Future Horizons

Super Intelligence by executive order: the new federal name and the word's long history

The executive order swaps "AI" for "SI" in executive-branch non-statutory documents and, for now, keeps the statutory definition. Superintelligence, from I. J. Good to Nick Bostrom and today's labs, names something else. What changes, what

Gabriele Masetti · 2026-09-30

What the order changes

The word "artificial", President Donald Trump told the UN General Assembly on 22 September, "makes it sound fake"; US documents, he said, would move to "the much more accurate term 'super' as opposed to 'artificial'." On 29 September he signed an executive order directing the executive branch to use a new term in place of "artificial intelligence".

Section 2 directs departments and agencies, "to the maximum extent permitted by law", to use "Super Intelligence" and "SI" in official correspondence, public communications, websites, reports, policy documents and other non-statutory documents. The fact sheet adds that agencies will "no longer acknowledge" the old terms.

Section 1 explains why: the capabilities of today's frontier systems "do much more than imitate or automate discrete aspects of human intelligence", and as they improve they "increasingly represent not merely artificial intelligence, but a new era of Super Intelligence."

In legal terms, the order renames a category and, for now, keeps its definition. The name it borrows already had a meaning: for three decades researchers have used "superintelligence" for something far more demanding than the systems it now labels, and the idea is older still. For the foreseeable future the word will carry two meanings in Washington at once. The question is which institutions keep them apart.

What the order keeps

The decisive sentence is the definition. Section 3 says "Super Intelligence" and "SI" mean "the technologies and systems encompassed by the term 'artificial intelligence' as defined in section 9401(3) of title 15, United States Code." That provision, enacted in the National Artificial Intelligence Initiative Act of 2020, covers "a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments."

The legal referent does not move. For purposes of the order, and except where otherwise provided by law, anything that falls within the section 9401(3) definition, today or later, counts as SI, and nothing else does. The statutory definition sets no capability threshold, so it covers a wide range of systems, from established predictive models to frontier language models.

Section 2(b) adds that nothing in that section requires changing "previously issued regulations, Presidential actions, contracts, grants, or other historical documents." Statutes are outside its reach by construction: the order governs non-statutory documents, and Congress alone amends the United States Code.

The open question comes with a deadline. Within 60 days, by 28 November, the Assistant to the President for Science and Technology, currently OSTP director Michael Kratsios, must send the President proposed legislative language for a federal definition of Super Intelligence "that reflects the capabilities described in section 1". The proposal must assess whether that definition should "modify, expand upon, or otherwise supersede" the existing statutory definition of artificial intelligence.

AI vs SI in one minute

What "artificial" meant in 1955

The order's Section 1 recalls that the field was "born in the United States, where American scientists first gave the field its name." The document that gave it that name is dated 31 August 1955 and signed by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon.

We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. — John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon, Dartmouth proposal, 31 August 1955

In that text "artificial" marks where the intelligence would come from: a machine rather than an organism. A note filed with the proposal in McCarthy's archived copy gives the working definition: "the artificial intelligence problem is taken to be that of making a machine behave in ways that would be called intelligent if a human were so behaving."

McCarthy later gave a second, institutional reason for the label. Reviewing Brian Bloomfield's edited volume The Question of Artificial Intelligence, he wrote that "one of the reasons for inventing the term 'artificial intelligence' was to escape association with 'cybernetics'". The name also marked a research programme off from its neighbours.

A word older than the computer

"Superintelligence" predates computing. Merriam-Webster dates its first known use to about 1822. Brigitte Nerlich, writing on the University of Nottingham's Making Science Public blog, traces the Oxford English Dictionary's 1822 citation to a passage on providence that speaks of "the Divine Superintelligence". In her reading, the word's meaning then shifted "from the divine to the spiritual and from the spiritual to the computational or digital".

Who first applied it to machines is not established. The idea has a clearer origin. The statistician I. J. Good set it out in "Speculations Concerning the First Ultraintelligent Machine", a paper in Advances in Computers conventionally cited as 1965, though Good's own footnote says the first draft was finished in April 1963.

Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an "intelligence explosion," and the intelligence of man would be left far behind … — I. J. Good, "Speculations Concerning the First Ultraintelligent Machine", Advances in Computers, vol. 6

Good's word was "ultraintelligent", and his concept already contained both halves of the modern debate: the promise and the problem of control. The same page calls the first such machine "the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control."

From Vinge's singularity to Bostrom's definition

In March 1993, at the VISION-21 symposium co-sponsored by NASA's Lewis Research Center, Vernor Vinge, then at San Diego State University, presented "The Coming Technological Singularity". The abstract opens: "Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended."

Vinge quoted Good at length and used the noun in passing: "Most speculations about superintelligence seem to be based on the weakly superhuman model." The word was already circulating before anyone had pinned it down.

The philosopher Nick Bostrom supplied the working definition. In "How Long Before Superintelligence?", dated 1997 and published in 1998 in the International Journal of Futures Studies, he wrote: "By a 'superintelligence' we mean an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills."

His 2014 book Superintelligence, which Nerlich credits with putting the word "squarely on the map" of AI debate, restated the definition early in its second chapter:

We can tentatively define a superintelligence as any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest. — Nick Bostrom, Superintelligence: Paths, Dangers, Strategies (Oxford University Press, 2014 hardback), ch. 2, p. 22

Two qualifiers carry the weight. "Greatly exceeds" rules out parity with experts, and "virtually all domains" rules out a system that is superhuman at chess, protein folding or code while ordinary elsewhere. By that standard, a system counts only when it outperforms the best people at nearly everything, not when it outperforms most people at something.

How the builders use the word now

Frontier labs use the word for a destination, though not always in Bostrom's full sense. On 22 May 2023 Sam Altman, Greg Brockman and Ilya Sutskever published "Governance of superintelligence" on OpenAI's site, describing their subject as "future AI systems dramatically more capable than even AGI". They wrote that "it's conceivable that within the next ten years, AI systems will exceed expert skill level in most domains", and proposed "something like an IAEA for superintelligence efforts".

Sutskever left OpenAI in May 2024 and on 19 June announced Safe Superintelligence Inc. with Daniel Gross and Daniel Levy. The company's page describes "the world's first straight-shot SSI lab, with one goal and one product: a safe superintelligence."

Meta and Microsoft followed in 2025. In a 30 June memo to employees, Mark Zuckerberg wrote that "developing superintelligence is coming into sight" and created Meta Superintelligence Labs, led by Alexandr Wang as chief AI officer. On 6 November Mustafa Suleyman announced Microsoft's MAI Superintelligence Team, working towards what he called "humanist superintelligence".

The phrasing agrees: "within reach" (the first line of SSI's page), "coming into sight", "working towards". Each company treats superintelligence as something not yet built, ahead of current products: a gap the 28 November proposal may need to address.

Two senses, six days apart

On 23 September Sen. Bernie Sanders and Rep. Greg Casar introduced the Ban Artificial Superintelligence Act, which their announcement summarises as covering, among other things, "an AI that exceeds human cognitive performance and capabilities across most domains". That is close to Bostrom's sense: a capability threshold. On 29 September the executive order attached "Super Intelligence" to every system inside the 2020 statutory definition of AI.

The spelling differs as well. The order writes two capitalised words, "Super Intelligence"; the research literature, the bill and the companies write one, "superintelligence". Federal documents now use the phrase for today's technology; research papers are likely to keep using it for a level that, by the builders' own account, has not yet been reached.

What is likely to happen next

Federal documents

Federal prose is changing quickly; the federal record will not. On the day of the UN speech, according to an email obtained by the Associated Press, a deputy assistant secretary in the State Department's Bureau of International Organization Affairs told staff to "change all references" to Super Intelligence.

The order's own fact sheet already describes the July 2025 "America's AI Action Plan" as a list of actions "to win the SI race", keeping the document's title and changing the prose around it. Because Section 2(b) exempts earlier documents, expect a mixed federal corpus for years.

The 28 November definition

The proposal will show whether the change stays one of vocabulary. The closest precedent is Executive Order 14347 of 5 September 2025, which allowed "Department of War" as a secondary title in "official correspondence, public communications, ceremonial contexts, and non-statutory documents", stated that statutory references "shall remain controlling until changed subsequently by the law", and asked for legislative recommendations within 60 days.

The Super Intelligence order follows the same architecture, down to a closely matching list of document types and the 60-day clock. If the 28 November proposal keeps the section 9401(3) definition, the change stays lexical; if it proposes a capability-based definition, Congress would be asked to create a legal category distinct from AI. The order's own wording, "the capabilities described in section 1", points towards the second.

Standards bodies

Standards bodies face a practical collision. NIST's Center for AI Standards and Innovation, renamed from the US AI Safety Institute on 3 June 2025, carries "AI" in its title, and the order says nothing about the names of offices. NIST also publishes Special Publication 330, The International System of Units (SI), the measurement system that the International Bureau of Weights and Measures presents "with the international abbreviation SI". Technical documents will likely spell out "Super Intelligence" to avoid the clash.

Other governments

So far, other governments show little sign of following in their own texts. The White House fact sheet of 25 September on President Xi Jinping's state visit says the two leaders "agreed to use the term 'super intelligence' rather than 'artificial intelligence'" and announces a "U.S.-China Super Intelligence (SI) Dialogue". China's Ministry of Foreign Affairs, listing the same deliverables on 26 September, calls it "the China-U.S. AI Dialogue".

The EU AI Act, Regulation (EU) 2024/1689, is built on the defined term "AI system" in Article 3(1), and changing a defined term there takes the EU's ordinary legislative procedure.

Search

Search engines follow usage. When the Gulf of Mexico was renamed in federal usage, Google changed its maps for US users only after the Geographic Names Information System was updated, citing a "longstanding practice of applying name changes when they have been updated in official government sources."

No official register exists for a common noun, so there is nothing equivalent to update. Rankings track the words people type and publish, and "SI" already competes with the metric system and Slovenia's .si domain. How quickly "SI" comes to signal federal technology policy in search results will depend on how widely the term is adopted outside government.

Industry

For now, industry still uses the word mainly for what it has not built. The labs named above already use "superintelligence" as a stated goal, so the term fits their announcements. Applying it to released products would depart from that framing. Model documentation and benchmarks, which the order does not reach, will likely keep the engineers' vocabulary.

Two dates will show whether the meanings stay apart: the proposed definition, due 28 November, and the next session of the dialogue that Washington calls SI and Beijing calls AI, due by November 2026. Until then the word does two jobs. In federal prose it names the systems already in use; in the research literature it still names Good's machine, "the last invention that man need ever make".