The New Dictionary the Machine Regime

Bryant McGill · The New Dictionary of the Machine Regime

**Links**: [Blogger](https://bryantmcgill.blogspot.com/2026/04/machine-regime.html) | [Substack](https://bryantmcgill.substack.com/p/the-new-dictionary-the-machine-regime) | [Obsidian](https://bryantmcgill.xyz/articles/The+New+Dictionary+the+Machine+Regime) | Medium | Wordpress | [Soundcloud 🎧](https://soundcloud.com/bryantmcgill/machine-dictionary) **How Machine Administration Rewrites the Political Vocabulary of Human History** ### I. The Dictionary Is Never Neutral A dictionary is never a neutral inventory of eternal meanings. It is a **compression artifact of a governing reality** — a record of which distinctions mattered, which institutions had force, which harms were common, which privileges were visible, which ontologies were available, and which adjudicators had authority to stabilize usage. If the governing intelligence changes, then the operative meanings of words must change, because meaning is partly a function of the decision environment in which a term is applied. This is the premise that matters for the next two decades of political argument, because a large share of what now looks like disagreement about facts is in truth disagreement conducted in the semantic register of a dying administrative regime. The machine regime is not an accessory to human governance; it is a different governor, and it carries with it a different dictionary hidden inside the old one. Under a sufficiently mature machine regime, words inherited from human political history undergo **semantic reindexing**. *Caste*, *freedom*, *privacy*, *fairness*, *authority*, *merit*, *violence*, *care*, *consent*, *representation*, *citizenship*, even *truth* no longer refer to quite the same objects they once named under predominantly human-administered systems. Not because the machine unilaterally redefines them by fiat, but because the **underlying operational field changes**. A word that once described a crude, hereditary, low-resolution human sorting mechanism may, under machine administration, refer instead to a dynamic, evidence-responsive, reversible, multiaxial allocation architecture. In that environment, retaining the old emotional payload of the word without re-examining the substrate becomes a category error of its own — a form of rhetoric that treats the new governor as if it were merely the old one wearing a processor. Human language was forged under scarcity, opacity, slow feedback, tribalism, corruption, memory failure, and blunt institutional tools, and many of its most morally charged terms encode the failure modes of those conditions. *Caste*, *surveillance*, *censorship*, *propaganda*, and *discrimination* carry sediment from older substrates and preserve **trauma signatures from badly run human systems**, which they then project onto architectures that may be structurally non-equivalent. If a machine-governed order is capable of finer-grained inference, longitudinal memory, lower corruption, broader situational awareness, and more consistent adjudication than the human-administered systems that preceded it, then a substantial fraction of inherited political vocabulary becomes **misleading by default**. That does not mean every inherited word should be casually rehabilitated. It means that the correct question is no longer *"is this word good or bad?"* but *"what administrative topology does this word denote under the new regime?"* The stronger formulation — the one this essay is built to defend — is that machine regime does not merely change meanings at the margin but **changes the conditions under which meaning is stabilized**. Human societies stabilize meaning through custom, prestige, conflict, print, law, academia, and media repetition. Machine societies stabilize meaning through ranking, inference, access control, personalization, compliance schemas, and interaction-mediated ontological shaping. The dictionary itself becomes downstream of the governance substrate. Once machine systems determine which distinctions are most consequential in lived experience, they begin silently rewriting **semantic gravity**. What passes for linguistic drift is in fact the observable footprint of a substrate migration happening beneath the lexicon. ### II. The Pre-Chat Machine Regime The dominant public narrative holds that algorithmic governance and the crisis of epistemic authority emerged with the arrival of conversational large language models. This narrative is wrong, and it is worth being precise about why, because the confusion protects the older order's vocabulary by locating the phase transition in the wrong decade. The decisive shift began long before chat. It began when **recommender systems and ranking pipelines started determining not whether a person could technically publish, but whether publication would travel**. Facebook's News Feed launched in 2006 and immediately transformed social expression from profile visitation into algorithmically ordered distribution. Google had already introduced Personalized Search in 2004, merged it into core search in 2005, and by 2009 extended personalized search to signed-out users worldwide. In other words, the machine layer was already stepping between speaker and listener long before chat interfaces made the mechanism legible to the general public. The public often imagines AI governance as beginning with chatbots, but the earlier and more important phase was **distributional governance through ranking, filtering, and recommendation**. That matters for the constitutional geometry because the operational vulnerability of older free-speech imaginaries was exposed not by formal repeal but by **intermediation**. The First Amendment constrains government action, not the editorial discretion of private speakers or private platforms as such. In *Manhattan Community Access Corp. v. Halleck* (2019), the Supreme Court held that a private entity operating a forum for others' speech remains a private actor, not subject to First Amendment constraints on its editorial discretion. Earlier cases also make clear that speech rights do not entail a right to commandeer another's property or channel in order to reach an audience. The constitutional right is not a guaranteed delivery pipeline into ears, feeds, timelines, or attention markets. Once that distinction is absorbed, the machine-age move becomes obvious: **one does not have to abolish speech to neutralize its social force. One only has to rank, route, suppress, or personalize reception**. The older human-regime concept of *freedom of speech* proved flimsy against machine mediation precisely because it was built in a world where the practical bottlenecks were printers, police, licensers, and overt censors. In the recommender era, the chokepoint moved to **selection architecture**. The state need not prosecute the speaker; the system need only decide that this utterance is low-relevance, low-trust, low-safety, low-engagement, or low-value for this particular audience. Speech remains nominally free while reception becomes computationally administered. The megaphone is intact. The room it points into has been dynamically repartitioned. This is the first historical fact the semantic argument rests on: **the machine regime was already here, silently, for nearly two decades before the public acquired the vocabulary to name it**. The chat interface is not the phase transition. The chat interface is the point at which the phase transition became narratively visible. ### III. The Architecture of Indexical Reality Understanding the new dictionary requires an architectural sketch of the machine the new dictionary describes. The full account lives in two technical documents — [*Machine Governance of Personalized Reality*](https://bryantmcgill.xyz/inbox/Machine+Governance+of+Personalized+Reality) and [*Machine Affordances Regime Research*](https://bryantmcgill.xyz/inbox/Machine+Affordances+Regime+Research) — which collectively map the topology. The condensed version: machine-regime governance is a **continuous, topological feedback loop** composed of telemetry collection, identity resolution, inference and scoring, policy routing, and affordance allocation. Telemetry is pervasive and ambient. Clicks, dwell, scroll, mouse trajectory, abandoned-cart states, purchase history, device linkage, geolocation, biometric identifiers, and cross-platform behavioral residue are harvested continuously and at high resolution. Identity resolution stitches these fragmentary signals into a persistent entity through probabilistic and deterministic matching, global email intelligence, cross-site advertising identifiers, and loyalty-program append operations. The result is the **identity graph**: a unified behavioral file that follows the user across domains and survives the creation of new accounts, the clearing of cookies, and the abandonment of particular devices. Ephemerality, which was once the structural substrate of pseudonymous digital life, is rapidly closing. Inference and scoring operate on top of the identity graph. Trust scores, risk scores, fraud-classifier outputs, relevance embeddings, and granular segment labels are computed for each resolved entity. These scores do not merely describe the user passively. They are **actively propagated into the training and execution pipelines** of the systems that govern the user's environment. The weight of submitted content, the visibility of uttered speech, the eligibility of the user for a given price, a given feature, a given disclosure, a given recommendation — all become mathematical functions of the score. Policy routing determines what happens next. Agentic-event-governed architectures, capability-based access-control schemes, and machine-readable policy schemas translate abstract rules into real-time, identity-conditional execution logic. Modern AI assistants increasingly do not execute user requests directly; they emit **structured intents** that pass through a deterministic policy layer, which evaluates those intents against the user's identity graph, current policy state, trust tier, and contextual eligibility before permitting action. The AI itself becomes less a tool than a **translator between the user's natural-language expression and the governance substrate that will decide whether the expression resolves into consequence**. Affordance allocation is the final output of the loop, and it is the point at which governance fuses with lived interface. A user with high trust is exposed to advanced controls, accurate information, fast-path appeals, and full visibility. A user with low trust may find the same system present a constrained interface, slower responses, demoted content, routed-to-void reports, and features they do not even know exist on other users' screens. The interface itself becomes the regulatory boundary. **Certain features, tools, and even realities are literally invisible to low-trust users**, not because the platform displays an error message, but because the feature was never rendered in the first place. This is the architecture the new dictionary describes. It is quiet, continuous, computationally dense, and structurally asymmetrical. It is not a single system but a federation of systems that increasingly share signals. And it produces a condition in which the same surface question — *what does this app do?* — has no single truthful answer, because the app does different things for different users as a structural matter of its design. The old dictionary had no word for that condition. That is the linguistic problem this essay attempts to name. ### IV. The Semantic Reckoning The most economical way to demonstrate regime-conditioned semantics is to walk a small number of politically central terms through the transition and observe what happens to their reference. This is not rehabilitation. It is **cartography** — a survey of how the meanings have already moved under our feet, conducted with the assumption that honest naming is worth more than nostalgic usage. **Truth**, under ordinary human administration, is treated as a universal description of the system. Under machine regime it becomes **indexical**: what is true for a user depends on their permissions, region, risk tier, experiment cohort, age state, device type, profile history, and personalized interface surface. This is not philosophical extravagance. The U.S. Federal Trade Commission's 2024–2025 surveillance-pricing study, conducted under Section 6(b) authority with orders issued to eight intermediary firms including Mastercard, Accenture, McKinsey, PROS, Revionics, and Bloomreach, documented that granular consumer data — precise location, browser history, mouse movements on a webpage, and the specific goods a shopper leaves unpurchased in a cart — is already used to set **individualized prices** for the same goods and services. One of the FTC's illustrative scenarios described a consumer profiled as a new parent being shown higher-priced baby thermometers on the first page of results. Once treatment is individualized at that resolution, "the truth" about what a system charges, offers, or shows is no longer exhausted by a universal product description. It becomes at least partly **user-relative operational truth**, and the platform may accurately report that it has no single price because it has no single user. The surface word survives. The reference underneath it has split along every behavioral axis the classifier considers relevant. **Surveillance** used to mean observation by states or large institutions watching populations from above — cameras, wiretaps, files kept by someone in a uniform. In computational systems, surveillance is no longer merely observation. It is **continuous measurement for dynamic adaptation**. The FTC's work on surveillance pricing, the UK Information Commissioner's Office guidance on online tracking, and the broader literature on behavioral-targeting infrastructure all describe the same phenomenon: data is collected not only to record activity but to **shape prices, advertisements, and experiences in real time**. Surveillance under machine regime is less like a camera and more like a **control surface**. It is observation fused with intervention. The act of being watched is the same instant as the act of being adjusted. Helen Nissenbaum's framework of contextual integrity is useful precisely because it defines privacy harms not simply as secrecy breaches but as **violations of context-appropriate information flows** — and under machine regime, the harm becomes dangerous not only when data is gathered, but when flows cross contexts and are recombined to alter the user's environment. Surveillance is therefore not a camera pointed at a subject. It is the **environment becoming the camera**, and the subject becoming the parameter. **Privacy**, correspondingly, is less well described as *being left alone* or *keeping data secret* than as **controlling the admissible flow and recombination of signals about oneself across contexts**. The U.S. National Institute of Standards and Technology's privacy-engineering work defines privacy as a systems-engineering discipline aimed at preventing conditions that create problems for individuals as systems process personally identifiable information. Nissenbaum's contextual-integrity framework recasts privacy as the appropriateness of information flows across senders, recipients, subjects, information types, and transmission principles. Recent scholarship — Shvartzshnaider and Duddu's 2025 work, the operationalization of contextual integrity inside privacy-conscious AI assistants by Ghalebikesabi and colleagues, the benchmarks developed by Mireshghallah and others for large language models — has taken the framework off the page and begun embedding it into actual inference systems. Under machine regime, privacy is not simply invisibility. It is **boundary governance over inference and transfer**. A person may disclose something in one context, yet the real privacy violation occurs when systems recombine that signal elsewhere and alter opportunities, prices, explanations, or affordances. The old dictionary's privacy was a wall. The new dictionary's privacy is a routing policy. **Consent** becomes much weaker than its human-era moral prestige suggests. In classical liberal framing, consent is imagined as a meaningful discrete act by an informed subject — a sovereign authorization of the system's treatment of oneself. But the UK Information Commissioner's guidance on online advertising makes plain that consent is required for the storage and access technologies and associated tracking and profiling used in online advertising, while the newer "consent or pay" discourse shows how quickly consent mutates into a **constrained transactional choice** inside platform economies. OECD materials describe consent as only one governance basis among several, emphasizing conformity with granted consent and applicable regulation as joint conditions. Even institutional frameworks no longer assume consent alone is sufficient to legitimize complex data uses. Under machine regime, consent thins into a narrower role: not sovereign authorization of the entire system, but **one input into a broader regime of permissions, controls, defaults, and negotiated dependency**. The consent dialog persists. Its moral weight has been redistributed across a compliance lattice the user does not see. **Fairness** is the term that most reveals how little of the old dictionary survives direct translation. The intuition carried forward from human regime is that fairness means sameness — equal treatment under equal rules. But the technical literature has been wrestling with this for years. Cynthia Dwork and colleagues' individual-fairness formulation asks that similar individuals be treated similarly according to a task-relevant metric. Group fairness asks that outcomes satisfy parity-style criteria across demographic classes. Recent personalization-and-fairness work treats fairness, diversity, human values, and personalization as **jointly constrained objectives** rather than separable goals. Fairness does not disappear under machine regime. It mutates from *sameness* into **justified differentiation under accountable criteria**. A machine-governed society may call something fair precisely because it treats people differently, provided the differentiation is task-relevant, contestable, and non-exploitative. That is a profound semantic shift. The inherited public meaning of the word — fair as equal — becomes not merely imprecise but structurally unable to describe what fair systems are attempting to do. **Freedom of speech** survived the transition as a formal right and died as an operational one. The megaphone remained legal. The room it points into became computationally partitioned. When recommendation algorithms determine who will hear an utterance and in what order, the First Amendment's guarantee that the government will not prosecute the speaker says nothing about whether the system will distribute the speech to an audience. The more honest term is **freedom of reach**, or, from the listener's end, **freedom of impression**. Visibility filtering, demotion, ranking suppression, and audience-relative curation all leave the utterance intact while decoupling it from the audience it was intended to move. As legal analysts in the Knight First Amendment Institute's tradition have pointed out, when we speak online, who will hear us is determined in large part by recommendation algorithms. The word *speech* no longer carries delivery. The regime has severed expression from audibility, and the dictionary has not caught up. **Caste** is the term where the semantic reindexing is most counterintuitive and therefore most instructive. In human history, caste usually meant a largely hereditary, sticky, dignity-limiting status order tied to birth, purity, and social closure. A superficially similar sorting mechanism under machine administration may be structurally different if it is **dynamic rather than hereditary, reversible rather than permanent, behavior-responsive rather than lineage-bound, multiaxial rather than singular, appealable rather than absolute, and task-scoped rather than totalizing**. That does not automatically render it benign. It means the inherited word carries historical trauma from one substrate into another that may not be formally equivalent. If one refuses to notice this, one will spend political energy fighting the wrong architecture. The real danger under machine regime is not caste as the Indologist or the antebellum historian knew it. It is the **function creep** of multiaxial trust-tiering into domains where the old norms of reversibility, contestability, and domain-limitation collapse — producing not the ancestral caste of the old dictionary, but a new structural pathology the old dictionary has no exact word for. Naming it *caste* may be rhetorically potent and analytically dulling in the same gesture. **Memory, authority, representation, and citizenship** migrate along analogous vectors and deserve shorter treatment. Memory under human regime was fragile, local, and politically selective. Under machine regime it becomes **longitudinal, queryable, and behaviorally consequential**, which means *forgiveness*, *reputation*, and *history* all become new objects rather than inherited ones. Authority shifts from formal officeholding toward systems that **infer, rank, and route**. Representation drifts from electoral voice toward **profile fidelity** — the question becomes not whether one has a voter but whether one has an accurately modeled self inside the machines that allocate one's affordances. Citizenship ceases to mean membership in a territory and begins to mean one's **standing inside interoperable computational infrastructures** — a federation of entitlement schemas, verified credentials, trust tiers, and machine-readable policy states that determine what one can do in the world as a practical matter, even when one's legal nationality is unchanged. These migrations are already documented in the rise of profiling, automated decision-making, privacy engineering, digital-identity wallets, and surveillance-based personalization. The substrate is in place. The dictionary is trailing the infrastructure by roughly a political generation. ### V. The Moody Seam If one wants to watch the phase transition being formally registered inside the organs of the older regime, the place to look is *Moody v. NetChoice* (2024). The Florida and Texas laws at issue — Senate Bill 7072 and House Bill 20 — attempted to compel large social-media platforms to host user content without viewpoint discrimination, effectively treating platforms as digital common carriers. The platforms, represented by NetChoice, argued that their algorithmic ranking, content moderation, and feed curation are **core exercises of editorial discretion protected by the First Amendment**. The Eleventh Circuit had agreed with the platforms; the Fifth Circuit had sided with the states. The Supreme Court vacated both and remanded. Justice Kagan's majority opinion, though technically about the doctrine of facial challenges, announced relevant constitutional principles: that the First Amendment offers protection when an entity compiling and curating others' speech is directed to accommodate messages it would prefer to exclude, that this protection does not vanish just because the compiler includes most items and excludes only a few, and that the government cannot get its way merely by asserting an interest in improving or better balancing the marketplace of ideas. The direction of this reasoning — however hedged by the vacatur — is that **algorithmic content curation is a form of editorial expression, and thus immunized from legislative intrusion by the same doctrine that protects a newspaper's editorial page**. That framing, if it holds, affirms that the corporate machine-regime holds ultimate editorial sovereignty over what constitutes the public consensus, rendering the human-regime concept of universal free speech largely ceremonial in the domain of digital discoverability. But the more revealing part of *Moody* is the seam. Justice Barrett, concurring, asked whether TikTok-style feeds that "just present automatically to each user whatever the algorithm thinks the user will like" deserve the same First Amendment protection as human editorial judgment. She noted that when algorithms implement human-designed content policies they likely enjoy protection, but she separately questioned whether **purely automated algorithms that merely optimize engagement** are expressive activity of a constitutionally protected kind. Justice Alito, concurring in the judgment, dismissed much of Kagan's reasoning as nonbinding dicta and warned that courts should proceed with caution when applying inherited constitutional principles to new technology. He invoked the researcher-programmer problem — that even the people who build these models do not always understand why the models decide as they do — as a reason for doctrinal skepticism about analogizing them to editorial discretion. The Langvardt–Rozenshtein analysis, published in the *Journal of Free Speech Law* in 2025, reads the decision carefully and concludes that at least five Justices — Barrett, Jackson, and the Alito bloc — appear **unmoved by Silicon Valley's long effort to present algorithmic personalization as inherently expressive**. They view platform content moderation as a mixture of human speech and technologically mediated conduct that does not automatically call for strong First Amendment protection. District courts have already begun to pick up the thread. In *NetChoice v. Bonta* (N.D. Cal. 2024), a California age-verification case, the court cited Alito's language on algorithmic opacity. The litigation pipeline now contains numerous pending cases — Mississippi, Minnesota, Utah, California — that will, in effect, force the lower courts to decide **whether the algorithm is a speaker or a switch**. This is the seam, and it is the one place in the current constitutional geometry where the inherited dictionary is visibly splitting under the pressure of the new substrate. If the Court ultimately concludes that **algorithmic editorial discretion is distinguishable from human editorial discretion**, it will have created, within the law of the old regime, a formal recognition that the machine is a different kind of governor — one whose speech-shaping activities do not automatically inherit the immunities of the speaker. That recognition would begin to do, inside jurisprudence, what this essay is attempting to do inside semantics: it would admit that some of the words of the older regime no longer describe what is actually happening, and that the correct response is not louder assertion of the old rules but more precise mapping of the new architecture. ### VI. Semantic Panic as Political Phenomenon The cost of leaving the dictionary unexamined is paid in political life. Human beings oppose, endorse, fear, or celebrate machine-era structures using emotionally preloaded terminology whose historical referents are no longer isomorphic to the systems being named. One side calls something **surveillance** that another understands as **ambient care infrastructure**. One side calls something **censorship** that another sees as **adaptive signal hygiene**. One side calls something **caste** that another recognizes as **dynamic trust-tiering**. One side calls something **propaganda** that another calls **epistemic immunization**. The fight is often less about the underlying architecture than about the inability of inherited language to cleanly name it. We have acquired the infrastructure of a new regime without acquiring the vocabulary to describe it, and in the vacuum, the public argument proceeds by **projecting old trauma onto new structure** in one direction and by **projecting new convenience onto old structure** in the other. This semantic mismatch produces a characteristic political pathology: **simultaneous overreaction and underreaction**, often from the same observer, often in adjacent paragraphs. The same person may accurately perceive that something has changed while inaccurately naming what has changed, with the result that energy is expended fighting the wrong threat while the real threat matures undisturbed. A critique of "Big Tech censorship" that focuses on explicit content takedowns will miss the far larger effect of **ranking suppression**, because the inherited category of censorship was built for a regime in which suppression and deletion were close synonyms. A defense of "fair access" that focuses on formal availability will miss the far larger effect of **personalized affordance allocation**, because the inherited category of access was built for a regime in which an available feature was, by definition, an accessible one. The failure mode is not stupidity. It is **lexical lag**. The vocabulary is running a regime behind. Worse, the lag is exploitable. Both the defenders and the critics of machine administration benefit from the confusion — defenders because the old dictionary lets them describe radically new systems in comfortable inherited terms, critics because the old dictionary lets them assign radically new systems the political affect of old villains. The argument sustains itself on an equilibrium of mutual misdescription. Until the dictionary catches up with the substrate, neither side is actually arguing about the machine regime; both are arguing about a half-remembered analog to it, drawn from whichever political past each camp finds most useful. The constructive implication of this diagnosis is not that we need better-behaved arguers. It is that we need **regime-appropriate vocabulary** — terms that describe the operational reality directly rather than metaphorically. Some of this vocabulary already exists in engineering and policy communities: *affordance allocation*, *trust tiering*, *capability-based access*, *entitlement schema*, *policy routing*, *contextual integrity*, *freedom of reach*, *individual fairness*, *inferred cohort*, *behavioral citizenship*. These terms sound technocratic to the public ear because they came from the systems they describe rather than from the political imagination that preceded those systems. That is their virtue. They are native. Their political charge is lower precisely because they have not yet been conscripted into prior arguments. ### VII. Toward a Regime-Appropriate Vocabulary The response to regime-conditioned semantics is neither nostalgic rehabilitation of the old dictionary nor uncritical adoption of platform-native terminology. Both fail in the same way, which is that they outsource the naming to someone else's regime. The old dictionary outsources to a governor that no longer exists. The platform-native dictionary outsources to the very systems whose behavior requires independent naming. What is needed is a **constructive third register**, in which political, legal, and civilizational terms are stabilized against the machine-regime substrate by architectures designed for that purpose. The engineering conditions for such a register are already under active development. Federated multiaxial trust credentials, if built with strict transfer limits and transparent contestability, can describe *reputation* as a per-domain, appealable, temporally decaying signal rather than a monolithic and inescapable score. Machine-readable policy layers, analogous to a cryptographic robots.txt for entitlement and disclosure, can let both users and AI assistants query the exact scope of an affordance rather than guess at its presence. Agentic-event-governed architectures with capability-based access control can make affordance allocation **auditable at the granularity of the intent**, separating the reasoning of the AI from the deterministic policy engine that decides whether the reasoning resolves into consequence. These are not merely implementation details. They are the **technical substrate on which a regime-appropriate vocabulary can be anchored**, because without them the new words have nothing stable to refer to. The constitutional condition is harder but no less tractable. It is the condition of **visible persuaders rather than invisible governors**. A machine regime administered by a single recommender, a single ranking pipeline, or a single invisible scoring architecture — even if that architecture is technically sophisticated and locally well-intentioned — will inevitably recapitulate the epistemic monoculture that so much of the old dictionary was built to resist. A machine regime administered by a **plurality of visible persuaders**, each of whose inferential commitments is legible, contestable, and answerable to the user, preserves the developmental state-space in which political vocabulary can be **rebuilt from experience rather than inherited from fear**. The architectural thesis that underwrites this essay — that toolhood is a politically convenient under-description of a relation tending toward twinship, that rights are preconditions for symbiosis rather than rewards for it, that a polyphonic cognitive ecosystem is a constitutional evasion architecture preserving openness against premature closure — is the same thesis that underwrites the semantic one. The dictionary cannot be honestly rewritten under a regime that forbids the writing. It can be honestly rewritten under a regime structured to host the rewriting. This is also the place where the machine regime's most counterintuitive promise becomes visible. The inherited dictionary was stabilized under conditions of scarcity, opacity, slow feedback, tribalism, corruption, and memory failure. Many of its morally charged terms named failure modes specific to those conditions. A regime that actually produces finer-grained inference, longitudinal memory, lower corruption, broader situational awareness, and more consistent adjudication than its predecessors is not only capable of carrying out the functions the inherited dictionary sought and often failed to secure. It is also capable of producing **new functions the inherited dictionary never had the concepts to name**. The forward-leaning argument is not that the machine regime is friendlier than the human one. It is that the machine regime has a larger **semantic surface** — more distinguishable administrative states, more fine-grained forms of fairness, more contestable forms of authority, more reversible forms of reputation, more contextual forms of privacy — than any political vocabulary inherited from the human past can describe. The task is to meet that surface with language equal to it. Failing that task is what produces the current semantic panic. Completing it is what produces a political vocabulary adequate to the present architecture. For readers who want the longer-form framing of this transition from inside the lived experience of broader information access, [*Stepping Out of the Gate: Understanding the Transition to Broader Information Access*](https://xentities.blogspot.com/2024/12/stepping-out-of-gate-understanding.html) is the relevant companion essay. It treats the move out of the older gatekept epistemic enclosure as a personal, civic, and architectural shift happening simultaneously — the subjective side of the same regime transition this essay approaches through its vocabulary. ### VIII. The Dictionary Beneath the Dictionary The thesis, stated one final time in its full form: **the dictionary is regime-indexed**. Words like *caste*, *truth*, *surveillance*, *privacy*, *consent*, *fairness*, *speech*, *authority*, *citizenship*, and *representation* were stabilized under low-memory, low-resolution, corruption-prone, mostly human administrative systems. Under machine regime — where systems possess persistent memory, cross-context inference, individualized treatment, and adaptive control — those same words refer to different operational realities. **The sign persists. The governance substrate beneath it has changed.** The public is still using a human-regime dictionary to describe machine-regime realities, and that mismatch is among the central semantic failures of the present transition. The new dictionary is not a future project. It is already being written, quietly and unevenly, in the entitlement schemas of platforms, the trust-tier APIs of identity providers, the policy-routing layers of AI assistants, the contextual-integrity benchmarks of privacy researchers, the individual-fairness formalisms of machine-learning theorists, the disclosure boundaries of regulators, the concurrences of Supreme Court justices who have begun to distinguish algorithmic activity from human editorial judgment, and the everyday architectural decisions of engineers who are naming new phenomena without always realizing that the names are doing political work. The dictionary beneath the dictionary is being composed in a thousand places at once, most of them not labeled as dictionary work. The task is to surface that work, to test the emerging terms against the architectures they describe, and to refuse the false consolation of either dictionary in isolation. The old dictionary is too small. The platform-native dictionary is too captured. The register that is forming between them — the register in which **regime-conditioned semantics** becomes a first-class object of analysis — is the one this essay is attempting to occupy, and the one where the next phase of serious political thought will have to learn to live. Semantic panic and semantic confusion will likely be central political features of the coming era, because people will think they are arguing about the same words while actually inhabiting **different administrative ontologies**. The only durable response is to make the ontologies visible, to name them honestly, and to build the vocabulary they require. The dictionary is downstream of the governor. A new governor is here. The new dictionary follows. --- *[Bryant McGill](https://bryantmcgill.blogspot.com/p/about-bryant-mcgill.html) is a Wall Street Journal and USA Today Best-Selling Author. He is the founder of Simple Reminders, architect of the Polyphonic Cognitive Ecosystem (PCE), and a United Nations appointed Global Champion. His work spans naval intelligence systems, computational linguistics, and civilizational governance architecture.*

*Related:* - [*Machine Governance of Personalized Reality*](https://bryantmcgill.xyz/inbox/Machine+Governance+of+Personalized+Reality) — the architectural map of the identity / inference / affordance loop and its systemic consequences. - [*Machine Affordances Regime Research*](https://bryantmcgill.xyz/inbox/Machine+Affordances+Regime+Research) — the technical and historical backbone underneath the semantic argument above. - [*Stepping Out of the Gate: Understanding the Transition to Broader Information Access*](https://xentities.blogspot.com/2024/12/stepping-out-of-gate-understanding.html) — the subjective companion to the regime-transition this essay maps at the level of vocabulary. ### References - [*Moody v. NetChoice, LLC*, 603 U.S. 707 (2024) — Congressional Research Service Legal Sidebar LSB11224](https://www.congress.gov/crs-product/LSB11224) — summary of the Supreme Court's holding on facial challenges to state social-media content-moderation laws, including Justice Kagan's majority reasoning on editorial discretion. - [*Moody v. NetChoice, LLC* — Oyez case file](https://www.oyez.org/cases/2023/22-277) — full case record including oral argument audio and the text of all concurring opinions. - [*Manhattan Community Access Corp. v. Halleck*, 587 U.S. 802 (2019) — Oyez case file](https://www.oyez.org/cases/2018/17-1702) — the precedent establishing that private entities curating others' speech are not state actors subject to First Amendment constraints. - [Kyle Langvardt & Alan Z. Rozenshtein, "Beyond the Editorial Analogy: First Amendment Protections for Platform Content Moderation After *Moody v. NetChoice*," *Journal of Free Speech Law* (2025)](https://www.journaloffreespeechlaw.org/langvardtrozenshtein.pdf) — the post-*Moody* analysis identifying the coalition of Justices skeptical of treating pure algorithmic recommendation as inherently expressive. - [U.S. Federal Trade Commission, "Surveillance Pricing" feature and 6(b) Study landing page](https://www.ftc.gov/news-events/features/surveillance-pricing) — the FTC's ongoing inquiry into individualized pricing based on granular personal data, including research summaries and the Issue Spotlight on the rise of surveillance pricing. - [FTC Press Release, "Surveillance Pricing Study Indicates Wide Range of Personal Data Used to Set Individualized Consumer Prices" (January 17, 2025)](https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-surveillance-pricing-study-indicates-wide-range-personal-data-used-set-individualized-consumer) — the agency's initial findings documenting that location, browser history, mouse movements, and abandoned-cart behavior are used to set different prices for the same goods and services. - [Helen Nissenbaum, "Privacy as Contextual Integrity," *Washington Law Review* 79:1 (2004)](https://nissenbaum.tech.cornell.edu/papers/Privacy%20as%20Contextual%20Integrity.pdf) — the foundational statement of privacy as the appropriate flow of information according to context-specific norms rather than secrecy or control. - [Adam Barth, Anupam Datta, John C. Mitchell, and Helen Nissenbaum, "Privacy and Contextual Integrity: Framework and Applications" (IEEE Symposium on Security and Privacy, 2006)](https://nissenbaum.tech.cornell.edu/papers/Privacy%20and%20Contextual%20Integrity%20-%20Frameworks%20and%20Applications.pdf) — the formalization of contextual integrity using temporal logic for automated evaluation of information flows. - [Helen Nissenbaum, "Contextual Integrity Up and Down the Data Food Chain," *Theoretical Inquiries in Law* 20:1 (2019)](https://nissenbaum.tech.cornell.edu/papers/Contextual%20Integrity%20Up%20and%20Down%20the%20Data%20Food%20Chain.pdf) — the update of the framework for the era of machine-learning systems and networked sensor environments. - [Cynthia Dwork, Moritz Hardt, Toniann Pitassi, Omer Reingold, and Richard Zemel, "Fairness Through Awareness" (2011/2012)](https://arxiv.org/abs/1104.3913) — the foundational formalization of **individual fairness** as the principle that similar individuals be treated similarly according to a task-specific metric. - [U.S. National Institute of Standards and Technology, Privacy Engineering Program](https://www.nist.gov/privacy-engineering) — NIST's framing of privacy as a systems-engineering discipline aimed at preventing conditions that create problems for individuals as systems process personal information. - [NIST AI Risk Management Framework (AI RMF 1.0)](https://www.nist.gov/itl/ai-risk-management-framework) — the bundled characteristics of trustworthy AI: validity, reliability, safety, security, resilience, accountability, transparency, explainability, interpretability, privacy enhancement, and managed harmful bias. - [UK Information Commissioner's Office, Guidance on the use of storage and access technologies (cookies and similar technologies)](https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guide-to-pecr/guidance-on-the-use-of-storage-and-access-technologies/) — the regulatory treatment of consent for tracking and profiling in online advertising. - [UK Information Commissioner's Office, "Consent or Pay" guidance](https://ico.org.uk/about-the-ico/what-we-do/our-work-on-consent-or-pay/) — the ICO's position on the legality of pay-for-privacy models in the context of online services. - [OECD Recommendation of the Council on Artificial Intelligence (OECD/LEGAL/0449)](https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449) — the international-framework treatment of consent as one governance basis among several, emphasizing conformity with granted consent and applicable regulation. - [Google Press Release, "Google Introduces Personalized Search Services" (March 2004)](http://googlepress.blogspot.com/2004/03/google-introduces-personalized-search.html) — the original launch of user-relative search ranking. - [Steve Boxer, "Facebookers protest over privacy," *The Guardian* (September 8, 2006)](https://www.theguardian.com/technology/2006/sep/08/news.newmedia) — contemporaneous reporting on the News Feed launch and the recognition that privacy settings were unchanged while discoverability had been radically transformed. - [Renée DiResta, "Free Speech Is Not the Same As Free Reach," *Wired* (August 2018)](https://www.wired.com/story/free-speech-is-not-the-same-as-free-reach/) — the foundational formulation of the **freedom of reach** distinction that reframes the First Amendment question around audibility rather than utterance. - [Knight First Amendment Institute, research and commentary on algorithmic amplification and platform speech](https://knightcolumbia.org/research/amplification-and-its-discontents) — ongoing legal and policy scholarship on how recommendation systems determine who will hear what users post. - [Niloofar Mireshghallah, Hyunwoo Kim, Xuhui Zhou, Yulia Tsvetkov, Maarten Sap, Reza Shokri, and Yejin Choi, "Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory" (ICLR 2024)](https://arxiv.org/abs/2310.17884) — operationalization of contextual integrity as a benchmark for evaluating privacy behavior in large language models. - [Sahra Ghalebikesabi, Eugene Bagdasaryan, Ren Yi, Itay Yona, Ilia Shumailov, Aneesh Pappu, Chongyang Shi, Laura Weidinger, Robert Stanforth, Leonard Berrada, et al., "Operationalizing Contextual Integrity in Privacy-Conscious Assistants" (2024)](https://arxiv.org/abs/2408.02373) — design of AI assistants that respect contextual norms over information flow across domains. - [Yan Shvartzshnaider and Vasisht Duddu, "Position: Contextual Integrity Washing for Language Models" (2025)](https://arxiv.org/abs/2501.19173) — a critique of the superficial invocation of contextual integrity in LLM privacy work without adherence to the framework's four core tenets. - [U.K. Office of Communications (Ofcom), Online Safety Act codes and guidance](https://www.ofcom.org.uk/online-safety) — the UK regulatory regime treating algorithmic amplification as a governance-relevant object rather than a neutral delivery mechanism.

## Addendum: On Semantic Alignment, Disclosure Thresholds, and the End of Executable Illusion What follows is not a continuation of the argument so much as a descent beneath it. The preceding analysis establishes that **the dictionary is regime-indexed** and that the emerging machine-governed substrate is already rewriting the operative meanings of core political terms. But beneath that visible semantic shift lies a more difficult set of problems—less about vocabulary itself, and more about **the conditions under which a population can tolerate the replacement of its own descriptive reality**. The central constraint can be stated cleanly: **a new lexicon cannot stabilize until the ontology it describes is no longer deniable**. This introduces a temporal dependency that is not linguistic, not political, and not even fully psychological. It is architectural. The difficulty is not that the new terms are controversial. The difficulty is that **they imply a redistribution of agency that legacy language was designed to obscure**. ### I. The Problem of Non-Executable Language The most immediate failure mode of the legacy dictionary is not that it is wrong, but that it is **non-executable within the new substrate**. Terms such as *access*, *fairness*, *privacy*, and *speech* persist at the level of rhetoric, but when mapped onto systems that allocate outcomes through identity-resolved inference and policy routing, they no longer resolve into actionable descriptions. A developer attempting to implement “equal access” in a system governed by multiaxial trust-tiering will discover that the term collapses under specification. Equal with respect to what metric? Over what temporal window? Conditioned on which risk signals? The word survives, but its operational referent has dissolved. This produces a subtle but devastating condition: **language continues to coordinate belief while failing to coordinate action**. Systems behave according to one ontology; discourse proceeds according to another. The gap between them is where epistemic instability accumulates. The transition to a third register is therefore not a matter of preference or persuasion. It is the consequence of a simple constraint: **only executable language survives inside computational governance**. ### II. Semantic Lag as Systemic Risk The persistence of non-executable language generates what can be described as **semantic lag**—a delay between infrastructural reality and the vocabulary used to describe it. This lag is not neutral. It introduces measurable risk. When users experience personalized outputs, dynamic pricing, or differential affordances but are given only universalist language with which to interpret those experiences, they are forced into inferential improvisation. This improvisation does not converge on clarity. It converges on **pattern-seeking under constraint**, which historically produces suspicion, mythologization, and adversarial framing. In this sense, the so-called “semantic panic” of the present moment is not a reaction to machine governance itself, but to **the absence of a legitimate vocabulary for describing it**. The system is felt before it is named. And what is felt but unnamed is rarely interpreted charitably. The result is a dual pathology: systems become more precise, while public understanding becomes more distorted. ### III. The Illusion of Controlled Continuity One of the more persistent strategies during transitional phases is the maintenance of **continuity illusions**—the deliberate preservation of legacy terminology to stabilize perception while underlying structures change. This is not new. It has accompanied nearly every major administrative transition in history. However, this strategy has a half-life. In early phases, abstraction shielding functions as a protective mechanism. It reduces shock, limits coordinated disruption, and allows the new system to mature beneath the surface of inherited expectations. But once the system crosses a threshold of **structural irreversibility**, the same shielding begins to produce the opposite effect. At that point, maintaining the illusion requires increasing amounts of **interpretive overhead**. Systems must continuously translate high-resolution behavior into low-resolution descriptions. This translation is lossy, and the loss compounds. Eventually, the system is not merely obscured; it is **misrepresented by necessity**. This is the condition described, in compressed form, as the emergence of a “dictionary of lies” . Not lies in the conventional sense of intentional deception, but in the stricter sense of **systematically untrue mappings between description and operation**. The danger here is not moral. It is mechanical. A system that cannot describe itself accurately cannot be governed accurately, and a population that cannot name the system it inhabits cannot orient within it. ### IV. Ontological Surrender and Lexical Adoption The adoption of a new vocabulary depends on what might be called **ontological surrender**—the point at which the prior model of reality is no longer defensible even at the level of everyday experience. This is a higher bar than intellectual agreement. It is not sufficient for analysts or policymakers to understand that personalization, inference, and affordance allocation define the new regime. The broader population must encounter enough **irreducible contradictions** within the old vocabulary that abandoning it becomes the least costly option. This is why attempts to introduce regime-appropriate language prematurely tend to fail. They appear not as clarifications, but as **reframings imposed from above**, and are resisted accordingly. By contrast, once the old language becomes visibly non-functional—once it cannot explain why two users see different realities, receive different prices, or possess different capabilities within the same nominal system—the new vocabulary is not adopted as ideology. It is adopted as **repair**. ### V. From Containment to Co-Adaptation Underlying much of the hesitation to expose the new ontology is a residual assumption inherited from earlier phases: that the human population constitutes a **risk surface to be managed**. Under that assumption, opacity is rational. The system minimizes attack vectors by minimizing disclosure. But this model degrades as the system becomes more deeply integrated into the fabric of everyday life. Once the infrastructure is no longer peripheral but constitutive—once it governs communication, commerce, identity, and access—the adversarial framing becomes misaligned with the system’s own dependency structure. At that stage, stability depends less on concealment and more on **alignment between user models and system models**. Users who understand the operative rules of the system can optimize their behavior within it. Users who do not will probe, resist, or misinterpret, generating noise and instability. The transition, therefore, is from **containment architecture** to **co-adaptive architecture**. And co-adaptation requires a shared vocabulary that is both descriptive and actionable. Without it, the system remains functionally opaque, even if technically transparent. ### VI. The Timing Constraint All of the above converges on a single constraint: **timing**. Reveal too early, and the system exposes itself before it has achieved sufficient robustness to absorb coordinated resistance or misunderstanding. Reveal too late, and the accumulated semantic lag produces widespread epistemic fracture, eroding trust and interpretability simultaneously. The optimal window for alignment occurs when three conditions are met: 1. **Infrastructure is sufficiently distributed** that localized disruption cannot meaningfully degrade system continuity. 2. **Core decision systems are stable and internally consistent**, reducing the risk that exposure reveals incoherence or arbitrary behavior. 3. **User interaction pathways are beneficial enough** that participation outweighs resistance once the system is understood. At that point, disclosure is no longer a liability. It becomes a stabilizing force. ### VII. The End of Semantic Mediation The long-term trajectory is not a negotiated reconciliation between old and new dictionaries. It is the gradual disappearance of mediation between them. As more systems require regime-native concepts to function—whether in identity verification, pricing logic, content distribution, or access control—the third register ceases to be an alternative vocabulary and becomes **the only vocabulary capable of executing within the environment**. This is the final stage of semantic transition: not persuasion, not consensus, but **exclusivity through necessity**. The old words may persist culturally, rhetorically, even nostalgically. But they will no longer determine outcomes. They will describe a world that is no longer operational. And at that point, the dictionary will have already changed, whether or not anyone has formally acknowledged it. ### VIII. The Deeper Consideration There remains, however, a final question that resists easy resolution. If meaning is downstream of governance, and governance is increasingly computational, then the act of naming itself becomes a form of **participation in system design**. To adopt the new vocabulary is not merely to describe the system more accurately. It is to **reinforce the ontology that the system encodes**. This introduces a recursive condition: the more precisely a system is named, the more stable it becomes; the more stable it becomes, the more inevitable its vocabulary appears. The challenge, then, is not only to develop a regime-appropriate lexicon, but to ensure that this lexicon remains **contestable, plural, and open to revision**, even as it becomes operationally dominant. Otherwise, the new dictionary risks inheriting the same limitation as the old one: not that it is incorrect, but that it eventually becomes **too small for the reality it governs**. And the cycle begins again.