This week’s Binding makes one argument from the record. Three writers to read alongside it end the issue.
On Sept. 22, at the United Nations General Assembly, Trump gave his reason for retiring a word. “The use of the word artificial makes intelligence fake,” he said, as UN News reported it. “It is not fake. It’s actually amazing.”
That is an argument about what a thing is and what it can do. It is an old kind of argument. Scholars who study the social construction of technology have mapped arguments like it since the 1980s: who argues, what the machine means to each side, and how the argument ends. This Binding borrows their method.
I made a version of this argument in March, before the order. In “The Demo Is Not the Definition,” I wrote that the name “Artificial Intelligence” matters because it “front-loads an answer to a question nobody has agreed on,” and that “every person in the room is watching the same demo and seeing a different thing.” Trump found the same name wrong from the other side. I thought “intelligence” claimed too much. He argues “artificial” claims too little.
This Binding also holds a line of my own. There are arguments about what something is or what it is capable of. Science creates the properties that need to be met.
Wiebe Bijker, one of the method’s founders, put the hard question to his own field in 2017. “Can we do without science?” His answer came in the next line: “No, we cannot do without science.” (Bijker 2017, p. 318)
This week the new word reached a proposed rule. No government document in this record identifies a level a system must reach to earn it.
From a speech to a rule
The word took three steps in 34 days, and each step left a page.
On Sept. 22, at the UN, Trump spoke of “the Artificial Intelligence being spoken of so much now,” then said it is “hereinafter officially called ‘Super Intelligence.’” That wording comes from the White House’s own excerpt of the speech.
On Sept. 29, he signed Executive Order 14434 (91 FR 63129). Sec. 2(a) tells agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI,” “to the maximum extent permitted by law,” in “official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents within the executive branch.” Sec. 1 gives the reason. The new term “more appropriately captures the promise, potential, and rapidly advancing capabilities of these technologies.”
Then Sec. 3(a) defines the new word with the old one. For purposes of the order, “SI” means “the technologies and systems encompassed by the term ‘artificial intelligence’ as defined in section 9401(3) of title 15, United States Code.” That section comes from the National Artificial Intelligence Initiative Act of 2020. Its definition reads: “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments.”
Read that sentence for anything super. The system it describes serves goals people set. It sets no bar for how well the system performs.
On Oct. 26, Treasury published a 38-page proposed rule on U.S. investment in China and other countries of concern (91 FR 67396). On page 67402, Treasury says the most notable change among the sectors it already covers “involves what was defined as AI systems in the Part 850 Regulations and is referred to as ‘super intelligence systems’ (SI systems) in the Proposed Rule.” One sentence holds both names.
See for yourself. Open the order and read Sec. 1 and Sec. 3(a) side by side. Write down the capability Sec. 1 claims. Then look for it in the definition Sec. 3(a) adopts.
What the proposal does
Treasury’s stated basis for the new word sits in footnote 1, on page 67406. The footnote cites Section 2 of the order, which it says “requires executive departments and agencies to use ‘Super Intelligence’ and ‘SI’ in place of ‘Artificial Intelligence’ and ‘AI’ in certain documents.”
Treasury keeps both words on purpose, and says so on the same page: “The narrative description of the Proposed Rule below refers to ‘artificial intelligence system’ or ‘AI system’ when discussing the applicable requirements in the Part 850 Regulations or Title VIII and refers to SI system and the abbreviation ‘SI’ when discussing the Proposed Rule.”
Title VIII is the law Treasury carries out. Congress enacted it on Dec. 18, 2025, through the COINS Act, inside the defense bill for fiscal 2026 (67396). By Treasury’s own account, the law names “AI systems” as a technology area and “does not further define” the term (67406).
Treasury calls its new definition “generally consistent with the existing definition of an ‘AI system’ found in the Part 850 Regulations, but with some modifications” (67406). The modifications reach systems that can “generate content,” plus “a harness for agentic capability,” “an embodied SI system,” and “robotic or other autonomous or unmanned systems” that run on one.
The proposal also changes what the rule does. Today’s rules, in Part 850, sort these deals by the system’s size and purpose. Deals tied to the largest systems, trained with more than 10²⁵ computational operations, or to systems built only for, or intended for, military, intelligence or mass-surveillance use, are prohibited. Deals tied to systems above a lower line, 10²³ operations, or meant for uses such as cybersecurity, must be reported to Treasury. The rest fall outside the program.
The proposal drops those lines. It “lists the development of SI systems within the definition of prohibited technology, and not within notifiable technology” (67402). Under the proposal, a U.S. person could not make certain investments in a covered foreign person, in a country of concern, that develops any SI system. Treasury explains why the size test goes. “Over the past two years, SI models have achieved notable performance gains through additional means other than the quantity of training compute. Therefore, no single measure, benchmark, or threshold is currently sufficient to distinguish advanced models” (67404). It gives national-security reasons for the wider ban. None of them cites the order or the change of name.
What the lens shows
Bijker once summed up a biogas study in a single sentence: “We identified different relevant social groups and mapped what biogas meant for each of them” (p. 316). Put “SI” where biogas sits, and this week’s record starts the map.
To the White House, the word marks a new era. The order’s title is “Inaugurating the Era of Super Intelligence.”
To Treasury, it is the headword of a proposed rule, with the old word kept for the statute and the rules on the books.
To Congress, as Treasury reads the law, the sector is still “AI systems” (67406).
To the Associated Press, it is “Trump’s favored term for AI” (AP, Oct. 22). To the law firm Freshfields, writing to clients on Oct. 26, it is “AI (now ‘SI’)” (Freshfields).
To the Coast Guard, on Oct. 26, it names a planned research agreement: “Super Intelligence (SI) Situational Awareness Tools” (91 FR 67324). One sentence in the same notice still describes “artificial intelligence enabled situational awareness tools.”
To the National Center for Education Statistics, on Oct. 29, the old word is past tense. Its request for information asks about “future Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI) applications and uses” (91 FR 68120).
Trevor Pinch and Bijker developed two terms for this pattern in 1984. The first, interpretative flexibility, means one artifact carries several meanings at once. The second, the closure mechanism, describes how such an argument ends (Pinch and Bijker 1984, abstract). In plain words, closure comes when the groups stop asking what the thing is.
Inside the executive branch, one group can try to close the question with a signature. Bijker later extended the method to take in power (p. 326). “Some groups can push their definition of the problem more than others,” he writes (p. 327). An executive order binds the agencies that work under it. Newsrooms, courts, companies and readers decide for themselves.
The closure question has a second half, and the method can’t answer it alone. Suppose every agency, newsroom and company said “SI” tomorrow. The systems would do what they did the day before. The word would be settled, and someone would still have to show the capability it claims.
What a name has to meet
Researchers have proposed properties for this word. Some have said what a test would take.
In 1998 the philosopher Nick Bostrom defined a superintelligence as “an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills” (Bostrom). That definition sets a system against the best of us, field by field.
In 2024, Meredith Ringel Morris and seven co-authors turned that kind of claim into levels a system can be scored on, in a paper at the International Conference on Machine Learning (Morris et al. 2024). Their first principle reads “Focus on Capabilities, not Processes.” Their top level, “Superhuman,” describes a system that “outperforms 100% of humans.” A general system at that level they call “Artificial Superintelligence (ASI),” and their table marks it “not yet achieved.” Then they say what sorting systems takes: “Unambiguous classification of AI systems will require a standardized benchmark of tasks” (arXiv version, Principle 1 and Table 1). The paper stops short of proposing that benchmark.
By properties I mean a stated capability and a test other people can run to see whether a system has it.
Bijker doesn’t exempt science from his method. “Science is socially constructed,” he writes, “and it has special value in society because of that construction process” (p. 319). Scientists argue too, over tests like that one. Bijker counts the argument as part of the process: “controversy between scientists is normal,” he writes, and “there is no reason to lose trust in science” (p. 319).
In a 2019 paper, P. M. Krafft and four colleagues compared how AI researchers and policymakers define the field (Krafft et al., abstract). Researchers “favor definitions of AI that emphasize technical functionality.” Policymakers “instead use definitions that compare systems to human thinking and behavior.” The authors pointed out that “definitions that emphasize human-like capabilities are most applicable to hypothetical future technologies.”
The order holds both styles in the same document. Its name compares the machine to human intelligence and claims more. Its legal definition, in Sec. 3(a), is functional and sets no level.
Treasury’s proposal would take the size measures out. In the same document that would write “SI system” into a federal rule, Treasury would drop the compute lines that sorted systems by size. Its stated reasons are security, simpler compliance and the measurement finding quoted above. Under the proposal, the ban would reach the development of any system in a category called “super,” with no capability measure to sort one from another.
The order itself asks for a new definition. Sec. 3(b) gives the President’s science adviser 60 days to submit “proposed legislative language to establish a Federal definition of ‘Super Intelligence’ and ‘SI’ that reflects the capabilities described in section 1 of this order.” That language is due to the President on Saturday, Nov. 28.
The new name reached a proposed federal rule in 34 days, and the test for what it claims has not arrived. Every Monday I take one word an institution bent and show the page it bent on. Subscribe and the next one is in your inbox, free.
What is settled
The strongest case for Treasury’s text runs like this, and most of it holds.
Treasury followed the order and went no further. It cites the order in a footnote. It says it keeps “AI system” when it describes the statute or today’s rules (67406). It leaves Part 850, with its “AI system” definition, in force for deals made from Jan. 2, 2025, until a final rule takes effect (67398).
The label keeps the old legal meaning. The order defines “SI” with the 2020 definition, and Treasury calls its own definition “generally consistent” with the old one (67406).
The wider ban stands on stated findings. Treasury says it acted “in consultation with interagency partners and the Executive Office of the President” and cites “the rapid advances in super intelligence capabilities of countries of concern” (67402). None of those reasons cites the order or the change of name. Its reason for dropping the size test rests on measurement. Models have gained performance by means other than training compute (67404).
Congress ordered the rework. The COINS Act told Treasury to update or replace the old rules (67397). And Treasury asks for comment on both of its choices, the ban (67404) and the definition (67406–67407).
The case runs thin in two places. The definition changes along with the label. It names content generators, agent tooling and embodied systems outright. And the order’s Sec. 1 goes past housekeeping. It sets as policy that the executive branch, “to the maximum extent permitted by law,” “will not acknowledge the usage of ‘Artificial Intelligence’ and ‘AI’ in any applicable setting.”
What I am not saying
I am not saying the name caused the ban. The proposal renames the sector and widens the ban in the same document. No document I read ties one to the other.
I am not saying Treasury only changed a label. The same proposal would end the reporting tier for these systems and drop the compute tests.
I am not saying anyone would be barred from building an AI system. The proposal would bar certain U.S. investments in covered foreign persons, in China and other countries of concern, that develop one.
I am not saying the order forced the rename. Sec. 2(b) reads: “Nothing in this section requires the alteration of previously issued regulations, Presidential actions, contracts, grants, or other historical documents.” Whether Sec. 2(a) reaches a new proposed rule is an open question. Treasury’s footnote cites Section 2 as the basis for the new word.
I am not saying a superintelligence exists. In the order, “SI” names today’s systems by the 2020 definition, and the proposal’s definition stays close to today’s.
And the method I borrowed maps how a meaning gets made. It shows no one’s intent.
What I am saying
My bottom line is that we are trying to slap a cookie label on broccoli. We are not dealing with Super Intelligence yet. I hardly think we are dealing with Artificial Intelligence. I think we have something far different that has not been properly labeled.
The record backs the label and the “yet.” The order defines “SI” with the 2020 definition, which sets no level, and the researchers who defined the top level marked “Artificial Superintelligence (ASI)” as “not yet achieved” in their 2024 paper. The rest is my judgment, the one I began in March, and it stands on the line I set down at the top. There are arguments about what something is or what it is capable of. Science creates the properties that need to be met.
The order settles the word for the executive branch, as far as the law allows. It has not said what level a system must reach to earn the word “super.” The order gives the science adviser until Nov. 28 to propose a definition built on capabilities. Until a definition names the capability and the test, “super” in a federal rule is a label with nothing to measure it against.
Two dates let you check whether that changes.
By Nov. 28, the President’s science adviser owes him proposed legislative language for a federal definition. If that language becomes public, read it for one thing. Does it name a capability a system must show, and does it say who tests for it?
By Dec. 7, anyone can comment on Treasury’s proposal at regulations.gov, docket TREAS-DO-2026-0595. Treasury “welcomes comment on the proposed definition of SI system” (67406–67407), and it says it will make every comment public. As of Oct. 28, the docket showed none posted. The window closes at 11:59 PM Eastern on Dec. 7.
See for yourself. Open page 67406 of the proposal. Find the sentence where Treasury tells you when it writes “AI” and when it writes “SI.” Then ask what a system would have to do to deserve the second word.
If your work runs on these systems, hit reply and tell me: what would you test before you let anyone call one super?
Read alongside
Three writers on Substack have worked the same ground, and each piece is free to read.
Deri Draper-Amason wrote “Calling It ‘Super’ Doesn’t Make It Safe” the day after the UN speech (Sept. 23). “Technology does not name itself,” she writes. “People name it, and the names we choose carry expectations, budgets, and risk tolerances with them.” Her advice to leaders matches the test I am asking for: “Define capability by evidence, not by label. Specify what the system does, under what conditions, with what error rates.”
Charley Johnson’s “The Inevitability Machine” (Untangled, June 27) is a field guide to the vocabulary of AI. His entry on “artificial intelligence” says the term “equates what the machine does with what a mind does,” and that “the word carries its premise forward like a stowaway.” He also offers a rule worth keeping: “Name the chooser and the future becomes something you can argue with again.”
Albert Tu’s “The Determinism Tax” (The Common World, Oct. 7) responds to an essay by Dario Amodei, the chief executive of Anthropic. Tu argues that the people who fear AI most and the people who cheer it loudest share four beliefs that make it look inevitable. He answers with history. New technologies “have had contingent paths,” he writes. “They were not predetermined, but rather products of actors in contest.”
I keep a list of writers who read AI as something people build, name and govern. It’s called AI Is a Choice.
Voters, Election Day is Nov. 3, and it falls inside Treasury’s comment window. See every race on your ballot at vote411.org.
I write What Time Binds with Claude. Every claim carries its page, and I approve every line. How I work: How the bread gets made.
Jerry W. Washington, Ed.D., is a Marine Corps veteran of twenty-three years, a combat engineer by trade, an instructor at UC Irvine’s Division of Continuing Education, and the writer of What Time Binds, a record of how institutions bend words and how the page holds them to account.
I advise school districts, community colleges, workforce boards, and veteran-service organizations when the words move faster than the coordination. AI adoption is the most common case. See the five ways to work together at jerrywwashington.com, or hit reply if you would like to talk.