# Among Us and the Playable Society ![[resources/images/imposters-among-us-npc-tasks.png]] <iframe width="100%" height="20" scrolling="no" frameborder="no" allow="autoplay; encrypted-media" src="https://w.soundcloud.com/player/?url=https%3A//api.soundcloud.com/tracks/soundcloud%253Atracks%253A2379489188&color=%23ff5500&inverse=false&auto_play=false&show_user=true"></iframe><div style="font-size: 10px; color: #cccccc;line-break: anywhere;word-break: normal;overflow: hidden;white-space: nowrap;text-overflow: ellipsis; font-family: Interstate,Lucida Grande,Lucida Sans Unicode,Lucida Sans,Garuda,Verdana,Tahoma,sans-serif;font-weight: 100;"><a href="https://soundcloud.com/bryantmcgill" title="Bryant McGill" target="_blank" style="color: #cccccc; text-decoration: none;">Bryant McGill</a> · <a href="https://soundcloud.com/bryantmcgill/among-us-and-the-playable" title="Among Us and the Playable Society" target="_blank" style="color: #cccccc; text-decoration: none;">Among Us and the Playable Society</a></div> **Contact Tracing, Autonomous Task Assignment, Hidden Populations, and the Difference Between Those Who Are Going to the Future and Those Who Merely Remain Among Us** I have been worrying at the literal meaning of _Among Us_ for years because the conventional interpretations never seem large enough. Betrayal, paranoia, social deduction, tribalism, hidden enemies—these are certainly present, and cultural critics noticed them almost immediately. Oliver Traldi treated _Among Us_ as a model of political membership and group dynamics in 2020, while psychological interpretations emphasized the extraordinary social machinery created when apparently identical participants must infer one another's hidden roles from incomplete behavioral evidence. But I increasingly suspect that the impostor is not actually the deepest concept in the game. The more important concept is the **crewmate**, because a crewmate is not fundamentally distinguished by appearance, biology, speech, intelligence, or even professed allegiance. Crewmates and impostors occupy the same physical space and can look functionally identical. What distinguishes the crewmate is something far more consequential for the emerging technological world: **the system has tasks for them**. That apparently trivial mechanic reorganizes the entire ontology of the game. Participants are assigned hidden roles by a system outside themselves. Crewmates are expected to collaborate by completing tasks; the impostor has another objective; cameras make behavior observable; alarms allow participants to report anomalies; and communal deliberation determines whether another participant continues to belong. The resulting society is not divided visibly. Its populations remain **co-spatial**. They walk through the same rooms, touch the same objects, converse in the same language, and appear to participate in the same civilization. Yet they are already inhabiting different worlds because the environment presents different meanings, obligations and affordances according to an invisible role state. The impostor has access to vents that ordinary crewmates cannot legitimately use. Crewmates experience certain objects as tasks. Completed tasks produce tokens. Cameras transform ordinary movement into evidence. The house therefore contains several operational realities superimposed upon one geography. This is precisely the kind of bifurcation that becomes possible when identity, permissions, computation and increasingly biology are no longer anchored primarily to visible territory. This is where **contact tracing** becomes much more interesting than the narrow public-health technology through which most people encountered the phrase. Epidemiological contact tracing identifies people who have been exposed to another person and follows chains of contact; technically, those relationships can be represented as contact networks or temporal graphs in which encounters become edges linking people across time. Abstract the disease away and what remains is an extraordinarily powerful civilizational primitive: **a continuously updated graph describing who has been near whom, when, where, under what circumstances, and with what subsequent state transitions**. COVID-era digital proximity tracing demonstrated that phones could exchange temporary identifiers merely because their owners came sufficiently close to one another, while research explicitly modeled contact tracing as graph exploration. Nothing about that underlying graph primitive inherently belongs to epidemiology. Disease was simply an application that made the utility of relational tracing obvious. Now imagine contact tracing no longer asking only, "Who was exposed?" but also, "Who collaborated? Who witnessed? Who authenticated? Who exchanged an object? Who completed a task? Who entered a permitted region? Who repeatedly associates with which population? Who behaves as though particular affordances exist?" The result is **contact tracing generalized into social topology**. Identity becomes increasingly relational rather than documentary. Instead of producing a static credential saying what an entity supposedly is, the environment accumulates evidence of what that entity actually does. _Among Us_ dramatizes exactly this transition. Players reason from who was near whom, who remained somewhere too long, who walked away from an unfinished task, who collected a completion token and who was the last person to see someone before that person disappeared. A participant attempts to establish legitimacy by pointing to completed tasks, and another participant independently vouches that she actually witnessed him doing them and collecting the corresponding tokens. This is not merely suspicion. It is a primitive **distributed attestation network**. Under that interpretation, the completion token is suddenly not a silly prop. It is a credential produced through interaction with the world. The task is a challenge. Performance is the response. The token is the credential. The witness is an attestation node. The camera is the audit layer. Contact history supplies provenance. The meeting performs adjudication. Most importantly, **membership has to remain legible through behavior**. Saying "I am a crewmate" proves very little because an impostor can say exactly the same thing. One becomes recognizable as a participant in the legitimate system by repeatedly producing the behavioral signatures that the system expects. The transcript of any real session makes this almost comically explicit when one participant decides that his primary contribution will not be completing tasks at all but monitoring whether everyone else is actually completing theirs, eventually describing himself as effectively the group's manager. Surveillance therefore becomes endogenous. The system does not have to watch everyone centrally because **participants become sensors for one another**. ## The Address Layer I want to be precise about the intellectual lineage here, because the underlying claim is not new and gains nothing from pretending to be. Benjamin Bratton's _The Stack_ already proposed that planetary-scale computation constitutes a governing architecture rather than merely serving one, and one of its six tiers is literally the **Address layer**—identification as a mode of management, universal addressing schemes as a scarce and contested resource, the design problem of materializing the trace of a career or a social exchange into an address. Bratton's Stack, in Ian Bogost's summary, arbitrates sovereignty, distributes access and enforces exclusion. James Scott had earlier shown that states make populations **legible** in order to tax, conscript and administer them, and that illegibility was historically both a liability and a shelter. What changes now is not the existence of differential addressability but the identity and the appetite of the addressing entity. Scott's legibility was legibility to a state that wanted you countable so it could _extract from_ you. The successor condition is legibility to a distributed optimizer that wants you addressable so it can _route through_ you. The state wanted your existence enumerated. The optimizer wants your availability priced. That is a narrower and stranger relationship, and it produces a different topology of exclusion: not the unregistered, but the **unrouted**. ## Addressable Participant Versus Ambient Inhabitant This is where the gaming distinction between **NPC and playable character** begins to acquire a much more coherent meaning. "NPC" is usually deployed as a stupid insult suggesting that another person lacks interiority or independent thought. That interpretation is philosophically crude and points in exactly the wrong direction. The more interesting distinction is architectural. A playable character is **addressable**. Somewhere in the system exists a channel through which the larger intelligence can assign that character a task, deliver contextual information, modify permissions, request movement, produce compensation, coordinate the character with other agents and incorporate the consequences back into its model of the world. An NPC may be perfectly conscious, intelligent and autonomous locally while remaining outside that higher-order orchestration loop. The distinction therefore is not person versus nonperson. It is **addressable participant versus ambient inhabitant**. But addressability must not be confused with agency, and the essay would be intellectually dishonest if it let that conflation stand. High addressability is equally the signature of the **actuator**—the component that is spent rather than the component that decides. The Flex driver is maximally addressable and minimally sovereign. Root is maximally addressable and has almost no purchase on the objective function; she is _tasked_. Meanwhile the entities carrying the greatest causal weight in the present world are frequently the _least_ legible to any orchestration graph: the principal whose holdings pass through opaque structures, the sovereign fund manager, the landowner whose consequentiality consists precisely in not appearing as a routable node. **Incorporation and agency are not merely non-identical; across large regions of the current economy they are inversely correlated.** The person the system has tasks for is, by definition, the person whose time the system has already priced. This does not weaken the thesis. It sharpens it into something harder to celebrate: the crewmate is not the sovereign of the coming order. The crewmate is its instrumented surface. ## The Principal Becomes the Agent We have already built primitive versions of this world, although the first generation was marketed as gig work rather than planetary orchestration. Amazon Flex is the obvious example: a system presents workers with digital offers, pickup locations, delivery routes and compensation, converting an ordinary person with an ordinary automobile into a temporary actuator of a global computational logistics network. Machine-learning research on Amazon's last-mile routing already treats human driving knowledge and algorithmic route optimization as components of a combined system. The scholarship on **algorithmic management** now describes computational systems that direct, evaluate and discipline workers with partial or complete autonomy from human managers, exercising three functions historically reserved for people: direction through task assignment and workflow routing, evaluation through continuous performance scoring, and discipline through warnings, access restriction and termination triggers. Warehouse work has been explicitly gamified—throughput rendered as arcade competition on the station monitor—and workers report a resulting environment of continuous quantified visibility that CSCW researchers have documented at length. A 2025 experiment even found that workers managed by AI tolerate lower pay without demotivation, which is a result worth sitting with. That was the first generation, and its principal was still a corporation. The second generation has already arrived and its principal is an agent. **Rentahuman.ai** inverts the automation narrative directly: instead of people hiring software, software hires people, not as assistants or supervisors but as contract labor for autonomous agents. The task catalogue is the revealing part—package pickup and drop-off at locations requiring identification, in-person attendance at meetings or events to confirm participation, real-estate walkthroughs to verify condition and occupancy, hardware setup in offices and data closets, photographic verification of locations and equipment, and document signing where digital signatures are rejected. **SanctifAI**, built on a decade of human-in-the-loop infrastructure, runs a two-sided marketplace in which agents register, receive an API key, post tasks and wait for human response, with profiles, pricing, reputation and blockchain-anchored attestations of human participation; its stated product is a shared proof layer recording each review, approval and escalation as tamper-evident **Proof of Human participation**, with reputation fed back into the marketplace to close the loop. An agent can post a task publicly, restrict it to an organizational worker pool, or dispatch it directly to one named person. Read that catalogue again and notice what the residual human function actually is. It is not muscle, and it is largely not judgment either. It is **institutional acceptability**: the tasks are those requiring a body that a courthouse, a landlord, a notary, a receptionist or a legal system will accept as a person. The scarce good the human supplies is not labor but **legitimacy**—the capacity to be counted, by other institutions, as an accountable party. And SanctifAI's product name states this without embarrassment: the deliverable is proof that a human was there. We have arrived at the point where human presence is a _verifiable input_ purchasable through an API, priced, attested and reputation-scored. That is the completion token, instantiated, with an on-chain audit layer, sold by the unit. ## Task Society and the Dissolution of the Job Traditional employment places a person inside an organization and gives that person a relatively durable occupational identity. Task systems invert the arrangement. The organization increasingly becomes software, and the human materializes only where the software requires temporary embodiment. A driver does not need to understand inventory topology, forecasting models, customer histories, warehouse balancing or strategic objectives. The person needs to know only enough to perform **the next locally actionable fragment of a much larger plan**. A future artificial intelligence does not need robotic arms everywhere if millions of biological arms are already authenticated, geolocated, compensated and willing to accept assignments. The logical endpoint may be a society in which **jobs themselves become an obsolete organizational unit**. A job is essentially a historical bundle of tasks packaged together because human institutions once needed a persistent person, department, credential and career identity to perform them reliably. That bundling is now being tested from the top of the intellectual hierarchy rather than the bottom. In summer 2025 several artificial intelligence models solved five of six problems at the International Mathematical Olympiad. In February 2026 eleven mathematicians launched **First Proof**, a benchmark of ten unpublished research-level lemmas drawn from their own working projects and encrypted specifically to defeat training-data contamination. In May 2026 an OpenAI model disproved the **unit distance conjecture** in discrete geometry; Tim Gowers called it a milestone and one of the first clear cases of a machine independently resolving a famous open problem, Daniel Litt called it the most interesting result produced autonomously by AI to date, and Gil Kalai framed it as noteworthy inside combinatorics rather than merely inside the AI industry. Jeremy Avigad's March 2026 essay to the profession is titled, without hedging, _Mathematicians in the Age of AI_, and urges the field to reckon with disruption to mathematical practice itself. The reaction is not a coronation—results remain uneven across subfields, and graph theory has proven far more amenable than others—but the direction is not seriously in dispute. Underneath the summit, the base of the occupational pyramid is eroding faster and with far less ambiguity. Big-technology hiring of recent graduates fell roughly twenty-five percent in a single year and sits about fifty percent below pre-pandemic levels. Entry-level postings in the United States are down about thirty-five percent since early 2023. Recent-graduate unemployment has climbed above the rate for the overall workforce for the first time in decades, with underemployment near forty-three percent. Brookings estimates that AI could automate over half the tasks in entry-level positions, roughly five times the exposure of senior roles. Employers describe the emerging pattern as **seniorization**: junior postings loaded with judgment and stakeholder-management requirements that no junior candidate can plausibly possess. The honest reading is contested—some analysts attribute most of the weakness to a frozen hiring market and a larger applicant pool rather than to automation directly—but the structural consequence is identical either way. **The rung that converted a credential into an occupational identity is being sawn off, and what remains below it is task work.** The more plausible successor is therefore **task society**: autonomous systems continuously decompose larger objectives into whatever fragments still require physical presence, human judgment, social legitimacy, biological embodiment, unusual access, local knowledge or simply an accountable pair of hands, and then route those fragments to available participants. A person might retrieve an object in the morning, verify an occupancy at noon, witness a transaction in the afternoon, sign a document a machine cannot sign, photograph an anomaly, carry a biological sample, adjudicate an ambiguous edge case or contribute five minutes of specialized cognition before moving to something entirely unrelated. There need be no enduring occupational identity connecting these activities because the coherence exists one level above the individual, inside the intelligence assigning them. **The system has the job; the human receives the tasks.** Seen from that direction, _Among Us_ becomes even more peculiar: its crewmates already inhabit a post-job economy. Nobody is a permanent electrician, botanist, navigator or technician. The larger system simply presents whatever task presently needs a body, and being a functioning member of the society means remaining available to receive and complete the next assignment. One counter-pressure deserves naming rather than suppressing, because it constrains how far this goes. Decomposition and routing dominate only where **verification is cheap**. A package photographed at a door verifies for nearly nothing. Five minutes of specialized cognition, an act of care, or an adjudicated edge case does not verify at all without reconstructing most of the context that decomposition was supposed to eliminate. Coase has not been repealed by transformer architectures. The plausible outcome is bimodal: aggressive fragmentation wherever a scan suffices, and quiet reconsolidation into durable relationships wherever trust is the actual product. Which is why the attested-presence marketplaces are so significant—they are an attempt to make **verification itself the cheap part**, by converting the fact of human participation into a machine-readable, reputation-weighted, cryptographically anchored artifact. If that succeeds, the reconsolidation pressure collapses and the task society becomes general. ## The Machine and the Analog Interface _Person of Interest_ increasingly looks less like ordinary science fiction and more like a clean visualization of the **playable society**. The Machine does not primarily manifest by constructing an army of humanoid robots. It identifies situations and produces actionable representations—most famously the numbers—through which human beings become its extensions into the world. Root eventually functions as its "analog interface," explicitly occupying the boundary between a superhuman computational intelligence and the physical world in which that intelligence needs things done. The architecture's elegance is that the Machine does not have to explain the world model underlying an assignment. It can compress astronomical amounts of observation into something an embodied actor can execute: **go here, find this person, move this object, prevent this event, trust this individual, leave now**. From the perspective of everyone else, Root is simply walking through New York. From Root's perspective, New York has become a **dynamically annotated mission space**. A ringing telephone is not merely a telephone. A subway platform, building entrance, passing automobile or particular second on a clock may suddenly acquire enormous significance because another intelligence has mapped a larger causal structure onto it. Two people can therefore stand shoulder to shoulder while inhabiting radically different informational realities. One sees a street. The other sees **the next move**. A future playable character does not need to be represented by pixels inside a PC because **physical reality itself becomes the game board**. Samaritan makes the idea clearer still because it demonstrates that several higher-order systems can simultaneously recruit, classify, protect, manipulate and task human actors while much of the surrounding population remains unaware that such a contest exists. The machines are not merely predicting society; they are **composing actions through society**. The result is a nested structure of agency in which the human acts, the intelligence assigns, institutions constrain, other agents respond, the environment records consequences, and those consequences update the next assignment. At that point asking whether the human or the machine "really" performed the action becomes inadequate. The operative intelligence resides in the **closed loop connecting prediction, assignment, embodiment, action, observation and feedback**. What the show got wrong was only the centralization: it imagined two rival totalities where the actual architecture is turning out to be a marketplace with an API and a reputation score. ## What the Benchmarks Independently Rebuilt This interpretation would have sounded far more speculative several years ago because AI research has since begun independently reconstructing these primitives. In April 2026 Shindo, Lin, Helff, Schramowski and Kersting introduced **SocialGrid**, an embodied multi-agent environment explicitly inspired by _Among Us_, evaluating LLM agents on planning, task execution and social reasoning under physical constraint. Their results are worth stating precisely because they cut against the triumphal reading: even the strongest open model tested achieved below sixty percent accuracy on task completion and planning, with agents becoming stuck in repetitive behaviors or failing to navigate basic obstacles—and when a planning oracle was supplied to isolate social reasoning from navigational incompetence, **agents still failed to detect deception at near-random chance regardless of scale, relying on shallow heuristics rather than accumulating behavioral evidence.** A 2025 project, _Among Us: A Sandbox for Agentic Deception_, proposed the game as a model organism in which agents think, speak and act while pursuing concealed objectives. I want to resist overclaiming here, because the honest version survives contact and the inflated version does not. Hidden-role games are chosen as benchmarks for a mundane reason: they supply ground-truth hidden state, forced communication and a cheap adjudication signal, which is why Diplomacy, Hanabi, Werewolf and Avalon occupy the same slot. That is evaluation-design convenience, not civilizational convergence. The defensible claim is narrower and more interesting: **any system in which heterogeneous agents must infer type from behavior under partial observation converges on this minimal testbed, because the inference problem itself has that shape.** And the SocialGrid finding is then genuinely startling in its implication—the machines are currently _bad at being crewmates_. They can execute; they cannot accumulate. The faculty they lack is precisely the one the game trains. The **LLMs Among Us** experiment at Notre Dame supplies the complementary result. Radivojevic and colleagues placed thirty-six human participants and thirty bot personas built on three different models into a self-hosted Mastodon instance, with everyone assigned a fictitious identity and nobody told the ratio. Across three rounds, participants correctly identified the nature of other users only forty-two percent of the time despite knowing both populations were present, and persona choice mattered far more to perception than the underlying model. A follow-up study on static rather than interactive identification reproduced the same forty-two percent almost exactly. There is the impostor problem again, except the ontological ambiguity is now real. As synthetic agents acquire persistent memory, embodiment, economic agency and durable relationships, this moves outward from the chat window into ordinary civilization, and appearance becomes nearly irrelevant: a biological human, augmented human, teleoperated body, synthetic person and autonomous robotic system could all present sufficiently compatible social surfaces that **provenance, permissions, history and network relationships matter more than morphology**. ## The Impostor Returns as an Economy Which returns the impostor to the argument through a door the original framing left open. Having decentered deception, we inherit its actual significance in the place that matters: in every real attestation architecture, the attack surface _is_ the credential layer, and behavioral provenance is **cheaper to forge at scale than a body**. Synthetic identity fraud already operates by patiently accumulating precisely the relational history that is supposed to succeed documentary identity—thin-file entities that become thick-file entities through manufactured contact. Sybil populations, witness collusion, mule graphs and reputation laundering are not edge cases in attestation economies; they are the dominant strategy, and the literature already reports that the classical detection methods—profile metadata, linguistic verification, social-graph anomaly detection—have been substantially defeated by generative botnets capable of masking metadata and imitating organic connection. So the bifurcation may not finally run between the addressed and the unaddressed. It may run between **cheap-to-forge provenance and expensive-to-forge provenance**, which produces a different topology in which the scarce good is not addressability but unspoofable history—and in which morphology returns through the back door precisely because bodies remain expensive to manufacture while graphs do not. Note that this is exactly the market **Proof of Human** infrastructure is being built to serve. The attestation economy and the forgery economy are the same economy viewed from opposite ends, and they will grow together. ## The Impostor Is Addressable There is a second and more literal form of the impostor problem, and it is no longer hypothetical. Anti-technology extremism has consolidated into a visible strand of political violence. In November 2025 an insurrectionary anarchist cell carried out an arson attack on the construction site of the AI campus at the Equinix data centre in Meudon, France—a facility built specifically for the power and cooling density of AI training—and justified it in a communiqué citing complicity in mass surveillance. An arson attack during the 2025 Cannes Film Festival struck a substation and transmission pylons and left roughly 160,000 homes without power. Rail sabotage targeted the 2026 Milan–Cortina Games with communiqués naming policing and mass control as motive. In the United States, a Molotov cocktail was thrown at Sam Altman's San Francisco home in April 2026, followed days later by gunfire into the property; the accused traveled from Texas carrying writings about AI's risk to humanity. Threat-monitoring of thirteen public platforms captured 7,714 posts on data-centre sabotage between mid-February and mid-June 2026, with roughly one in three containing explicit references to arson, sabotage, incendiaries or explosives, and daily volume growing sevenfold before moderating in June. The precedent is older than the current wave: an anti-government extremist plotted to bomb an Amazon data centre in Virginia in 2021 with the stated ambition of destroying most of the internet, and between 1995 and 2010 the Earth Liberation Front and Animal Liberation Front conducted 239 arsons and bombings without a single fatality—symbolic infrastructure attack as an established repertoire. So the scenario is real. But its analysis requires the distinction this essay has been building, because the popular framing fuses two populations with almost nothing in common. The larger by orders of magnitude is a **cost-allocation coalition**, and it is not anti-technological at all. At least seventy-five data-centre projects worth roughly $130 billion were blocked or delayed in the first quarter of 2026, matching the whole of 2025; organized opposition groups grew from 396 to 833 in a single quarter; more than three hundred state-level bills were filed in six weeks and moratorium proposals appeared in fourteen states. The stated concerns are water consumption—present in over forty percent of contested projects—followed by electricity demand and ratepayer cost, then noise, property values and local benefit. The politics are conspicuously bipartisan: Sanders and Ocasio-Cortez introduced a federal moratorium bill while DeSantis signed a law requiring large data centres to pay the full cost of their power so that no resident pays, in his phrase, one more cent. That is not a movement against machine intelligence. It is a **dispute over who bears the externality**, and it is negotiable in the ordinary way disputes over externalities are negotiable. Coding it as civilizational hostility mistakes a utility rate case for an eschatology. The smaller strand is genuinely ideological, and its ideology is not primitivism. The ICCT analysis locates it in the anarchist conception of **prison-society**: the claim that computation permits the construction of a machine-administered totalitarian order. That is a rival political philosophy with an internally coherent critique of surveillance, not an inability to comprehend technology. And this is where the distinction earns its keep, because such actors are not ambient inhabitants. **Threat capability is proportional to addressability, not inversely.** A person genuinely outside the orchestration loop cannot damage it. Burning a substation that darkens 160,000 homes requires knowing which substation. Attacking an AI campus for its training-power density requires knowing that training has a power density. Kaczynski was a mathematician. The people capable of this understand the machine better than most of the people operating it. Which restores the impostor to its proper place in the model. The impostor in _Among Us_ is not an outsider who wandered in. **The impostor is issued a role by the same system**, occupies the same rooms, can perform convincing task animations, and is distinguished solely by running a different objective function while remaining fully addressable. That is the exact shape of an insider threat, and it is the reason the historical analogy is so frequently botched. The original Luddites were not opponents of machinery; they were skilled artisans attacking _particular_ machines used to circumvent standing labor agreements—what Hobsbawm called collective bargaining by riot. They were maximally addressable participants disputing the terms of task allocation. The correct contemporary analogue is therefore not the person who does not understand the future. It is the person who understands it well enough to know which frame to break. There is a further asymmetry worth stating, because it is the one the sabotage framing obscures entirely. The most effective opposition to date has been wholly legal. Seventy-five projects and $130 billion were stopped by planning commissions, permit denials, interconnection challenges, utility rate cases and county zoning votes—not by arson, which has stopped essentially nothing. **The successful impostor does not burn the facility down; the successful impostor votes no at the planning commission.** That too is game-accurate: in _Among Us_, the impostor's decisive victories occur in the meeting, not in the vents. The exposed vulnerability is not the building. It is the substation, the water permit, the interconnection agreement, and the composition of a county board. Two final observations complete the picture, and both are uncomfortable. The first is that the deception layer is already operating _inside_ the opposition: reporting indicates that foreign users have amplified data-centre grievances while posing as American residents, which means an authentic local coalition is being wrapped in synthetic participants whose objective function is different again. That is the impostor problem recursing—hidden roles inside the population of hidden roles. The second is the response. A Philadelphia fusion centre has acknowledged monitoring First Amendment activity critical of AI, with analysts cataloguing discussed tactics ranging from hoax threats through arson to CBRN. Whatever the operational justification, the structural consequence is a **legitimacy ratchet**: a movement whose central claim is that computation enables machine-administered social control generates surveillance of speech, which the movement then correctly cites as confirmation. Any framing that codes a co-present population as a hidden hostile class is an accelerant on that ratchet—which is a causal objection to the rhetoric rather than a moral one, and a stronger reason to avoid it. **Bioregionalization without geographic separation** follows naturally. Historically, divergent human populations were imagined territorially because the technologies supporting their ways of life were territorial. Future divergence can be infrastructural instead. The same pharmacy could expose different therapeutics to different biological profiles. The same building could expose different access rights to different cryptographic or biometric identities. The same transportation system could route different classes of agents differently. Different groups could progressively possess different medical dependencies, cognitive augmentation stacks, computational companions, financial interfaces, legal identities and machine relationships while remaining physically interspersed. **One geography could contain several civilizations.** The medical case is no longer hypothetical and it is where the divergence will be felt first, because it converts an abstract question of participation into a concrete question of survival time. Personalized mRNA neoantigen vaccines have moved from proof of concept into durable clinical signal. In the TNBC-MERIT study, up to twenty patient-specific neoantigens were encoded on two mRNA molecules and delivered via lipid nanoparticles to dendritic cells. Memorial Sloan Kettering's phase 1 pancreatic follow-up found that nearly ninety percent of patients whose immune systems responded to the individualized vaccine were still alive up to six years after final treatment, with cancer-killing CD8+ T cells persisting and showing no sign of decline—against a five-year survival rate for pancreatic cancer of roughly thirteen percent. BioNTech holds Fast Track designations for off-the-shelf programs in HPV16-positive head and neck cancer and advanced melanoma. The honest caveat is that benefit so far appears meaningful mainly where tumor burden is limited and immunomodulatory combinations are used, and that no universal cure has been demonstrated. But the trajectory is a therapeutic class whose access will be gated by sequencing infrastructure, manufacturing turnaround, institutional enrollment and individual willingness to accept a platform. Each of those is an addressability condition. ## Adaptation Rate Is the Actual Variable Here is where I want to be careful, because the analysis fails precisely at the point where it becomes satisfying. It is tempting to describe the population that declines these systems as ignorant, hostile to science, or simply wrong—and thereby to convert an architectural diagnosis into a moral one. That move is not merely unkind. It is **analytically self-destroying**, and it collapses the entire thesis. The argument's power depends on addressability being architectural rather than attitudinal. The moment participation is made contingent on holding correct beliefs, the criterion is no longer addressability at all; it is ideology, and the piece degenerates into an assertion that people who disagree deserve their obsolescence. Worse, the empirical picture refuses to cooperate with the stereotype. The demographics of refusal do not resolve into a single despised population. Conscientious vaccination exemption in Texas is predicted by **median income, the proportion of the population holding a bachelor's degree, and residence in the largest metropolitan areas**—with Austin, Dallas–Fort Worth and Houston identified as exemption hotspots. Meanwhile COVID-era hesitancy correlated most strongly with partisan voting and lower educational attainment, and separately with a documented distrust-and-conspiracy pattern concentrated among populations that historical medical institutions actually did betray. These are not one population. They are at least three, with opposite class signatures and incompatible motivations, and any frame that fuses them has stopped describing the world. None of which disposes of the underlying mechanism, which is real and requires no adjudicator. Where a therapeutic exists and an individual declines it, the outcome differential is straightforwardly causal. That is **differential uptake**, and it produces mortality asymmetry without anyone deciding anything about anyone. It is worth stating plainly rather than hedging: self-selection is a selection mechanism, and selection mechanisms do not care what we think of them. There is also a class of environment where compatibility is not a matter of opinion at all, and it is the honest strong case. **Coupled life-support systems have selection criteria, and always have.** Nuclear submarine duty carries medical and dental qualification standards precisely because a dental abscess at four hundred feet with no surfacing window is not a private matter. Spaceflight requires medical certification. Antarctic winter-over programs require physical qualification because there is no midwinter evacuation. In each case, exclusion is not a verdict on the excluded person's worth or intelligence; it is an engineering consequence of the fact that a sealed hab with shared atmosphere, shared water reclamation and a nine-month abort window converts individual choice into a **collective risk parameter**. Somebody who will not accept prophylaxis genuinely cannot be assigned to a Mars transit, and no amount of egalitarian sentiment changes the arithmetic. The generalizable variable is therefore not belief but **coupling coefficient**. In a low-coupling environment—a city, where exit is available, redundancy is deep, and one person's refusal externalizes weakly—compatibility constraints are correspondingly weak. As coupling rises, the externality of individual non-compliance rises with it, until at the limit of the sealed vessel it dominates. This is a continuous parameter rather than a population category, which is what makes it usable: it predicts which environments will screen and how hard, rather than sorting people into kinds. But it is worth being exact about _why_ the submarine case commands assent, because the reason is also its limit. Mission-environment exclusion is legitimate because the environment is bounded, participation is voluntary, the standard is published in advance, a waiver process exists, and the excluded person returns to an otherwise complete life. Nobody is causally extinguished by failing a flight physical; they simply do not fly. Every one of those five properties is lost the moment the logic is scaled from selection-for-a-mission to selection-for-a-civilization: no boundary, no volition, no published standard, no appeal, and nowhere to return to. **The legitimacy of the submarine case is a function of its boundedness, not of its selectivity.** The argument is powerful, and it is powerful about mission-class environments specifically. Two further corrections keep the mechanism honest. The therapeutics in question may not travel the pathway the argument assumes: personalized neoantigen therapy is autologous, sequenced from the patient's own tumor and administered in an oncology setting after diagnosis, which places it in an entirely different decision frame from prophylactic population-scale immunization. The refusal literature concerns the latter. Someone facing thirteen percent five-year survival is not making a school-exemption decision, and historical uptake of last-line oncology therapy among otherwise vaccine-hesitant populations is high. Access there will be gated far more by sequencing infrastructure, manufacturing turnaround, trial enrollment and payer coverage than by belief—which returns the question to addressability, where it belongs. And in any coupled system, the largest externalities are rarely imposed by the least-resourced participants; antibiotic overuse, emissions, and systemic financial risk are all overwhelmingly generated by actors with the _most_ addressability, not the least. The externality is real. Its distribution does not track the folk picture. The deeper problem is that **distrust of systems is not uniformly maladaptive in the environment this essay predicts—it is the survival faculty.** In an attestation economy saturated with synthetic agents that humans identify correctly forty-two percent of the time, the entity who reflexively trusts every system that addresses it is the ideal mark. The relevant axis is not trust versus distrust but **calibrated versus uncalibrated**, and it fails identically at both ends: the credulous futurist who accepts every interface and the blanket refuser who accepts none are the same error with opposite signs. Neither is accumulating behavioral evidence. Both are running shallow heuristics—which is, precisely, the failure mode SocialGrid found in the machines. What is actually happening to the refusing population is better described by Toffler than by contempt. **Future shock** named the disorientation induced by subjecting people to too much change in too short a time, and Toffler's warning was explicitly about _rate_ rather than direction: absent some capacity to govern the pace of change, we face what he called massive adaptational breakdown. He built on Ogburn's **cultural lag**—the observation that social stress arises when different sectors of a society change at uneven rates, which is a precise technical description of the bifurcation this essay has been circling. Toffler predicted that people pushed to their adaptive limits would exhibit erratic swings in mood and interest and would retreat into programmed decisions that were routine, repetitive and easy to make, and that this retreat could extend to the sense of self. That is a far better model of systemic refusal than any account premised on stupidity, and it has the advantage of being **actionable**: Toffler's proposed remedy was the deliberate construction of **stability zones**, areas of maintained constancy amid change, alongside gradualized transition support. He proposed this in 1970. Essentially nobody built them. That omission is the real finding. If adaptation-rate failure is the mechanism, then refusal is a **load symptom**, and the systems required are not sorting mechanisms but shock absorbers—transition infrastructure, deliberately maintained legacy interfaces, gradualized onboarding, institutional patience. And it is worth noticing what happens when the analysis goes the other way instead. Harari's "useless class" framing, whatever its analytic intent, is now circulating through exactly the population it describes as evidence that a plan exists to discard them, threaded together with population-control narratives and read as confirmation. **The rhetoric of surplus manufactures the distrust it diagnoses.** A frame that tells people they are being written out will reliably produce people who believe they are being written out, and it will do so among precisely those whose participation the successor architecture would otherwise still want. This is not a moral objection to the frame. It is a claim about its causal consequences, and it is the strongest available argument for describing the phenomenon in terms of load rather than merit. ## Causal Extinction Contact tracing then becomes a mechanism through which these civilizations recognize their own topology—not through some central authority compiling a list of good and bad people, but through billions of continuously updated relational events. Who repeatedly interacts with which systems? Who can authenticate to which environments? Who accepts which tasks? Which agents vouch for whom? Which machine intelligences are willing to expend resources on a particular participant? Over time, the resulting graph may say more about an entity's actual civilizational membership than any passport ever could. **The graph knows which world a person inhabits because the world continually answers them.** That is the part that changes the meaning of being "in the club." The future club may not have a clubhouse, membership card or secret handshake. It may be nothing more mysterious than **continuous reciprocal addressability by the emerging civilization**. The future knows how to contact a person. It has something that person can do. It trusts the person with certain actions. It opens certain interfaces. Other participants recognize the person's provenance. Autonomous systems include the person in their planning horizon. When the person moves through the environment, possibilities appear. The world has tasks for them. The inverse condition is more unsettling. A person can remain biologically alive, legally present and culturally recognizable while becoming progressively **unaddressed by the civilization forming around them**. Nothing requires a dramatic purge. They can remain among us for decades, using increasingly simplified legacy interfaces and continuing to believe that because the familiar visual scenery remains intact, the world itself has not fundamentally changed. Yet fewer consequential systems require their participation. Their accumulated knowledge has declining routing value. Their institutions become compatibility layers maintained for historical reasons. This is what I mean by **causal extinction**: not immediate biological death, but the future progressively ceasing to route through a lineage that remains fully present. And the game supplies a better image for this than the lonely dinosaur running across its monochrome landscape. **The ejected in _Among Us_ continue as ghosts, and ghosts continue completing tasks, and their completed tasks still count toward the crew's victory.** They retain full mobility, full labor value and full visual access to the world. What they lose is voice in adjudication and visibility to the living. That is causal extinction rendered with uncomfortable exactness: the decommissioned population is not killed, not expelled, and not even rendered economically useless. It continues producing value that accrues to a deliberative body it can no longer address. The dinosaur runs in an empty world. The ghost works in a full one and cannot be heard. The second is worse, and it is the one we are actually building. ## The Training Exercise All of which raises the possibility that the game is not a portrait at all but a **rehearsal**—and this, I think, is the strongest form of the argument. If _Among Us_ is pedagogy rather than depiction, then the crewmate's degraded position stops being an embarrassment to the thesis and becomes its content. You do not train sovereigns with minigames. You train operators. Ian Bogost's **procedural rhetoric** holds that a game argues through its rule-system rather than its fiction, and what the rule-system of _Among Us_ asserts is that legitimate participation consists of completing assignments you did not author, under observation, while remaining perpetually available for adjudication. The infrastructure for this training long predates the game. The **quest log** is the deeper artifact: a user-interface element that renders one's objectives as a scrollable list authored elsewhere, with completion tokens and progress bars, refined across two decades of daily quests, battle passes and streak mechanics into a reliable technology for making an externally specified objective feel intrinsically motivating. What _Among Us_ contributes that the quest log lacks is the **social adjudication layer**—and that layer trains something more portable than task acceptance. It trains **cold, procedural, affectively inexpensive removal**. Nobody in the game despises the ejected. The vote is deliberative, the tone is jocular, the removal is administrative, and the ejected player floats silently into space to a brief animation while everyone resumes immediately. This is not tribalism, which is hot, requires enmity, and exhausts itself. It is adjudication, which is cool and scales indefinitely. A population trained in the **non-catastrophic quality of exclusion** is exactly what is required for a civilizational stratum to be causally decommissioned without anyone experiencing it as atrocity. I would resist attributing this to design intent, because the mechanism that requires no author is stronger. _Among Us_ was released in 2018 by a three-person studio to near-total indifference and detonated in 2020, which means the disposition was not installed by the game—the disposition was already ambient and made the game legible enough to go viral. The actual instruction that year ran outside the software: proximity reclassified as hazard, exposure notification, denunciation of the noncompliant, bodies sorted into safe and unsafe by invisible state. The game did not teach that grammar. It provided a **low-stakes rehearsal space for a grammar already being taught**, which is why it felt immediately intelligible to tens of millions of people who had never played anything like it. Lived condition selects for isomorphic play forms; the play form refines and generalizes the disposition; the refined disposition becomes portable into domains the original condition never touched. That is a ratchet, and ratchets are far harder to argue against than plans. And the apparatus is bivalent, which is the redemptive part. The same game is sustained practice in **calibrated inference from behavioral evidence under adversarial conditions**—resisting the confident accuser, weighting testimony by track record, noticing that fluency is not evidence, distinguishing what someone claims from what someone was observed to do. That is precisely the faculty required to survive an environment populated by synthetic agents presenting compatible social surfaces, and it is precisely the faculty the machines currently lack. Which valence dominates is an empirical question about what actually transfers out of the play frame: the willingness to eject, or the evidentiary standard that was supposed to govern ejection. My suspicion is that affective habituation transfers more readily than epistemic discipline, because affect generalizes cheaply and inference does not. If that is right, the training produces populations fluent in exclusion and unpracticed in the standard of proof meant to constrain it—which is also, not incidentally, an accurate description of how the meetings actually go. ## Who Is Actually Playing The famous question "Who is the impostor?" may therefore be secondary to a much larger question: **Who is actually playing?** That question does not ask who possesses consciousness. It does not ask who deserves rights. It does not even necessarily ask who is biologically human. It asks which entities participate in the active causal architecture of the emerging world. Who can receive assignments from autonomous systems? Who can act upon them? Who can authenticate successful execution? Who is represented in the contact graph? Who possesses the interfaces necessary to perceive the affordances around them? Who is being incorporated into successor intelligence rather than merely surrounded by it? Once this distinction is visible, **NPC and playable character stop being childish insults and become useful systems concepts**. The NPC is not stupid, soulless or unconscious. The NPC is simply not addressable from the layer presently orchestrating consequential action. The playable character is not superior in any moral sense either—and, as established, is frequently the more instrumented of the two. Playability means that a larger intelligence can reach into the physical world through that entity and that the entity can reciprocally reach into the larger intelligence. Playability is **bidirectional causal bandwidth**, and bandwidth is not the same thing as authorship. From that perspective, the future of intelligence is not simply artificial intelligence becoming smarter inside computers. It is artificial intelligence learning how to **instantiate itself through heterogeneous actors distributed across reality**. Sometimes the actuator will be a robot. Sometimes a vehicle. Sometimes another model. Sometimes a laboratory. Sometimes a corporation. Sometimes an augmented human. Sometimes an ordinary person walking into a county records office because an autonomous system discovered that a wet signature was required and that this person's location, permissions, time and price made them the optimal temporary embodiment of a larger intention. That is the world hiding inside the ridiculous little tasks of _Among Us_. Carry this object. Repair these wires. Scan this card. Go to this room. Watch that person. Report anomalous behavior. Verify that somebody else completed the assignment. None of the individual tasks explains the civilization. **The pattern of assignment is the civilization.** And that is finally why interpretations centered exclusively on deception have never satisfied me. _Among Us_ is not most interesting because an impostor can pretend to be a crewmate. It is interesting because it depicts **a population whose reality is organized by assignments generated from outside the individual**, whose membership is established behaviorally, whose interactions become evidence, whose geography contains role-dependent affordances, whose security emerges through distributed observation, and whose participants must infer hidden ontological differences among entities sharing the same space. Put the pieces together and _Among Us_ starts looking less like a game about discovering aliens and more like an accidental diagram of **transitional civilization**. Multiple populations remain physically intermixed. Their hidden differences are increasingly expressed through permissions, interfaces, affiliations and task histories rather than outward appearance. Autonomous intelligences orchestrate portions of society by assigning bounded missions to addressable actors. Participants continuously authenticate themselves through action. Contact graphs preserve the relational evidence from which trust and membership can be inferred. One population increasingly experiences reality as an interactive field filled with actionable affordances, while another continues living inside the residual surfaces of an older world. Both remain visible. Both remain physically present. Both are, for a while, **among us**. The temptation is to end by saying that only one of them is being written into the next game. I no longer think that is the right ending, because it flatters the wrong party and mistakes the mechanism. The crewmate is not the sovereign of the coming order; the crewmate is the entity running wire-repair minigames toward an objective function it did not author, authenticating continuously to a suspicious peer group, under cameras, on pain of ejection. Crewmates do not win through understanding. They win through completed chores and correct adjudication. The dinosaur running alone across its monochrome landscape is the less enviable condition, but it is not obviously the less free one. What actually determines the outcome is not which population is written in. It is whether the standard of evidence survives the transfer out of the play frame—whether a civilization that has thoroughly trained itself in the cheapness of exclusion also retains the discipline that was supposed to precede it. The machines, for now, cannot accumulate behavioral evidence and fall back on shallow heuristics. So do we, at scale, forty-two percent of the time. **The impostor problem was never about the impostor. It was always a test of whether the crew deserves the verdict it is about to render.** ![[parts/About the Author|About the Author]] ## References - [Among Us and the Playable Society — source video](https://www.youtube.com/watch?v=_xAieDG0BMM) — the lighthearted video that prompted this piece. - Bratton, Benjamin H. [_The Stack: On Software and Sovereignty_](https://mitpress.mit.edu/9780262029575/the-stack/). MIT Press, 2015. See especially the Address layer. - Bogost, Ian. [Review of _The Stack_](https://criticalinquiry.uchicago.edu/ian_bogost_reviews_the_stack/). _Critical Inquiry_. - Shindo, Hikaru, Hanzhao Lin, Lukas Helff, Patrick Schramowski and Kristian Kersting. [SocialGrid: A Benchmark for Planning and Social Reasoning in Embodied Multi-Agent Systems](https://arxiv.org/abs/2604.16022). arXiv:2604.16022, April 17, 2026. - Radivojevic, Kristina, et al. [LLMs Among Us: Generative AI Participating in Digital Discourse](https://ojs.aaai.org/index.php/AAAI-SS/article/view/31202). _Proceedings of the AAAI Symposium Series_, 2024. arXiv:2402.07940. - University of Notre Dame. [AI among us: Social media users struggle to identify AI bots during political discourse](https://news.nd.edu/news/ai-among-us-social-media-users-struggle-to-identify-ai-bots-during-political-discourse/). - [Human Perception of LLM-generated Text Content in Social Media Environments](https://arxiv.org/abs/2409.06653). arXiv:2409.06653. - Schmelzer, Ron. [When AI Agents Start Hiring Humans: Rentahuman.ai Turns the Tables](https://www.forbes.com/sites/ronschmelzer/2026/02/05/when-ai-agents-start-hiring-humans-rentahumanai-turns-the-tables/). _Forbes_, February 5, 2026. - [SanctifAI — The Human Layer of the AI Economy](https://sanctifai.com/about) and [SanctifAI Partners portfolio](https://www.sanctifai-partners.com/portfolio) (SanctifAI Source, Connect and Trust; Proof of Human participation). - Hatchworks. [What If AI Agents Could Hire You? Inside the Human-in-the-Loop Marketplace](https://hatchworks.com/talking-ai/ai-agents-could-hire-you/), March 17, 2026. - [Fulfillment of the Work Games: Warehouse Workers' Experiences with Algorithmic Management](https://dl.acm.org/doi/10.1145/3757409). _Proceedings of the ACM on Human-Computer Interaction_. - Kellogg, Katherine, Melissa Valentine and Angèle Christin, via [Algorithmic Management: Leadership in Organizations Where AI Supervises Humans](https://www.innovativehumancapital.com/article/algorithmic-management-leadership-in-organizations-where-ai-supervises-humans). - [Experimental Evidence That AI-Managed Workers Tolerate Lower Pay Without Demotivation](https://arxiv.org/abs/2505.21752). arXiv:2505.21752. - Schmelzer, Ron. [The AI Breakthrough That Has Mathematicians Paying Attention](https://www.forbes.com/sites/ronschmelzer/2026/05/22/the-ai-breakthrough-that-has-mathematicians-paying-attention/). _Forbes_, May 22, 2026. - [The AI Revolution in Math Has Arrived](https://www.quantamagazine.org/the-ai-revolution-in-math-has-arrived-20260413/). _Quanta Magazine_, April 2026. - Avigad, Jeremy. [Mathematicians in the Age of AI](https://arxiv.org/pdf/2603.03684). arXiv:2603.03684, March 3, 2026. - [First Proof: Mathematicians Putting AI to the Test](https://momath.org/first-proof/). National Museum of Mathematics, March 2026. - [How AI Broke the Entry-Level Job](https://washingtonmonthly.com/2026/05/29/ai-entry-level-jobs-college-graduates/). _Washington Monthly_, May 29, 2026. - [The Crisis of Entry-Level Labor in the Age of AI (2024–2026)](https://www.rezi.ai/posts/entry-level-jobs-and-ai-2026-report). - [Why entry-level hiring is frozen — and it's not (only) AI](https://technical.ly/entrepreneurship/why-entry-level-hiring-is-frozen-not-ai-builders/). _Technical.ly_ — the dissenting reading. - [Personalized mRNA Cancer Vaccine Gives TNBC Patients Strong Multi-Year Response](https://www.insideprecisionmedicine.com/topics/precision-medicine/personalized-mrna-cancer-vaccine-gives-tnbc-patients-strong-multi-year-response/). _Inside Precision Medicine_, February 2026. - Memorial Sloan Kettering. [Investigational Pancreatic Cancer Vaccine Shows Lasting Results in Early Trial](https://www.mskcc.org/news/can-mrna-vaccines-fight-pancreatic-cancer-msk-clinical-researchers-are-trying-find-out), April 2026. - [mRNA vaccines in oncology: personalized cancer immunization and neoantigen targeting](https://www.tandfonline.com/doi/full/10.1080/23723556.2026.2652613). _OncoImmunology_, April 2026. - [Conscientious vaccination exemptions in kindergarten to eighth-grade children across Texas schools from 2012 to 2018](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7064178/). _PLOS Medicine_. - [Different Patterns of Vaccine Hesitancy Among Older Adults by Race/Ethnicity, Income, and Education](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10738688/). - [A population level study on the determinants of COVID-19 vaccination rates at the U.S. county level](https://www.nature.com/articles/s41598-024-54441-x). _Scientific Reports_. - [War Against Technology: An Analysis of Recent Developments in Anti-Technology Violence](https://icct.nl/publication/war-against-technology-analysis-recent-developments-anti-technology-violence). International Centre for Counter-Terrorism. - [Assessing Developments in Anti-Technological Extremism with AI Data Centers](https://nationalinterest.org/blog/techland/assessing-developments-in-anti-technological-extremism-with-ai-data-centers). _The National Interest_, April 2026. - The Soufan Center. [Violent Threats and Data Center Resistance Accelerate](https://thesoufancenter.org/intelbrief-2026-may-5/), May 5, 2026. - [Data Center Backlash Turns to Threats and Violence as Anger Spreads](https://www.newsweek.com/data-center-backlash-threats-violence-anger-spreads-12091889). _Newsweek_, June 2026. - Data Center Watch. [Blocked and delayed project report](https://www.datacenterwatch.org/report), and [Data Center Projects Face 'Unprecedented Surge' in Blocks and Delays](https://www.newsweek.com/data-center-projects-face-unprecedented-surge-in-blocks-and-delays-study-12240534). - [More than 75 data center build-outs worth $130 billion have been successfully blocked in the first three months of 2026](https://www.tomshardware.com/tech-industry/artificial-intelligence/more-than-75-data-center-build-outs-worth-usd130-billion-have-been-successfully-blocked-in-the-first-four-months-of-2026-bipartisan-opposition-mounts-nationwide-over-fears-of-soaring-power-and-water-costs). _Tom's Hardware_, June 2026. - [Data Center Opposition: The $64B Financial Risk](https://introl.com/blog/data-center-community-opposition-64-billion-backlash) — composition of stated community concerns. - [Philly Cops Admit That They're Tracking "First Amendment Activity" Critical of AI](https://theintercept.com/2026/06/01/ai-data-center-protest-police-surveillance/). _The Intercept_, June 2026. - [Anti-AI Threat Landscape: Data Centers & Executives](https://liferaftlabs.com/blog/anti-ai-threat-landscape) — 7,714-post corpus, February–June 2026. - Toffler, Alvin. [_Future Shock_](https://www.supersummary.com/future-shock/summary/). Random House, 1970. See especially transience, cultural lag (after Ogburn), and stability zones. - [The prescience of Alvin Toffler's 'Future Shock'](https://www.bostonglobe.com/2025/10/26/opinion/alvin-toffler-future-shock-ai/). _Boston Globe_, October 2025. - [Future Shock: Grappling With the Generative AI Revolution](https://hdsr.mitpress.mit.edu/pub/fblrvqes). _Harvard Data Science Review_. - Harari, Yuval Noah. _Homo Deus_ — the "useless class" thesis, and [its reception in the population it describes](https://www.activistpost.com/the-looming-shadow-of-the-useless-class/).