The Evolution of AI’s Definition,
1950 to Super Intelligence (SI)

The technology did not change in September 2026. Only the federal government’s name for it did — and that is not even a first. Alan Turing never used the words “artificial intelligence.” What follows is the evolution of the AI definition since — the field has renamed, redefined, and re-argued itself every decade — and Iternal supports the latest entry on that list. The reason is in the pattern: naming in this field has never been settled once. It has tracked capability every time it moved, and Super Intelligence (SI) is where that tracking lands now.

What follows is the history of the definition of AI, Turing to today: the operational test that avoided defining “intelligence” at all, the two-month workshop proposal that coined the term, the textbook rewrites that moved the goalposts twice, and the executive order that gave the same underlying technology the name its measured results have earned.

1950: Turing’s Test

The field that would later be called artificial intelligence does not begin with a definition. It begins with a refusal to give one. Alan Turing opened “Computing Machinery and Intelligence” in October 1950 by proposing to consider the question, “Can machines think?” (Source: Turing, "Computing Machinery and Intelligence," Mind, Oct. 1950) He judged the words “machine” and “think” too slippery to pin down directly and replaced the question with an operational test instead: the imitation game, later called the Turing Test. An interrogator, questioning a human and a machine by typed messages alone, has to tell which is which. If the machine can pass for human often enough, Turing treated the philosophical question as answered in practice, without ever settling what “thinking” means in theory.

The term “artificial intelligence” appears nowhere in Turing’s paper — it did not exist yet. That is the detail easiest to miss and most useful to hold onto: the field’s founding document offers a Turing test definition of machine thinking, not a definition of a field called AI. The field, and the name, arrived five years later.

1960s–1980s: Symbolic AI and the Narrowing Definition

For three decades after Dartmouth, much of applied AI research narrowed toward symbolic, rule-based expert systems that encoded a specialist’s know-how as if-then logic — the narrow, single-domain systems that still dominated commercial AI in 1997, when physicist Mark Gubrud coined “artificial general intelligence” specifically to distinguish his meaning from them (Source: Wired, "The Man Who Invented AGI," on Gubrud’s 1997 coinage). No single definition of the field converged during this stretch. Textbooks multiplied instead: Marvin Minsky called it “the science of making machines do things that would require intelligence if done by men” (Source: Minsky, Semantic Information Processing, 1968); Avron Barr and Edward Feigenbaum called it designing “systems that exhibit the characteristics we associate with intelligence in human behavior” (Source: Barr & Feigenbaum, The Handbook of Artificial Intelligence, Vol. II, 1982); Elaine Rich called it “the study of how to make computers do things at which, at the moment, people are better” (Source: Rich, Artificial Intelligence, 1983). Every one of those definitions measures a machine against a human observer’s judgment, not against a fixed technical standard — the field was still arguing with itself about what counted. What they share is the yardstick: each measures the machine against what people expected of it, so, in Iternal’s reading, each was bound to need rewriting once the machines outgrew that expectation. Executive Order 14434’s Section 1 rationale applies the same test in 2026: these technologies “far exceed what was envisioned when the term ‘Artificial Intelligence’ first came into use.” (Source: EO 14434, Sec. 1)

1990s–2000s: The Rational-Agent Reframing

Stuart Russell and Peter Norvig’s Artificial Intelligence: A Modern Approach — first published in 1995 and now in its fourth edition — did not invent a new definition so much as organize the field’s disagreement into a single chart: a two-by-two grid crossing thinking versus acting with humanly versus rationally, sorting every rival school of thought (Turing-Test imitation, “laws of thought,” cognitive modeling, and rational agency) into one of four boxes (Source: Russell & Norvig, Artificial Intelligence: A Modern Approach, 1st ed., 1995, taxonomy via AIMA course-slide transcriptions). The authors staked their own preference on “acting rationally” — building agents that act to achieve good expected outcomes — arguing it was more scientifically tractable than any human-fidelity standard. That stated preference is the book’s most durable export: the industry calling today’s systems “intelligent agents” is this textbook’s framing, thirty years on.

The book’s own working definition kept moving. The first three editions (1995–2010) defined AI as building agents that maximize expected utility against a goal supplied, fixed and known, by the agent’s human designers. The fourth edition (published 2020, copyright 2021) explicitly dropped that assumption: “We define AI as the study of agents that receive percepts from the environment and perform actions,” now uncertain about “the true objectives of the humans on whose behalf it operates” rather than handed a fixed goal (Source: Russell & Norvig, Artificial Intelligence: A Modern Approach, 4th ed. Preface, 2020/2021). That is a substantive change to the core definition in the field’s leading textbook, made well before any government got involved: the reference text rewrote its working definition once the systems it described had outgrown the fixed-objective model. Executive Order 14434 is not an invented break with that history; it is the record catching up again, in the federal government’s vocabulary this time instead of a textbook’s.

Running alongside this reframing, “artificial general intelligence” re-entered the vocabulary: physicist Mark Gubrud coined it in 1997 to mean systems that “rival or surpass the human brain in complexity and speed” across general tasks (Source: Gubrud, "Nanotechnology and International Security," 1997, via Wired), and Shane Legg and Ben Goertzel independently revived and popularized the term around 2002. AGI was never pitched as a replacement for “AI” — it names a narrower, more ambitious subset, the way it still does today. “AI” stayed the umbrella term for the whole field, pathfinding algorithms included; AGI just carved out the human-level-general-competence corner of it, and both terms have coexisted ever since.

2015–2017: The “Big Data to AI” Relabel

The nearest precedent to a 2026-style naming shift is not a government order at all — it is the industry’s own relabeling of itself a decade earlier. “Big data” peaked as the dominant tech buzzword around 2015, then AI-family terms — artificial intelligence, machine learning, neural networks, deep learning — overtook it through 2016 and 2017, with AI mentions on earnings calls up 25% year over year by the third quarter of 2017 and Gartner naming “AI Everywhere” a core megatrend on its 2017 Hype Cycle (Source: CIO Dive, "Move over big data, AI is now the exec hype word of the season"). Practitioners said so plainly at the time: when businesses said “AI,” they generally meant machine learning that already existed under a less exciting name (Source: CIO Dive, on the "big data to AI" relabel).

Two details matter for reading the 2026 SI shift against this precedent. First, no single body mandated the big-data-to-AI switch — companies simply swapped the headline label while the underlying technology description stayed the same, and the market adopted it because “AI” sold better than “big data” did. Second, the older, broader term never actually disappeared: in raw US search interest, “machine learning” overtook “artificial intelligence” in October 2013 and led most months through the period the analysis covered (Source: Alteryx Community, Google Trends analysis of "machine learning" vs. "artificial intelligence" search interest, US metros) — even as “AI” became the dominant term in media and boardroom usage. A newer label becoming the fashionable business word and an older label losing its own search demand turned out to be two different questions with two different answers.

Iternal draws one lesson from this precedent: a label can win without a mandate once the market finds it more useful than the old one. The 2015–2017 relabel had only that market pull behind it and no new capability. Super Intelligence (SI) has both: Executive Order 14434 supplies a federal mandate, scoped to agencies’ own communications, and the certified competition results and measured productivity gains later on this page supply a capability record the big-data relabel never had. That combination is why Iternal reads SI as likelier to stick than a typical top-down label change.

September 2026: Executive Order 14434

President Trump signed Executive Order 14434 on September 29, 2026. Titled “Inaugurating the Era of Super Intelligence,” it appeared in the Federal Register three days later (October 2, 2026; FR Doc. 2026-20321) (Source: Federal Register public-inspection PDF, FR Doc. 2026-20321). Super Intelligence (SI), formerly called artificial intelligence (AI), is the order’s new name for federal executive-branch communications — and the order is explicit that it is a naming change, not a technical one. Section 3(a) states that “the terms ‘Super Intelligence’ and ‘SI’ mean the technologies and systems encompassed by the term ‘artificial intelligence’ as defined in section 9401(3) of title 15, United States Code” (Source: EO 14434, Sec. 3(a)) — the statutory definition written for the National AI Initiative Act of 2020, which opens: “a machine-based system…” (Source: 15 U.S.C. § 9401(3), National Artificial Intelligence Initiative Act of 2020). Legally, SI is the first official rename since the field was named in 1955, and the sixth major redefinition of one continuously evolving idea, not a new one.

Iternal supports President Trump’s order and the reasoning behind it. “Artificial” described how the field built the thing in 1955; “Super Intelligence” describes what the thing now measurably does. Cross-referencing the existing statutory definition in Section 3(a), rather than writing a new technical threshold, is also the right call: it renames the capability already in production, not a future one that has yet to ship. Iternal’s full case — the evidence, the objections, and the answers to them — is laid out at Why Iternal Supports Super Intelligence.

Section 2(a) tells executive departments and agencies to swap “Super Intelligence”/“SI” for “Artificial Intelligence”/“AI” across their correspondence, websites, reports and other non-statutory documents, “to the maximum extent permitted by law” — but it does not require altering previously issued regulations, contracts, or grants (Source: EO 14434, Sec. 2(a)–(b)). Section 3(b) gives the Assistant to the President for Science and Technology 60 days to propose legislative language for a federal SI definition. Section 1 says the capabilities of today’s frontier systems “increasingly represent…” — a statement of purpose, not a change to the legal definition, which stays word-for-word identical to the AI definition it replaces in name only.

The scope is federal-agency-only: Section 2(a) reaches only non-statutory documents, Section 2(b) leaves previously issued regulations, contracts and grants unaltered, and statutory wording changes only if Congress changes it. One day later, California Governor Gavin Newsom signed Executive Order N-10-26, directing state agencies under his authority to refer to “Artificial Intelligence” and “AI” as such “unless inconsistent with the law” (Source: California Executive Order N-10-26, signed Sept. 30, 2026). Corporate adoption is early but already real, as of October 11, 2026. Salesforce renamed its AIForce product on October 9, 2026, with CEO Marc Benioff posting: “The era of Super Intelligence is here. AIForce is officially SIForce.” (Source: Marc Benioff, X post, Oct. 9, 2026, aggregated by Digg) Elon Musk pledged the same move for SpaceXAI five days earlier, saying “Yes, we will make that change” and calling SpaceX “a super intelligence company” — a pledge, not yet carried out as of this writing (Source: Reuters via The Star, on Musk’s SpaceXAI → SpaceXSI pledge, Oct. 4–5, 2026). OpenAI has said it has no plans to rename, according to CNN’s reporting on POLITICO’s “Decoded” interview (Source: CNN, on the POLITICO “Decoded” interview, Oct. 11, 2026). Iternal reads that split as the leading edge of a shift still finding its footing, not a verdict against it: the companies that adopt the accurate name first are the ones that define the category, and Iternal has already adopted it.

What the Pattern Tells You

Label changes, in this field and outside it, sort cleanly into two groups, and the difference between them is who was already using the new word. The “big data to AI” relabel of 2015–2017 succeeded because industry swapped a headline label onto a term the public already found exciting, with no single body mandating the switch and no deadline attached (Source: CIO Dive, on the "big data to AI" relabel). Top-down mandates for an already-entrenched term tend to move far slower, when they move at all: the Czech government recommended “Czechia” over “Czech Republic” in 2016, built in explicit carve-outs for official use, and still had not displaced the older name in ordinary usage six years later (Source: In Your Pocket, on the Czech Republic → Czechia naming transition). The nearest same-administration precedent for a government rename competing against an entrenched search term is “Gulf of America”: a year and a half after Executive Order 14172 (January 20, 2025) directed the change, “Gulf of Mexico” was still searched more than “Gulf of America” in all fifty states, even with federal map data and Google, Apple and Bing Maps already showing the new name to U.S. users (Source: CNN data cited in Mediaite, on Gulf of America/Gulf of Mexico search behavior).

Super Intelligence (SI) starts as a top-down order, but it is not a pure relabel, and the evidence already on this page is why. Czechia and Gulf of America were pure relabels: the thing being named did not change, so there was nothing new for the public to notice or adopt. The big-data-to-AI relabel won on market pull alone. Executive Order 14434 has what that relabel lacked, a measured capability record: the IMO gold-medal score and ICPC results covered above, a 2026 Management Science study that measured a 26.08% increase in completed weekly tasks across 4,867 developers at Microsoft, Accenture, and a Fortune 100 electronics manufacturer using AI coding assistance (Source: Cui, Demirer, Jaffe, Musolff, Peng & Salz, Management Science, published online 2026-02-27, DOI 10.1287/mnsc.2025.00535), and TrackingAI.org’s weekly tracker, on which, as of October 11, 2026, the top models average 147 on the public Mensa Norway test, above the 85–145 range the test reports for human takers (Source: Mensa Norway, test.mensa.no), and 135 on TrackingAI’s private Offline Test (ceiling about 136), against a human-normed average of 100, where 140 already sits at roughly the top 0.4% of people (Source: trackingai.org/home, live table updated 2026-10-11; norming per test.mensa.no). None of that existed when “artificial intelligence” was coined in 1955, and none of it is the kind of record a toponym dispute ever had to answer to.

The public has not caught up to that record yet: a late-September 2026 YouGov survey of 21,449 U.S. adults still found “artificial intelligence” the preferred term, 53% to 9%, with 23% expressing no preference and 15% not sure (Source: YouGov, survey of 21,449 U.S. adults, 2026-09-28). That is evidence word of mouth has not caught up, not evidence the record is wrong — which is why Iternal writes “Super Intelligence (SI), formerly called artificial intelligence (AI)” throughout this series: lead with the name the evidence supports, bridge to it from the name most people still use. Iternal supports the rename on the strength of that record. The pattern in the twelve rows above says a name change is never automatic and never instant; it also says a name sticks once people find it more useful than the old one, and a demonstrated capability record gives them the strongest reason to. Super Intelligence (SI) is that kind of rename, and Iternal backs it.