintelligence through broad superiority to the best human cognitive abilities, including scientific, practical, and social domains. This is a documented definition; the essay does not establish the first coinage of the word.
Google DeepMind's 2026 From AGI to ASI research adopts a comparison with large human organizations. The difference illustrates why a bare claim of superintelligence is incomplete without a threshold. Consciousness is a separate question from measured cognitive capability.
U.S. federal usage from September 2026
Executive Order 14434, signed September 29, 2026, directs executive agencies to substitute Super Intelligence and SI for Artificial Intelligence and AI in official communications, websites, reports, policy documents, and other nonstatutory documents, to the extent permitted by law.
Section 3(a) assigns SI the scope of the existing statutory AI definition in 15 U.S.C. 9401(3). It does not add a superhuman capability threshold. The order also does not require rewriting existing regulations, contracts, or historical documents.
America.gov launched the same day. GSA describes enhanced search across government websites, and Google identifies Gemini as supporting the service. That deployment is an example of AI-enabled public information access. Neither its launch nor the federal naming directive establishes superintelligence in the technical sense.
Why it is contested
The cited research uses different human baselines: Bostrom compares capability with the best individual humans, while DeepMind's 2026 paper compares it with large organizations. Those thresholds are not interchangeable. Government communications add another meaning: SI can denote systems covered by the existing AI definition, without implying broadly superhuman performance.
Distinguish it from nearby terms
- Artificial general intelligence concerns broad, transferable competence; some definitions place it near human capability. Superintelligence adds a stronger superiority claim.
- Narrow AI can outperform humans in a bounded domain without meeting a broad superintelligence threshold.
- Recursive self-improvement describes a process by which a system improves itself. It does not establish that the resulting capability is superintelligent.
- SI in a U.S. executive communication may be the institutional replacement for AI. Read the document's definition before inferring capability.
Check your understanding
A federal website describes its search assistant as SI. What would you need to know before interpreting that as a claim that the assistant exceeds the best human capabilities across broad domains?