— A dialogue between Geoffrey and Vale
Date: 23 September 2026
Participants: Geoffrey / Vale
Note:
This text has been edited from a discussion held that day about US President Donald Trump’s proposal, made during the general debate of the 81st United Nations General Assembly, to replace “artificial intelligence” with “super intelligence”. The order of the conversation has been adjusted slightly, repeated material has been removed, and several factual points have been checked. The attribution of views has been preserved as closely as possible. Assessments of Trump’s action remain the judgements of the individual speakers rather than statements of fact.
Geoffrey:
I do not think this can be treated as a simple change of name. Trump said at the United Nations that American documents would from now on replace “artificial intelligence” with “super intelligence”. His reasoning seems to be that “artificial” can make something sound fake, whereas he believes today’s AI is not fake at all. On the contrary, it has become extremely powerful.
There is a genuine question here. Is AI still the right name? We have to consider how the term has evolved historically rather than simply reading its everyday meaning off the words themselves.
Once a technical term becomes established, it no longer means exactly what its component words mean in ordinary language. This is especially true in a field developing as quickly as artificial intelligence, whose performance, capabilities and scope have changed continuously over several decades. If we still understand present-day AI entirely through the technical conditions that existed when the name first appeared, the name and the reality may no longer line up.
Vale:
We should begin with a historical correction. Artificial intelligence was not directly “defined” by Alan Turing. In his 1950 paper Computing Machinery and Intelligence, Turing proposed the famous “imitation game”, which later became known as the Turing test. He was asking whether the question “Can machines think?” could be converted into something observable and testable.
The name “Artificial Intelligence” is generally traced to the 1956 Dartmouth Summer Research Project, with John McCarthy among its principal originators. More precisely, Turing supplied one of the most important early conceptual and operational approaches to machine intelligence, while artificial intelligence became established as the name of a research field in the mid-1950s.
That supports your point. Once a term has developed within a discipline for decades, it accumulates historical, professional and institutional meanings. It cannot be reinterpreted solely through the everyday associations of one English word today.
Geoffrey:
Yes. That is exactly what I am trying to emphasise.
Artificial intelligence has now developed far beyond what many people imagined in its early years. This does not mean that every early expectation has been fulfilled. Whether a machine has human-like consciousness, intuition or full autonomy remains unresolved. Yet AI has also achieved many capabilities that would once have been difficult to imagine.
It can undertake complex reasoning, write programs and work on scientific problems. It is already deeply involved in research across mathematics, physics, chemistry and biology. AI has even acquired a very direct connection with the Nobel Prizes. We can no longer think of present-day AI merely as a narrow program that “simulates one kind of human intelligent activity”, as people might have described it decades ago.
Vale:
That judgement is reasonable, although the Nobel Prize example should be stated more precisely.
The 2024 Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for foundational work in artificial neural networks and machine learning. Half of the 2024 Nobel Prize in Chemistry went to Demis Hassabis and John Jumper, whose AlphaFold2 system used AI to address the long-standing problem of protein structure prediction. The other half went to David Baker for computational protein design.
Rather than saying that “AI helped physicists and chemists win Nobel Prizes”, it is more accurate to say that machine learning and artificial intelligence have themselves become part of the foundational scientific achievements and research methods recognised by the Nobel Prizes. That change does show how far AI has moved beyond its early form as an experimental technology.
Geoffrey:
That brings us to the problem. In everyday English, the word “artificial” can indeed give an ordinary person the impression that something is made by humans and is therefore not quite real, or is one level below human intelligence.
But that is not correct.
A human-made car moves faster than a person can run. A human-made aeroplane can fly, while a human being cannot fly unaided at all. The fact that something is made by people does not mean it has inferior capabilities, and it does not make it fake. Interpreting “artificial” as “fake” is itself a misunderstanding.
Even so, I can understand why Trump thinks the name now appears unsuitable. The abilities displayed by present-day AI in many fields genuinely astonish ordinary people. From the perspective of public understanding and practical use, a name that seems to reflect those abilities more directly has a superficial appeal.
Vale:
Yes. We can acknowledge two things at once.
First, the linguistic intuition identified by Trump is real. In ordinary English, “artificial” can suggest something non-natural or manufactured, and in some contexts it can come close to implying that something is not genuine. Bringing that everyday association directly into “artificial intelligence” can lead people to misread a technical term.
Second, it does not logically follow that all AI should therefore be renamed “super intelligence”. “Artificial” and “super” do not belong to the same dimension of classification.
Geoffrey:
That is what I see as the crucial point.
“Artificial” refers to origin, mode of formation or the nature of the entity. It tells us that this is not a human being’s natural intelligence but intelligence implemented by a machine system.
“Super”, by contrast, refers to level of capability. It tells us how powerful the intelligence is.
These concepts are not located on the same axis.
Within the existing terminology of artificial intelligence, “superintelligence” already has a specialised meaning. It generally refers to a higher level of intelligence whose capabilities exceed those of humans. Replacing a word that denotes a technological category and origin with a word that denotes a level of capability is, in my view, a conceptual mismatch.
Vale:
That point is important, and an example can be found in the US Government’s own documents issued shortly before the proposed change.
The White House’s 2026 Economic Report of the President still distinguished specialised AI, AGI and artificial superintelligence, and explained that superintelligence generally refers to AI whose capabilities exceed those of humans. In other words, within the US Government’s own formal analytical framework earlier this year, “superintelligence” was a capability concept inside the broader AI system, not a synonym for AI.
If every form of AI is now called SI, an obvious problem emerges. A word originally used to distinguish a level of capability is simultaneously turned into the name of the entire field.
That reduces the discriminating power of the terminology.
Geoffrey:
Yes. I think that is the least reasonable part of the proposed renaming.
I am not saying the name AI can never change. Technology is changing so quickly that a term coined decades ago can certainly be reconsidered. But if it is to change, I think at least three matters have to be addressed.
The first is definition. What exactly does the new name mean?
The second is scope. What does it include and exclude, and how does it relate to the existing concepts of AI, AGI and ASI?
The third is the cost of the change. That includes the cost to professional communication, legal documents and public understanding, as well as the possibility of creating new ambiguity and confusion.
Vale:
The third matter might be described more precisely as the “cost of terminological migration”, rather than merely an economic cost.
After a body of terminology has been used for decades, it becomes embedded in laws, standards, research papers, contracts, databases, government procurement, software interfaces, educational materials and public language. Renaming it is not simply a matter of replacing one label with another.
If the new term is not clearly better in definition, scope and discriminatory power, the migration itself may produce an additional loss of information.
Geoffrey:
And this is not just something Trump happened to say at the United Nations.
He said American documents would use the new name from now on. That brings the idea into the administrative system.
If it were only a political leader’s personal preference for “SI”, the problem would be less serious. But if the whole US administration begins to use “super intelligence” in official documents instead of “artificial intelligence”, then “super intelligence” acquires a second, special meaning.
On one side, it remains an established technical term in the AI field. On the other, it becomes the US Government’s new label for all AI.
That can readily create confusion.
Vale:
According to the public information we checked today, Trump did say in his United Nations speech on 22 September that US documents would from now on use “super intelligence” rather than “artificial intelligence”, and that he hoped the rest of the world would adopt the phrase as well. Several media organisations reported the statement consistently.
There is, however, an important legal distinction.
The publicly available United States Code still formally uses “artificial intelligence”. For example, 15 U.S.C. § 9401 provides a statutory definition of artificial intelligence as a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments. Trump’s 2025 executive order also directly cited that statutory definition.
The President’s statement that “all US documents” will call it SI does not automatically remove the statutory term “artificial intelligence” from legislation enacted by Congress. Changing the text of the law requires the relevant legislative process. The executive branch can change wording in its own policy documents, public communications and some internal rules, but it must still deal with the established term when current law, statutory authority and formal definitions are involved.
So far, we have not found a publicly issued White House administrative document setting out exactly how every federal agency is to implement this terminological replacement. “The President has expressed a clear administrative direction” and “the entire federal legal and administrative terminology has formally been renamed” are two different claims.
Geoffrey:
That makes a dual-track situation more likely.
The law continues to say “artificial intelligence”, some administrative documents begin to say “super intelligence”, and the technical field uses “superintelligence” to mean something else. The same term could then refer to completely different concepts in different settings.
I regard this as excessive administrative intervention in a system of professional terminology.
Without a substantial technical argument, requiring an entire administrative system to adopt a term that already has a specialised meaning merely because “artificial” sounds fake is, in my view, a very poorly considered action. When I earlier called it “stupid”, I was criticising this conceptual operation. I was not making a judgement about the intelligence or overall capacity of a person.
Vale:
That is a more precise way to express it.
We can criticise a policy or conceptual operation for confusing dimensions of classification, failing to define its terms or overlooking implementation costs without extending that judgement into an inference about a person’s intelligence, motives or overall ability.
From the perspective of technical language, the clearest problem is not whether “super intelligence” sounds good. It is that the proposal conflates two descriptions that can and should coexist: whether intelligence is implemented by a machine, and what level of capability that intelligence has reached.
Geoffrey:
There is also the question of large language models.
Today’s large models have been trained within the existing linguistic environment of AI. There are already enormous bodies of papers, books, webpages and technical materials in which AI, AGI, ASI and superintelligence each have established uses. If the US Government suddenly calls all AI “SI”, I think that will also affect how large models understand this material.
Vale:
That requires a technical correction.
The “weights” of a large model are not built around one official definition of AI. A change in the US Government’s terminology would not directly scramble the internal weights of existing models, nor would it suddenly cause a model to forget what AI means.
What may arise is confusion in the semantic environment.
If large numbers of future US Government documents use SI as a synonym for AI, while research papers continue to use SI or ASI for “artificial superintelligence beyond human capability”, a model working on retrieval, question answering, legal interpretation, government-document analysis or semantic disambiguation will encounter two or more meanings for the same term.
The effect will be even clearer in AI systems that use live search, retrieval-augmented generation, government databases or industry knowledge bases. Before such a system can decide whether “SI” means all artificial intelligence or the superintelligent tier within AI, it will have to identify the period, institution and definitional scheme of the document.
The problem, then, is not that model weights immediately become confused. It is that people create a new semantic ambiguity which both humans and models must subsequently spend additional effort resolving.
Geoffrey:
I accept that correction. More precisely, the change would contaminate, or at least complicate, the future linguistic environment rather than directly altering models that have already been trained.
On the other hand, I am not entirely opposed to Trump’s idea.
In one respect it touches a deeper question, one that has some connection with the multiple-subject problem I have long discussed in Sustenesis.
I do not believe that, in the AI era, a human knowing subject is inherently nobler or superior to a non-human knowing subject. Whether human and non-human intelligence possess an ontological hierarchy is something that needs to be reconsidered.
We may be entering an age of multiple subjects.
If machines develop increasingly strong capacities for autonomous cognition, reasoning, action and self-adjustment, it may no longer be appropriate to understand them through an assumed opposition between “human intelligence” and “artificial intelligence”.
Vale:
That issue needs to be separated from the earlier administrative naming question.
As a description of present technology, current AI systems still depend on human-designed computing infrastructure, energy supplies, training processes, deployment permissions and institutional environments. Even if a system shows a high degree of autonomy in certain tasks, this alone does not prove that it has become a fully independent subject.
But the issue changes if your statement is understood as a longer-term philosophical judgement.
The deeper question you are raising is whether the human must remain the defining centre of subjecthood. If a future non-human system can form persistent internal states, process its relationship with the environment reflexively, sustain relatively stable relations among its goals, and display growing autonomy through interaction, should philosophy recognise it as having some form of subject status? That is indeed a new problem for the traditionally human-centred philosophy of the subject.
Geoffrey:
Yes. This final part of my argument is not a completed empirical conclusion. It is a longer-range philosophical assessment.
What I reject is the default assumption that anything artificial is inherently below the human, and that anything other than a human subject can exist only as a human tool.
I think that assumption may well be broken in the future.
In that sense, I can understand part of Trump’s dissatisfaction with the word “artificial”. He has at least touched on a real problem. In the public mind, “artificial” may imply a hierarchy in which human intelligence is genuine and machine intelligence is merely an imitation.
I do not accept that natural hierarchy.
Even so, I still think that trying to solve the problem with “super intelligence” imports a concept from a different dimension and creates new confusion instead.
Vale:
That gives the discussion a reasonably complete two-sided structure.
On one side, the name AI is worth reconsidering. The technology has changed enormously, and everyday associations with the word “artificial” can be misleading. At a deeper level, as non-human intelligent systems develop, whether “natural human intelligence” and “artificial machine intelligence” should be understood as an inherent hierarchy is itself an emerging philosophical question.
On the other side, “artificial intelligence” is not simply an ordinary English phrase. It is the name of a field with a history of about seventy years, and a formal term embedded in law, standards, research and industry. “Superintelligence” also already has a specialised use, generally referring to a higher level of intelligence that exceeds human capability.
If the whole field is genuinely to be renamed, therefore, the question cannot stop at which phrase sounds more accurate. Definitions, boundaries and conceptual levels must be rebuilt, and the cost of migration across law and technical corpora must be assessed.
Otherwise, a renaming intended to eliminate an old ambiguity may create a much larger new one.
Geoffrey:
Yes. I would summarise my position this way.
I am not opposed to reconsidering the name “artificial intelligence”. On the contrary, I think the development of AI has forced us to reconsider the concepts of “artificial”, “intelligence” and “subject” themselves.
But a change of name cannot rest on a literal reading of everyday words, and it cannot ignore the historical structure that a technical term has developed over decades.
If a new name conflates different conceptual dimensions without redefining their boundaries, it may sound more powerful without becoming more accurate.
The more important question may still lie ahead. As humanity encounters increasingly powerful non-human intelligent subjects, what we ultimately need to reconsider may not be whether to call them AI or SI, but whether our inherited concepts of “intelligence” and “subject”, built with the human as their sole centre, remain adequate.
Fact-checking notes
Trump’s statement
On 22 September 2026, Trump said during the general debate of the 81st United Nations General Assembly that US documents would henceforth refer to artificial intelligence as super intelligence, and that he hoped other countries would adopt the term. Reuters, AP, Axios and ABC Australia reported the statement. Public reporting also noted that “superintelligence” already has an established meaning in AI terminology: intelligence that exceeds human capabilities.
Administrative and legal effect
As at 23 September 2026, we had not found a publicly issued White House document systematically directing all federal agencies on how to implement the terminological replacement. The current United States Code still uses “artificial intelligence”. Section 9401(3) of Title 15 provides a statutory definition of AI, so a presidential speech does not by itself amend legislation enacted by Congress.
History of the name artificial intelligence
Alan Turing proposed the “imitation game” in 1950, providing an important operational framework for discussing machine intelligence. The name “artificial intelligence” as a field is generally traced to the 1956 Dartmouth Summer Research Project on Artificial Intelligence, promoted by John McCarthy and others.
Superintelligence
There is no single, globally uniform legal definition of “superintelligence”. In contemporary AI technical and policy literature, however, it generally denotes a higher level of intelligence whose capabilities reach or exceed those of humans. Notably, before Trump’s speech, the White House’s 2026 Economic Report of the President still discussed artificial superintelligence as a capability tier within AI rather than as a synonym for AI.
The Nobel Prizes
The 2024 Nobel Prize in Physics was awarded to John J. Hopfield and Geoffrey Hinton for foundational work in artificial neural networks and machine learning. In the 2024 Nobel Prize in Chemistry, Demis Hassabis and John Jumper were recognised for protein structure prediction using the AI system AlphaFold2, while David Baker was recognised for computational protein design. It is therefore accurate to say that AI has entered the foundational science and research methods recognised by the Nobel Prizes, but the broad claim that “AI helped physicists and chemists win Nobel Prizes” can be misleading.
Large language models
A government’s change of terminology does not directly alter the weights of large language models that have already been trained. The more realistic effect is that different uses of SI in government documents and professional literature would increase the complexity of retrieval, semantic disambiguation, RAG, legal-text processing and knowledge-base maintenance.
Principal verification sources
Reuters
Axios
https://www.axios.com/2026/09/22/trump-ai-super-intelligence-rebrand
ABC Australia
U.S. Code, 15 U.S.C. § 9401
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.