# Superintelligence A hypothetical AI system that surpasses the best human minds in every cognitive domain — science, strategy, creativity, social intelligence, general reasoning — by a significant margin. Popularized by [[Nick Bostrom]]'s 2014 book [[Superintelligence (Book)]], the concept is the theoretical ceiling of AI development and the scenario that motivates the [[Existential Risk]] discourse. ## The rhetorical function "Superintelligence" does enormous work in AI discourse despite describing something that doesn't exist and may never exist: **It maximizes stakes.** If the endpoint of AI development is a system smarter than all humans combined, then every decision about AI governance, safety, and development is a decision about the future of the species. This framing elevates AI lab leadership to world-historical importance and makes AI policy the most consequential policy domain in existence. **It justifies the present.** The prospect of superintelligence is used to justify present-day resource allocation (trillions in compute infrastructure), present-day deference to AI labs ("they're managing the most important technology in history"), and present-day risk tolerance ("compared to the benefits of superintelligence, current harms are manageable"). **It divides the discourse.** Whether superintelligence is imminent, distant, or impossible is the deepest fault line in AI discourse. [[AI Doomerism]] treat it as imminent and existentially dangerous. Accelerationists (e/acc) treat it as imminent and desirable. The [[Stochastic Parrot]] camp treats it as a distraction from present harms. Each position flows from a prior belief about whether superintelligence is real. OpenAI's "Superalignment" team (established 2023 to solve alignment for superintelligent systems, partially dissolved 2024) embodied the concept's organizational function: it positioned OpenAI as simultaneously building the most dangerous technology in history and heroically working to make it safe. ## The federal appropriation of the term (2026) On September 29, 2026, [[Executive Order 14434]] directed every executive department and agency to write "Super Intelligence" and "SI" in place of "artificial intelligence" and "AI" in all non-statutory usage. The order defines the term by reference to the existing statutory definition of artificial intelligence at 15 U.S.C. § 9401(3) — that is, it applies the word to **today's deployed systems**, not to the hypothetical ceiling this page describes. That is a collision worth naming. In the safety literature the word's entire force comes from marking something that does not yet exist: it is the counterfactual against which present systems are measured. An official vocabulary that calls current large models "super intelligence" removes the gap the word was built to hold open. The rhetorical functions catalogued above do not disappear — maximized stakes, justification of present allocation — but they transfer from a speculative future to an installed base, which is the opposite of what the existential-risk tradition intended the term to do. It also strands the discourse. If the government's word for a chatbot is "super intelligence," safety researchers lose the term they used for the thing that would actually be dangerous, and the two senses now have to be disambiguated every time either is used. ## See Also - [[AGI]] — the intermediate milestone - [[The Singularity]] — the post-superintelligence scenario - [[Existential Risk]] — the risk framework for superintelligence - [[AI Doomerism]] — the anxiety superintelligence generates - [[Executive Order 14434]] — the 2026 order making "Super Intelligence" the federal government's word for ordinary AI