AI to Super Intelligence: What the Rename Actually Changed

Editorial Note: This analysis examines what changes when a technology is renamed while its underlying definition remains largely unchanged. It focuses on terminology, systems, measurement, and reasoning rather than evaluating the policy itself.

The Rename Changed the Label, Not the Thing. A six-layer breakdown of AI → Super Intelligence

0. The name changed on day one. The definition is due in sixty days.

On September 29, 2026, an executive order titled Inaugurating the Era of Super Intelligence took effect. Federal executive agencies were directed, to the maximum extent permitted by law, to use "Super Intelligence" (SI) instead of "Artificial Intelligence" (AI) in official correspondence, public communications, websites, reports, policy documents, and other non-statutory executive-branch documents. The same day, leaders of several major AI companies signed a one-page White House Super Intelligence Compact. On October 4, a coordinating body called the Super Intelligence Force (SIF) was announced, led by Director of National Intelligence Jay Clayton, with a mandate to coordinate the federal response to AI and assess its risks and opportunities.

The interesting part is the gap between those dates. The label flipped immediately. What "Super Intelligence" will mean as a federal statutory term is still open. Within sixty days, the President's Assistant for Science and Technology is required to submit proposed legislative language that would establish a federal definition and assess whether it should modify, expand, or supersede the existing statutory definition of "artificial intelligence."

That gap is the entry point for everything that follows. A term's real power never comes from which document it appears in. It comes from how many downstream systems already depend on it. So this piece skips the question of whether the new name sounds better, and works down six layers instead. Each layer answers a different question:

The order matters. The semantic layer is only the doorway. Layer three is where the outcome gets decided. By the end, you'll see that the most consequential part of this rename lives in two dates that haven't arrived yet.

A scope note: this is an analysis of how terminology works, not an evaluation of any policy. The political timing shows up exactly once, as one explanatory variable for "why now."

1. AI vs. Super Intelligence: A Scope Expansion, Not a Synonym Swap

Start with what each word carries.

Artificial carries an important ambiguity. It can signal something human-made, constructed, or non-natural, while in some contexts it can also evoke ideas of imitation or inauthenticity. Those associations became especially salient in public discussions of generative AI after 2022, including debates around deepfakes, hallucinated outputs, and authenticity. The White House fact sheet says as much: the new name is meant to convey the technology's actual capabilities, while the old one "implies merely imitating or automating human intelligence" (White House, Sept 29, 2026).

Super is an unbounded intensifier. It doesn't qualify the origin of anything. It makes a claim about degree, and the direction of that claim is beyond.

So the first shift is precise: a qualifier became a degree word. The old term answered "where did this come from." The new one answers "how strong is this."

The second shift matters more, and it's about lineage. In existing literature, superintelligence has a substantially narrower referent. Bostrom defines superintelligence as an intellect that greatly outperforms the best human brains across practically every field. Some later discussions distinguish artificial narrow intelligence, artificial general intelligence, and artificial superintelligence as progressively broader capability categories, but that ladder is a conceptual framework rather than a universally accepted scientific taxonomy. Under those frameworks, superintelligence represents a hypothetical capability level beyond human general intelligence.

What the order does is take that top-of-ladder concept and use it to cover the whole set. In terminology terms that's a scope expansion: the extension grows, the intension thins out.

The result is one word with two legitimate readings, depending on who's speaking:

• In federal documents, SI refers to everything the current legal definition of AI covers — spam filters, recommendation engines, customer-service bots.
• In research literature, superintelligence still refers to a capability ceiling nobody has demonstrated.

Here's a concrete failure mode. Suppose a 2027 solicitation says "vendors must describe their SI capabilities." A bidder is entirely entitled to read that narrowly — I have no superhuman systems — while the buyer may intend the broad reading — list every model you run. The ambiguity isn't in the word. It's in whose definition wins.

A useful reference point: renaming is normal in this field. When "artificial intelligence" was coined at Dartmouth in 1956, it was competing with machine intelligence, cybernetics, and electronic brain. The one that survived seventy years did so because of citation networks, textbooks, and standards lock-in — not because it was the prettiest option. The real question was never "can you rename it." It was "which name sticks, and why."

2. Same Referent, New Label: Why the Definition Clause Matters More Than the Name

This layer is the one people skip, and it sets the boundary on every inference that follows.

Section 3 of the order maps SI onto the definition of "artificial intelligence" at 15 U.S.C. 9401(3), and explicitly states that previously issued regulations, presidential actions, contracts, appropriations, and other historical documents need not be revised (White House, Sept 29, 2026).

In engineering language: the pointer got renamed. The address it points to didn't move.

Three things follow from that:

1. The technical referent is unchanged. Architectures, benchmarks, capability curves, deployment patterns — continuous across September 29.
2. Existing legal obligations are not changed by the renaming order itself. The order does not rewrite previously issued regulations, contracts, grants, or other historical documents simply because the terminology changes.
3. A measured shift in attitudes or behavior cannot automatically be attributed to a capability change. If the technical referent is held constant, a controlled comparison can isolate the label as one possible explanatory variable. In real-world data, however, other events and changes in the technology can move attitudes at the same time.

Worth clearing up a common conflation here: treating a narrative upgrade as a capability jump. They require different evidence. A capability jump shows up as reproducible artifacts — scores on new evaluation suites, formal verification results, long-horizon autonomous task completion, third-party replications. A narrative upgrade shows up as changed wording. What's publicly visible right now is the second kind.

One practical takeaway for readers: when you pick up any document about this technology, go to the definitions section first. Definitions set scope. Scope determines your obligations and your rights. The name is just an index.

3. Why Renaming a Technology Is Hard: Every System Locked to the Word "AI"

This is the densest layer, and the one most coverage of the rename skips.

A technical term's real weight equals the number of systems coupled to it, multiplied by each one's switching cost. Here's the coupling map for "AI":

Three observations.

First, terminological authority is distributed. On September 30, California signed Executive Order N-10-26, directing state agencies to continue using "Artificial Intelligence" and "AI" in official business, unaffected by the federal naming (California EO N-10-26, Sept 30, 2026). The federal order binds federal executive paperwork. It doesn't reach states, companies, universities, or the press. That boundary is itself the finding.

Second, adoption is selective. On October 4, Elon Musk endorsed SI publicly and said SpaceXAI would become SpaceXSI — but as of October 5, the website, handles, and registrations were unchanged. Some vendors have picked up the new word when describing facilities. Most product lines haven't. Selective adoption is data in its own right: companies follow where renaming buys policy alignment, and don't follow where it costs brand equity.

Third, the non-retroactivity clause creates a corpus fork. Documents before September 29, 2026 say AI. Federal documents after that date say SI. The legal meaning is identical. Downstream, that's a real engineering problem: contract review, compliance scanning, retrieval systems, and RAG pipelines that match on a single term will miss things at the fork.

Two concepts worth naming:

• Terminology debt — the long-run maintenance cost created by a rename, distributed across every downstream system, while the benefit accrues to whoever issued it. This debt doesn't disappear when the document is signed.
• Dual register — SI for government, AI for market and users. It's the lowest-cost rational strategy, and it also means the new term faces almost no penetration pressure on the consumer side.

A counterintuitive corollary: the more thorough the rename, the heavier the terminology debt. If the definition due in sixty days turns out broader than the current AI definition, then every contract, standard, and precedent that leans on the old definition needs reinterpreting. If it turns out narrower, you get a gap — systems regulated as AI that don't count as SI. Both produce maintenance cost. The only difference is who pays it.

4. Does Renaming AI Change Behavior? Framing, Adoption, and Measurement Hazards

Straight answer: the changes we can confirm are at the wording level. No systematic behavioral change has been observed yet.

Adoption signals. Federal paperwork has switched. SIF carries the new word in its name. Individual executives use SI in public. What hasn't been observed: standards bodies changing language, corporate filings (10-K and equivalents) changing language, product naming changing, procurement specifications changing at scale. Those four are behavioral evidence. The first three are statements.

Attitudinal context. The rename landed during a period of net-negative public sentiment. A Reuters/Ipsos poll fielded September 17–20, 2026 (n=1,277, ±3) found 39% saying AI has a net negative effect on society — the highest since tracking began in March — against 11% net positive; 73% saying AI companies aren't doing enough to prevent serious harms; 55% saying slower development would be a good thing (Reuters/Ipsos, Sept 22, 2026). These numbers explain why now. They measure attitudes under the "AI" label, so they can't be used as evidence that the rename worked or failed.

On framing effects, the honest position is narrower than people assume. The literature supports the direction — labels shift judgments — but effect sizes are contested and highly context-dependent. Treating "renaming meaningfully improves public attitudes" as settled is unsupported. So is the opposite claim that renaming does nothing at all; the change at the federal-document level is real and durable.

One useful analogy is the euphemism treadmill. Steven Pinker describes a recurring pattern in which a replacement term can acquire some of the negative associations attached to the thing it describes, eventually creating pressure for another replacement. Examples often discussed in this literature include changes in terminology around nuclear waste and secondhand smoke. The analogy is imperfect here: "Super Intelligence" is not simply a euphemism for "Artificial Intelligence." The useful question is narrower — whether the new label can maintain a different set of associations while the underlying experience remains broadly continuous.

Does the pattern apply to "super intelligence"? It hinges on one variable — whether public day-to-day experience of the technology improves or worsens. If electricity prices, employment effects, and incident rates trend better, the new term has room to survive. If they trend worse, the new term loses neutrality faster than the old one did, because its expectation baseline was set higher.

The measurement trap — most useful item in this section for anyone doing research. If a follow-up survey swaps "AI" for "SI" in its question wording, longitudinal comparability breaks: pre- and post-September data may be measuring different conceptual frames. The fix is unglamorous — keep both labels in the question ("AI, also referred to as Super Intelligence"), and record the wording-change date in the dataset. Same for any text-mining study tracking public sentiment: mark a breakpoint at 2026-09-29, or you'll read a wording change as an attitude change.

5. Will "Super Intelligence" Stick? A Term-Lifecycle Test with Falsifiable Markers

Five variables predict whether a technical term survives:

1. Does it have an operational definition? Without one, it can't enter standards or contracts.
2. Is it bound to standards or contracts? Bound means durable; bound only to publicity means fragile.
3. Is it adopted by users rather than issuers? The people writing code, signing contracts, and publishing papers decide, not the people announcing.
4. Does it carry negative experience of the referent? If yes, it enters the treadmill.
5. Is there an incumbent with network effects? Stronger incumbent, higher replacement cost.

Five historical cohorts, five different endings:

• Information superhighway (1990s): strong political and media push, no operational definition, no standards binding. Dead.
• Web 2.0 (2004–2010): conference-marketing origin, briefly a funding vocabulary, no technical definition. Faded as a technical term; survives as a periodization label.
• Metaverse (2021–2023): heavy capital and brand push, real products, experience short of expectation, tied to negative associations. Sharp decline.
• Cloud computing (2006– ): operational definition (NIST SP 800-145), bound to procurement contracts and service-level agreements, adopted by users across the board. Survived and became the default.
• Artificial intelligence (1956– ): textbook and standards lock-in, dense citation network, seventy years of accumulated indexing and retrieval assets. Extremely hard to replace.

Place "Super Intelligence" in that coordinate system. Variable 1 is pending — the definition proposal is due in sixty days. Variable 2 is currently close to zero: the term has not yet established itself in major standards, and there is no evidence of broad contractual adoption. Variable 3 is near zero: enterprise-side signals are individual statements, not execution. Variable 4 depends on public experience over the next two years. Variable 5 is badly against it — "AI" is arguably the strongest-network-effect technology term in current use.

Three scenarios, each with what would falsify it:

• Scenario A — official documents only. Federal paperwork uses SI long-term; industry and academia keep AI. Trigger: the definition proposal isn't enacted, standards bodies don't follow. Falsified by: a standards body revising its language, or large-scale contract terminology change.
• Scenario B — durable dual register. Government documents use SI, products and markets use AI, indefinitely. On current evidence, this is the best-supported baseline. Falsified by: major vendors renaming consumer product lines, or a structural migration in search-volume distribution.
• Scenario C — full adoption. SI becomes the cross-context default. Requires: legislative confirmation of the definition, plus standards adoption, plus major vendors using it in contracts and products. Falsified by: a vague definition proposal, or the 120-day report recommending no terminology harmonization.

Two checkpoints: the sixty-day definition proposal — its breadth determines regulatory and compliance coverage — and the 120-day SIF report, which will indicate whether any terminology or statutory changes are recommended. Before those land, any verdict on whether the rename "worked" is unsupported.

6. How the Rename Shifts the Mental Model You Use to Reason About AI

This layer connects directly to how people think with and about intelligent systems, and it's the part most coverage leaves out.

Labels are compressed heuristics. Nobody evaluates every attribute of a technology from scratch; a name pulls up a bundle of default expectations. "Artificial" pulls up human-made, possibly not real, verify before trusting. "Super" pulls up high capability ceiling, leading edge, worth anticipating. Changing the label changes the starting point of the inference.

Three self-checks follow from that:

Salience reordering. An intensifier can make "capability" more salient in a judgment and may reduce the relative salience of "risk." That doesn't mean people become blind. It means the same person, given the same facts, may produce a different risk rating under two labels. The test is direct: swap the name back to "AI" and re-read. If your conclusion moves, part of it was resting on the label rather than on evidence.

Definition anchoring gets more important, not less. When one word has two legitimate readings — the narrow superhuman sense and the broad "all AI systems" sense — readers tend to default to whichever supports their point. That's the most common entry point for concept drift. The habit worth building: state which definition you're using before the argument starts, and state the scope.

Treat intensifiers as claims requiring evidence. "Super," "revolutionary," "surpassing human" are all testable propositions in technical terms. When you encounter one, the next question should be which evaluation, run by whom, replicated where — not acceptance as a premise.

One adjacent observation on human–AI collaboration: framing in a prompt does shift model output tendencies, and that's reproducible in practice. But the effect is context-dependent, and it shouldn't be over-extrapolated into "renaming changes model capability." Naming is worth testing as a variable. It isn't worth accepting as a conclusion.

7. Practical Checklist: Handling the AI/SI Terminology Shift in Docs, Contracts, Data, and Content

This section is for people who have to deal with the change in actual work. Each item includes the reason.

1. Build a term mapping table. List AI / SI / superintelligence / AGI / ASI with their respective scopes and the places they aren't interchangeable. Tag each with its source (statutory definition, academic definition, administrative usage) and effective date. Why: after September 2026 these four terms can't be used interchangeably, and mixing them creates real ambiguity in contract and compliance contexts.

2. Contracts and legal documents: explicit definition plus a successor clause. Define the referent scope in the definitions section, add "including any successor name, renamed expression, or equivalent term," and date-stamp the definition. Why: non-retroactivity protects historical documents, not future contracts. Without a successor clause, a single rename can put the interpretation of existing obligations in dispute.

3. Data and analysis: keep both labels, add a version field. Write survey items as "AI (also referred to as Super Intelligence)." Add a terminology-version field to datasets, mark 2026-09-29 as the switch point, and treat time series with a breakpoint. Why: otherwise you'll read a wording change as an attitude change, and your trend lines will be wrong.

4. Content and SEO: run both terms, with clear division of labor. Use the new term in titles and metadata to match emerging query intent; keep "AI" in body copy to preserve authority and existing long-tail coverage. During the transition, query volume splits across both — monitor them separately. Why: search behavior migrates gradually. Betting on one term loses traffic on both sides during the crossover.

5. Standards and compliance: track wording, distinguish audiences. Watch whether NIST and ISO/IEC documents revise their terminology. For documents aimed at US federal audiences, remember the order covers non-legal documents only; legal documents still use the current AI definition. Why: mistaking an administrative convention for a legal standard is the most common compliance error in situations like this.

6. Retrieval and automation pipelines: add synonym expansion. Put bidirectional AI ↔ SI mapping into contract review, compliance scanning, and RAG indexes, and time-tag historical corpora. Why: single-term matching misses items at the fork, and those misses typically surface during an audit rather than at build time.

8. Two Checkpoints, and One Question Left Open

Pulling the six layers together: the sign layer saw a scope expansion. The referent layer saw no change. The system-coupling layer remains almost entirely locked to the old word. The behavior layer currently offers wording evidence only. The survival layer has more unfavorable variables than favorable ones. The cognition layer introduces a testable possibility: the new label may change some readers' default reasoning path even when the underlying referent stays constant.

So the accurate summary is this: the rename changed the label and the register, not the thing. Its long-run effect depends on whether third parties — standards bodies, companies, researchers, users — adopt it, not on the issuer's intent. The sixty-day definition proposal and the 120-day risk report are the two real checkpoints.

One question for you, and the one I'm most curious about: in your own work, has a term ever been renamed "officially" while your actual usage never changed? If most people answer yes, the pattern may be more worth recording than "super intelligence" itself.

Read More of Intelligenr

• How AI Is Used in Quantitative Trading (and Where It Breaks)

• Does AI Make You Worse at Your Job? The Feedback Problem Nobody Measures

• How to Get Better Results from AI: Beyond Prompt Templates

• Why AI Gets Slower Over Time: Memory, Bandwidth & Physics

References

1. The White House (2026) Inaugurating the Era of Super Intelligence — Executive Order 14434 https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/

2. Reuters / Ipsos (2026) Three out of four Americans say AI firms not doing enough to prevent disaster, Reuters/Ipsos poll finds https://www.reuters.com/world/three-out-four-americans-say-ai-firms-not-doing-enough-prevent-disaster-2026-09-22/

3. Nick Bostrom (2014) Superintelligence: Paths, Dangers, Strategies https://www.oxfordmartin.ox.ac.uk/publications/superintelligence-paths-dangers-strategies

Author Note: The terminology discussed here is still evolving. The analysis reflects the evidence available at publication and treats the upcoming definition and policy checkpoints as open questions rather than settled outcomes.