What Gives AI Terminology Standing?

14 hours ago

20 min read

AI terminology illustration comparing Super Intelligence (SI) through government authority and legal scales with Artificial Intelligence (AI) through scientific research, technical measurement, and data analysis.

Renaming Artificial Intelligence as “Super Intelligence” is easy. Establishing the technical distinction is harder.

On September 29, 2026, President Donald Trump issued Executive Order 14434, directing executive departments and agencies to use the terms “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents. The order states that federal terminology should reflect what the administration describes as the transformative capabilities of contemporary systems. Within the executive branch and the limits stated in the order, this is an administrative direction governing which terminology agencies are expected to use. The change therefore has immediate institutional effect even though the systems being described remain the same systems that existed before the wording changed. [1] This analysis addresses the terminology and definitional state as of October 6, 2026.

What makes the order particularly useful as a case study in AI terminology is what happens next. For implementation, “Super Intelligence” and “SI” initially map onto the technologies and systems already encompassed by the statutory definition of artificial intelligence in section 9401(3) of title 15 of the United States Code. That statute defines 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. It further states that artificial intelligence systems use machine- and human-based inputs to perceive real and virtual environments, abstract those perceptions into models through automated analysis, and use model inference to formulate options for information or action. The definition describes a broad class of machine-based systems rather than a newly specified level of intelligence. The new terminology is therefore attached first to the existing statutory AI category, before any separate federal technical category is established. [1][2]

The sequence matters. The executive order requires the terminology substitution immediately within its stated scope, while separately directing the Assistant to the President for Science and Technology to develop proposed legislative language for a federal definition of “Super Intelligence” within 60 days. That later proposal is supposed to consider whether the new definition should modify, expand upon, or supersede the existing statutory definition of artificial intelligence. The order therefore separates the terminology imposed now from the definitional work contemplated afterward. The administrative designation comes first. Potential technical and statutory differentiation comes later. [1]

The National Institute of Standards and Technology provides an important second piece of evidence. NIST has stated that it is updating its communications to incorporate the term “super intelligence” as directed by the executive order. At the same time, NIST says the term will be added to its technical glossary when it is explicitly defined or cited in a final NIST technical-series publication. NIST is therefore complying with an administrative terminology requirement while maintaining a separate process for technical-glossary inclusion. The two actions operate under different kinds of authority. [3]

The larger problem begins with a simple separation. The September 29 directive changed the language required in a defined part of the federal government. A change in designation can be administratively legitimate without establishing a newly differentiated technical classification. The act of renaming left the underlying technical properties of the affected systems untouched. The order identifies no new training process, capability threshold, architecture, evaluation regime, or technical test that systems had to satisfy at the moment of the terminology change. A model did not acquire new weights, a system did not cross an identified benchmark, and an architecture did not become different because the executive vocabulary changed. Administrative classification changed through authority; whether that establishes a different technical classification is a separate question.

That question reaches far beyond one executive order. Artificial intelligence is a field in which terminology carries substantial weight. Terms distinguish systems, capabilities, levels of autonomy, technical architectures, risk categories, regulatory obligations, research objectives, and claims about what systems can do. Those distinctions influence how researchers compare systems, how policymakers write rules, how standards bodies organize concepts, and how organizations interpret capability and risk. When a word moves, the conceptual boundary associated with it can move as well. Deliberate, bounded changes can increase precision. A change that outruns an explanation of what changed in the object can create the appearance of a distinction before the distinction itself has been established. The central issue is what kind of standing newly introduced terminology possesses.

The Four Forms of AI Terminology Standing

For purposes of this article, standing is an analytical framework rather than a use of legal standing. It distinguishes administrative, statutory, technical, and evidentiary standing according to the basis on which a terminology claim operates.

Here, administrative standing means that an authority has sufficient institutional power within a defined jurisdiction to prescribe terminology for the people, documents, or processes subject to that authority. Executive Order 14434 establishes that kind of standing within the executive branch, subject to applicable law. Agencies can follow the terminology directive on the basis of that administrative authority; proof of a superior technical classification is a separate issue. Administrative standing answers a particular question: who has authority to require this language here? [1]

Statutory standing answers a different question. A legislature can define terminology for the purpose of a law, and that statutory definition establishes what the term means within the legal scope governed by that statute. Federal law continues to contain a statutory definition of “artificial intelligence.” The executive order recognizes that boundary by defining “Super Intelligence” through section 9401(3) for immediate implementation. It separately directs preparation of proposed legislative language if a future federal definition is to modify, expand upon, or supersede the existing one. Administrative and statutory terminology can therefore interact while remaining distinct. [1][2]

Technical standing concerns the object being classified. If one class of system is said to be technically different from another, the classification must identify the property, condition, capability, boundary, relationship, or other distinction doing the separating. That distinction can be multidimensional and context-dependent; it does not have to rest on a single numerical benchmark. A classification might depend on breadth of capability, depth of performance, autonomy, architecture, operating conditions, or some combination of those factors. What matters is that the distinction can be described well enough for another person to understand what places one system on one side of the boundary and another system on the other. That identifiable basis for membership is what allows terminology to add technical information.

Evidentiary standing asks whether a particular system satisfies the conditions of a defined category. A field can define a class before evidence establishes that any present system belongs inside it. A vendor can also apply a category before sufficient evidence supports the assignment. The difference matters because a coherent definition can exist independently of whether a current system has crossed its boundary. Evidentiary standing concerns the relationship between the classification and the observations, measurements, records, or other evidence used to apply it to an actual system. It asks what establishes that the thing being classified meets the category’s conditions rather than merely resembling the language used to describe them.

That requirement becomes more concrete when a classification is applied to an existing computational system. A theoretical category can exist before any present system satisfies it. An operational classification is different: the evidence has to attach to an identifiable realized system or configuration whose relevant state can actually be observed or measured. In this article, that is the limited sense in which instantiation matters. Instantiation does not prove the classification; it identifies the operational referent to which the observations, measurements, and evidence belong. What must be instantiated and recorded depends on the claim, but the object, relevant configuration, and operating conditions must be identifiable enough for the classification to be bounded and independently assessed.

Administrative, statutory, technical, and evidentiary standing can align in a mature classification. A government can adopt terminology that is already technically well defined. A statute can incorporate a recognized technical standard. Evidence can establish that a system satisfies the resulting criteria. In those circumstances, institutional authority, technical meaning, and evidentiary application reinforce one another while remaining analytically distinct. Authority to require a term establishes authority over terminology within that scope; the technical propositions carried by the terminology require their own support.

The executive order can be read purely as a federal administrative designation rather than as an attempt to establish a scientific taxonomy. On that reading, administrative standing is exactly what it establishes. A separate technical question arises when the designation is interpreted as describing a different technical condition. At that point, definition, boundary, criteria, and evidence become separately relevant.

The September 29 change deserves attention because it makes several layers visible at once. A government exercised administrative authority over language. Existing law still provides the statutory definition to which the new language initially points. NIST can update its communications while keeping technical-glossary inclusion tied to a technical publication. Those layers can coexist without contradiction because each answers a different question: what language is required, what the law defines, what a technical term means, and what evidence supports applying that term to a system. Terminology can evolve, but treating a terminology change as evidence of changed technical reality requires knowing which kind of change actually occurred. [1][3]

How AI Terminology Is Defined and Standardized

Institutions routinely develop, manage, revise, standardize, and harmonize technical language. ISO 704 establishes principles and methods for terminology work and explicitly links objects, concepts, definitions, and designations. ISO 29383 provides governments, administrations, nonprofit organizations, and commercial organizations with a methodology for developing terminology policies. ISO/IEC 22989 performs that work specifically for artificial intelligence by establishing terminology and describing concepts in the field. These processes exist because technical communication depends on more than choosing a preferred word. It depends on maintaining relationships among the object, the concept, the definition, the designation, and the context in which the term is used. Terminology governance is an established mechanism for creating enough consistency for different people and institutions to communicate about the same concepts. [4][5][6]

Governments, companies, standards bodies, and scientific communities all have legitimate reasons to define or refine terminology within their scopes. ISO’s terminology-policy work expressly anticipates different organizational settings and different terminology needs. That means institutional participation in terminology is a normal part of technical communication rather than a defect in itself. The analytical question begins one step later: what kind of claim is the resulting terminology entitled to make, how far does that claim travel beyond the institution that prescribed it, and what additional evidence is required when the term is used to describe a technical condition rather than an administrative convention? [5]

NIST’s own terminology work demonstrates why that distinction matters. Its Language of Trustworthy AI glossary is designed to improve common understanding and communication by recording and exposing multiple meanings used across an interdisciplinary field. NIST’s broader glossary likewise warns that terms can carry different definitions in publications written at different times and contexts, and it tells readers to return to the source publication for authoritative meaning. That approach preserves variation instead of hiding it. A reader can see that a term may carry different meanings and can trace each meaning back to its source. Technical vocabulary can therefore remain plural and meaningful when the definition, source, context, and purpose stay visible. [3][7]

The OECD takes a similar approach to the definition of an AI system. Its current definition describes a machine-based system that infers from inputs how to generate outputs such as predictions, content, recommendations, or decisions, while recognizing that systems vary in autonomy and adaptiveness after deployment. The accompanying memorandum makes clear that the definition is intended to remain broad enough to accommodate changing techniques and applications and that its relevant scope can depend on context. That design gives the definition enough stability to support policy while leaving room for technical variation among systems. The definition therefore performs two jobs at once: it identifies the object under discussion and marks the dimensions along which systems inside that object class may still differ. That is a boundary condition rather than a weakness: the definition tells the reader what the term is intended to capture and where contextual judgment remains necessary. [8]

Artificial intelligence began as a broad research designation. The 1955 Dartmouth proposal described a research program concerned with machines using language, forming abstractions and concepts, solving problems associated with human intelligence, and improving themselves. Its scope was not tied to a single architecture, algorithm, benchmark, or implementation. The proposal instead grouped multiple research problems under the larger category of artificial intelligence. [9]

That history shows why scientific terminology can be anticipatory. Artificial intelligence began partly as a research proposition about what machines might become capable of doing. A term can name a research objective, a hypothetical class, or an unresolved technical possibility before realization. The evidentiary obligation changes when the term moves from identifying a possibility to classifying an existing system. Pursuing a class of systems and establishing that a particular system belongs to that class are different propositions.

Bostrom’s 1998 treatment uses superintelligence to perform distinguishable conceptual work rather than treating it as another spelling of artificial intelligence generally. Nick Bostrom’s treatment defines superintelligence as intelligence substantially beyond the best human minds across essentially all important intellectual domains, rather than merely a system that performs individual tasks better than humans. His treatment also discusses superintelligence as a possible development following human-level artificial intelligence. The concept therefore carries a directional relationship to capability: it describes a proposed state beyond ordinary human intellectual performance across a broad field of activity. Whatever one thinks of the prediction, the term is doing more conceptual work than simply indicating that an artificial system is advanced or impressive. [10]

Bostrom’s definition is influential rather than controlling. Governments, technical bodies, companies, and other institutions remain free to define terminology within their own scopes. The capitalization and two-word construction used in Executive Order 14434 may itself signal an intended federal designation distinct from the historical one-word concept. The relevant fact is that the 2026 designation entered a vocabulary that already carried technical and philosophical meaning. [1][10]

This creates a semantic collision worth examining. In Bostrom’s treatment, superintelligence identifies an extreme capability distinction. In the September 29 order, “Super Intelligence” initially encompasses the systems already covered by the existing statutory definition of artificial intelligence. The same linguistic neighbourhood is therefore serving two different classification functions: one describing an advanced or hypothetical capability class, and one administratively replacing the broader AI designation within the executive branch. Both usages may be legitimate within their respective scopes, but they do different work. Readers therefore need the applicable scope and definition before treating them as the same technical category. [1][10]

What Makes a Technical Classification Meaningful?

The simplest way to test whether a new technical classification adds information is to ask what distinguishes membership in the new class. A category can be defined by one benchmark or by multiple dimensions, qualitative conditions, thresholds, or relationships. What matters is that some difference carries the classification. The label adds technical information only when that difference can be identified.

Google DeepMind’s work on levels of artificial general intelligence provides a useful example of this type of classification discipline. The researchers examined competing definitions, identified principles for a useful ontology, separated depth of performance from breadth or generality, proposed graduated levels, and discussed how future benchmarks might measure systems against those levels. They also treated autonomy and risk as related deployment considerations rather than collapsing every property into a single word. That structure allows disagreement about the thresholds while preserving visibility into the dimensions being classified. The resulting framework is one proposal, but it demonstrates what it means to make a classification inspectable. [11]

The importance of that example is methodological. DeepMind’s particular levels could later be revised or rejected while the underlying method remains useful. A classification becomes more technically useful when someone can ask what separates one category from another, what observations would support membership, what measurements remain inadequate, and where uncertainty persists. A category becomes more than a label when it provides a structure for distinguishing states of the world.

Apply that same test to the initial definition in Executive Order 14434. Section 3 states that, for purposes of the order, “Super Intelligence” and “SI” mean the technologies and systems encompassed by the existing statutory definition of artificial intelligence. At that point, back-substituting the older designation leaves the technical content unchanged; the order has yet to introduce a separate technical partition. The order provides a policy rationale based on advancing capabilities, while its operative initial definition carries forward the existing AI class and contemplates a different federal definition later. [1]

The finding is limited to that sequence. Questions about whether present systems could satisfy a defensible definition of superintelligence, or whether the federal government could later create a technically useful “Super Intelligence” category, remain open. What the order establishes at the moment of substitution is that the required nomenclature changes before a separately differentiated technical class is defined within the order.

The Autonomy of Technical Terminology

Existing terminology already explains much of the landscape. Terminology governance covers how institutions manage language. Standardization coordinates common vocabulary. Definitions describe intended meanings. Taxonomies organize classes. Harmonization reconciles competing terms or concept systems. Semantic consistency helps keep usage stable inside a defined context. Those concepts already explain how vocabulary is built, compared, maintained, and revised. The remaining question is more specific: how can the technical proposition embodied by a designation remain separately examinable when an institution has authority to prescribe the word? That residual concerns the relationship between external authority and the evidence supporting the technical meaning.

That residual is where autonomy becomes useful. Here, autonomy is article-specific: it refers to the independent examinability of the technical claim carried by terminology. The technical validity of a classification remains answerable to its concept, scope, criteria, and evidence even when people or institutions create, prescribe, standardize, or revise the language. An authority may determine what people inside its jurisdiction are required to call something. The technical claim represented by the word remains separately examinable.

Technical terminology is always produced and maintained through human and institutional processes: communities adopt or reject terms; institutions standardize them; legislatures define them; researchers revise them; autonomy protects the distinction between those processes and the evidence supporting the technical proposition. The relevant independence is the ability to examine technical validity separately from the status of the speaker.

Back-substitution helps test whether autonomy adds anything beyond established concepts. Semantic integrity addresses internal coherence; terminology governance addresses vocabulary management; classification validity addresses supportability; standardization can create common usage. Among the terminology concepts examined here, none by itself names the independence of technical validity from the status of the institution using the term. Autonomy adds that boundary: technical meaning remains separately examinable from the authority of the institution asserting it, even when the institution has legitimate power to prescribe language within its own scope.

A terminology-standing test can therefore be stated mechanically. What is the provenance of the term? What definition is being used? What is its scope? What separates it from adjacent categories? What criteria determine membership? What identifiable system or configuration is actually being classified? What operational state and conditions are being observed or measured? What evidence supports applying it to the system in question? What authority is asserting it, and what kind of authority is that? In what context does the definition operate? Is the usage internally consistent? What evidence would show that the classification does not apply to the system in question, or that the assignment should be narrowed? Could another competent evaluator reconstruct and independently assess the classification from the same definition, criteria, evidence, and conditions? How would a later revision be recorded? The value of the test is practical: it forces the reader to reconstruct the path from word to concept, from concept to criteria, and from criteria to evidence. That makes it possible to distinguish a legitimate contextual definition from a label whose apparent technical meaning rests mainly on repetition or institutional status. This article-specific test is derived from the distinctions above rather than from an ISO or NIST standard.

Applied to artificial intelligence generally, the test allows multiple valid definitions within explicit scopes. OECD can define AI for its recommendation. NIST can preserve multiple contextual definitions across technical publications. ISO can establish terminology for standards and cross-stakeholder communication. A legislature can define AI for the operation of a statute. Each can possess the form or forms of standing appropriate to its stated scope because the origin, purpose, definition, and context can be identified. Multiple definitions can therefore coexist without becoming arbitrary. [3][6][8]

Applied to Bostrom’s treatment of superintelligence, the test shows traceable provenance and an identifiable conceptual distinction. That does not establish that any present system satisfies the category; evidentiary standing remains separate. Applied to Executive Order 14434, the distribution is different: the order clearly creates administrative standing for “Super Intelligence” within its prescribed scope, while technical differentiation beyond the incorporated statutory AI category remains to be established through definition, criteria, and evidence. [1][10]

A classification can gain stronger technical standing as its distinctions are defined, its boundaries clarified, and its criteria specified. Its evidentiary standing strengthens as evidence establishes that relevant systems satisfy those criteria. Terminology therefore has room to evolve with technical progress. Autonomy permits that evolution while keeping the relationship among the word, definition, underlying concept, and evidence open to examination.

When AI Terminology Shapes Governance and Measurement

Why does this distinction matter beyond vocabulary? Artificial intelligence terminology now operates inside technical standards, laws, risk frameworks, and governance documentation. Once a category is used to determine which requirements apply to a system, terminology becomes part of the operating environment around that system. A word alone does not enforce a rule, but the category it represents can determine which rules, tests, disclosures, controls, or obligations are considered relevant.

The European Union’s AI Act provides a concrete example. It creates defined categories and attaches consequences to those classifications. A general-purpose AI model can be classified as a general-purpose AI model with systemic risk when specified conditions are satisfied. The Regulation identifies high-impact capabilities, technical tools and methodologies, indicators, benchmarks, training computation, model characteristics, reach, and other factors that can inform that classification. It also provides mechanisms for reassessment as technology changes. The value of the example is methodological: the classification is connected to identifiable criteria, evidence, and consequences, so a reader can inspect both the category and the process by which an object is placed inside it. [12]

Terminology becomes operationally consequential when it enters governance. The EU AI Act separates an AI system, a general-purpose AI model, and a general-purpose AI model with systemic risk because different classifications determine which provisions apply. A model and an AI system are treated as distinct regulated objects for those purposes. The definition determines the object, and the object helps determine the applicable requirements. [12]

The same principle operates outside law. ISO/IEC 22989 exists to establish artificial-intelligence terminology and describe concepts in the field partly so diverse stakeholders can communicate with sufficient consistency. NIST’s trustworthy-AI glossary likewise exists because organizations need enough common language to communicate about AI risk and trustworthiness. Its glossary deliberately exposes multiple meanings and directs readers to the originating publication for the authoritative meaning in context. This approach gives disagreement a structure: definitions can differ while provenance, context, and intended use remain visible. The objective is traceable meaning across plural definitions rather than forced uniformity. [3][6][7]

Contextual definitions are compatible with terminological discipline. A term can mean one thing inside a statute, another thing inside a technical standard, and something broader in ordinary discussion. The condition is that the context and definition remain recoverable. Problems emerge when the authority of the speaker obscures which definition is operating, what distinction the term is supposed to carry, or what evidence establishes that the classification applies.

Measurement: When Does a New Label Reflect a New State?

Measurement presents the same problem from another angle. If researchers claim that one system has crossed from one capability category into another, they need some way to explain the difference being measured. The measurement may be incomplete, the benchmarks may improve, and the category may be multidimensional. Precision can remain provisional while the classification stays answerable to identifiable criteria. The reader needs an observable difference to determine whether the name represents a change in technical state or only a change in language.

That also makes it useful to distinguish different kinds of transition. A designation transition changes the word being used. A definitional transition changes the boundary or rule attached to the word. An operational-state transition occurs when a property of the system relevant to the claim changes under the stated operating conditions. An operational-class transition occurs when the observed or measured state crosses the defined boundary of a category. These transitions can occur together, but none should be treated as evidence of another without showing the relationship. A change in designation therefore does not, by itself, establish a change in operational state.

The issue becomes sharper as AI vocabulary moves among communities. A researcher may use “reasoning” to describe performance on a particular class of tasks. A product company may use the same word to describe a feature. A policymaker may use it as evidence of increasing capability. A member of the public may hear it as a claim about human-like cognition. The same sequence can occur with “agent,” “autonomy,” “intelligence,” “understanding,” “general intelligence,” and “superintelligence.” Each transfer can preserve the word while changing the proposition attached to it. The important question is how much of the original claim travels when the word crosses contexts and whether the receiving context makes that shift explicit.

Terminology therefore belongs inside the larger governance conversation. Governance depends on identifying what is being governed. Measurement depends on identifying what is being measured. Assurance depends on identifying what proposition the evidence is supposed to support. Accountability depends on knowing what action, system, actor, or condition is being discussed. A moving referent can leave the reasoning grammatically coherent while the underlying proposition changes.

AI vocabulary can evolve while remaining accountable. NIST’s glossary preserves multiple definitions across publications and contexts and directs readers back to the originating source for authoritative meaning. That approach keeps meaning traceable as terminology changes. [3][7]

That is the relevant autonomy: technical propositions remain examinable independently of institutional preference. Someone must still be able to ask what the term means, what changed, which criteria distinguish the category, and what evidence supports its application.

Symmetry matters. The same requirement for appropriate evidentiary support should apply whether terminology comes from government, industry, academia, standards bodies, or another source. The relevant question is what type of claim is being made and what gives that claim standing within the context where it is being used. That symmetry is important because the article’s test is meant to separate source authority from claim support, not to rank institutions by prestige. A technical classification introduced by a laboratory, regulator, standards body, company, or government must be supported through the same relationship among definition, scope, criteria, and evidence.

The same standard applies to anyone introducing new technical terminology. If established terminology already performs the same mechanical work, it should ordinarily govern. A new term earns its place by adding identifiable meaning, staying within a bounded scope, and making its relationship to existing concepts clear. Otherwise semantic inflation becomes the problem rather than the solution.

The terminology-standing test is therefore a test of reconstructability.

A term can vary by context while the relevant form or forms of standing remain identifiable when its origin, boundary, and purpose are explicit. A definition can evolve while preserving its stated scope when the revision is visible. A classification can remain uncertain while its technical and evidentiary standing remain separately assessable. Technical or evidentiary standing weakens when the relationship between the language and the conditions supporting the classification becomes unclear.

Return, then, to September 29, 2026. The executive order directs federal executive agencies to use “Super Intelligence” and “SI” instead of “Artificial Intelligence” and “AI” in specified non-statutory communications. It provides a policy rationale centred on rapidly advancing capabilities. For implementation, however, it initially defines the new terms by reference to the technologies and systems already encompassed by the existing statutory definition of artificial intelligence. It separately directs the development of proposed legislative language for a federal SI definition. The case therefore returns us to the same layered structure established at the beginning: administrative terminology changes now, while distinct definitional work remains a separate process. Those are the facts of the sequence. [1][2]

NIST’s current treatment makes the separation even easier to observe. NIST says it is working to update its communications to incorporate “super intelligence” as directed by the executive order. The same glossary page says that when the new term is explicitly defined or cited in a final NIST technical-series publication, it will be added to the glossary. Administrative adoption and technical-glossary qualification are therefore occurring as separate events inside the same federal environment. That separation demonstrates how one institution can comply with required terminology while preserving a separate technical-definition process. The process itself makes the distinction visible. [3]

Conclusion: AI Terminology, Authority, and Evidence

The analysis leaves the future meaning of “Super Intelligence” and “superintelligence” open. A new classification can become technically meaningful if its definition establishes a real distinction, its scope is bounded, and the evidence needed to apply it becomes identifiable.

A theoretical category can be defined before any existing member is established. A claim that an existing computational system occupies that category does require an identifiable operational referent and evidence tied to its relevant operational state.

The distinction is straightforward: authority can create required usage; law can create statutory definitions; standards bodies can establish terminology for specified purposes; researchers can propose new classifications; communities can popularize them. Each of those mechanisms can legitimately influence language; however, evidence has a different job: it allows someone to examine whether the condition represented by the classification exists in the object being classified, and whether the technical difference carried by the term can be demonstrated rather than assumed.

That is the autonomy technical terminology needs: its technical meaning remains independently examinable even as people, institutions, and standards bodies revise the language around it. Authority may choose the term. Evidence still has to carry the technical claim.

When a government, company, research laboratory, standards organization, journalist, or commentator announces that AI has become something else, ask the mechanical questions first. What changed? What remained unchanged? What does the new category mean? Where is its boundary? What condition has to be satisfied to cross that boundary? What evidence shows that the condition has been satisfied?

If those questions have answers, terminology can evolve with the technology it describes; if those questions remain unanswered, the terminology may still acquire usage, popularity, institutional adoption, or administrative authority. Artificial intelligence does not become superintelligence merely because the words change; if the classification is different, the difference should be able to stand on its own.