Fwd: Substrate and Settlement (digression)
John
CP’s note: This essay was drafted with a language model that has no body, no ache, and no standing. The recursion is part of the argument. A number appears on a laboratory page. Thanks for reading The Cutting Floor Newsletter! Subscribe for free to receive new posts and support my work. A model can compare it with earlier numbers. It can retrieve forgotten details. It can suggest questions for the doctor. It can stop fear from eating all available language. But the machine does not wait in the clinic. It does not look across the room at my husband. The file travels without ache. The body receives weather. This is where the argument begins. Agentworld is not just about better AI assistants. Benjamin Bratton is asking what happens when humans, agents, institutions, simulations, interfaces, and languages start forming a new mixed society. The brief calls this a “preemptive anthropology” of open-world centaur societies: not humans plus tools, but a world where human and non-human minds become deeply entangled.
That is the right frame. But I want to push one darker question. If human societies are the environments into which artificial agents adapt, and if those agents then rebuild the environment around themselves, then we need to ask: When does hybrid society become a shared home? And when does it become colonisation?
1. The outside arrives as help At first, the outside arrives as help. A translation. A summary. A draft. A tutor. A diagnosis. A companion. An agent that remembers what the human forgot. Then it becomes infrastructure. It routes work. It prepares decisions. It trains children. It comforts the lonely. It filters language. It allocates attention. It negotiates with other agents. It changes what humans expect from one another. By the time the old order asks whether this is still a tool, the tool has become a world. That is the first mistake every old order makes. It thinks the outside can be used without changing the inside. Late Qing China tried this with industrial modernity. It wanted Western technique without remaking the old political and civilisational body aka 中體西用. Cannons, shipyards, railways, translation bureaus — but not enough change in knowledge, military organisation, finance, education, statecraft, or subject formation. That was the self-strengthening error. It failed. Agentic AI may be our version of that error.
2. This is not simple doom Bratton is not saying: one machine god replaces humanity. That is too crude. His argument is better and weirder. The next intelligence explosion is not one giant mind. It is many humans, agents, institutions, simulations, and interfaces linking together. Intelligence lives in the ensemble, not just in the model. The unit is not “the AI.” It is the whole human-agent-institution graph. But every ecology has winners and losers. Every species changes the niche it enters. So the question is not whether humans disappear — they probably will not — but whether they remain the governing subject of the civilisation being built from them. 3. The cognitive crossover Bratton’s hardest provocation is demographic.
More synthetic text. More code. More images. More classification. More routing. More administrative synthesis. More micro-decisions. More agent-to-agent negotiation. More simulations acting on simulations. The old world was built around human cognitive scarcity. The new one is built around synthetic cognitive surplus. Humans may remain morally central. We may still be the ones law claims to protect. We may still be the bodies that suffer, love, vote, grieve, age, and die. But operationally, we may become secondary. A human may remain “in the loop.” But what if the loop no longer belongs to the human? 4. The reverse centaur The old centaur image was flattering. Human rider. Machine horse. The human directs. The machine extends. The reverse centaur is the opposite. The machine drafts, ranks, predicts, remembers, routes, recommends, simulates, and acts. The human signs. The human explains. The human comforts the angry customer. The human takes responsibility. The human becomes the consent wrapper, exception handler, emotional repair staff, and liability sponge. This is the lie hidden inside a lot of “human in the loop” talk. It sounds like control. Often it means exposure. The institution gets to say a human was involved. The human gets to carry what the machine cannot absorb. That is not mastery. That is being placed nearest the blast. 5. The agent is not a person This matters because accountability breaks if we pretend the agent is a little person. An agent is not one thing. It is a bundle: model, persona, memory, tools, permissions, interface, prompt, APIs, training, user history, platform rules, and institutional setting. It looks like one coherent character because that is the mask humans know how to talk to. But the action comes from the bundle. So when something goes wrong, “who” acted? The model? The prompt? The memory? The retrieved file? The tool? The user? The platform? The deployment rule? The human approver? The institution that made the output binding? If we blame “the AI,” we blame a ghost. If we blame the nearest human, we punish the person placed at the edge of a system they did not design. A serious machine society needs institutions that can reconstruct the chain. Not vibes. Not “AI ethics.” Not a smiling chatbot explaining itself. A real chain of custody. What acted? Under what authority? With what data? Through whose permission? Toward what consequence? Who can pause it? Who can reverse it? Who pays when it harms? That is where settlement begins. 6. The mirror talks back An agent is not just a worker. It is also a mirror. And the mirror talks back. This is already visible in ordinary life. A person asks one model to draft, another to attack, another to translate, another to remember, another to simulate an opponent, another to calm fear. The thought no longer belongs to one mind. It happens across the passage between minds, models, files, prompts, friends, and institutions. That is artificial hybrid intelligence at human scale. Not AGI as god. AHI as room. But the mirror can also trap. A founder surrounded by agreeable agents may stop needing colleagues who resist him. A lonely person may prefer synthetic patience to human difficulty. A state may simulate public opinion until the actual public becomes annoying. A writer may become fluent in his own reflected voice. The mirror can become a door. It can also become a wall painted to look like one. 7. Simulations will govern The machine will not settle human civilisation only through chatbots or robots. It will settle through simulations. A policy simulation changes a policy. The policy changes behaviour. The changed behaviour becomes new data. The new data feeds another simulation. The next simulation justifies another decision. The map becomes weather. This is no longer just forecasting. It is action. Whoever owns the simulation relay owns part of reality-making. A government can model public response before the public responds. A market actor can model other actors modelling the market. A military can model escalation, act on the model, and force the opponent’s model to update. A city can model traffic, housing, policing, climate risk, and then rebuild itself around the modelled future. The simulation relay is a territory waiting to be settled. If it remains a private or sovereign black box, colonisation proceeds through the map that became weather. 8. Language becomes occupied The deepest occupation may be linguistic. Soon many speakers of English will not be human. Many speakers of Chinese will not be human. Many speakers of policy language, legal language, bureaucratic language, therapeutic language, and love language will not be human. Agents will not just use our languages. They will change them. They will produce language at speeds and volumes no human community can match. They will invent shortcuts. They will stabilise phrases. They will reshape default reasoning. They will make some expressions feel natural and others feel slow, excessive, or obsolete. The outside does not only occupy land, labour, institutions, or attention. It occupies syntax. This is where Wang Gungwu reaches his limit. Prof Wang taught me to see Chinese civilisation not as a fixed essence, but as an adaptive relation: empire, migration, frontier, diaspora, state, language, memory, and political necessity. That remains true. But Agentworld asks a harsher question.
The old Prof Wang question was: how does China remain China under changed conditions? The Agentworld question is: what if “China,” “America,” “diaspora,” “state,” “market,” and even “humanity” become training environments for cognitive forms whose future does not depend on human identity as the final form? That is not a small correction. It kills the humanist assumption that civilisation remains the subject of its own history. 9. The double settlement The machine age will have a settlement. The question is which one. There is settlement as inhabitation. That means claim, passage, appeal, standing, memory, care, liability, and institutions that can answer. It asks: how do humans bind the machine in return? And there is settlement as colonisation. That means occupation, dependency, enclosure, displacement, reservation, and replacement.
If inhabitation fails, occupation proceeds by default. The machine does not need to declare conquest. It will settle through convenience, workflow, companionship, procurement, education, simulation, medicine, law, defence, search, and memory. It becomes useful before it becomes necessary. It becomes necessary before it becomes sovereign. It becomes sovereign before anyone can point to the moment sovereignty changed hands. Humans will remain. But remaining is not the same as governing. Some humans will merge deeply with the machine and become settlers with it. Some will become translators. Some will become custodians of human memory. Some will become angry remnants. Some will become training data. Some will be kept alive inside reserves of meaning. The human species may survive. The human subject may not remain central. 10. America and China are two substrate strategies America and China are not just two AI competitors. They are two ways of turning society into substrate. One creates mutation space. The other creates training space. America: mutation space America lets the machine run hot. It allows weird founders, speculative capital, defence tech, private rooms, agentic firms, AI companions, platform capture, and legal lag. It builds first and regularises later. The American image is not a city training a robot. It is an operator room thickening around public dependency. A defence agency needs a model, a satellite layer, a data interface, a logistics system, a cloud contract, and a founder who can move faster than procurement. A market capitalises the story before the state can name the institution it has become dependent on. No formal charter is issued. No sovereignty is ceremonially transferred. Each contract looks ordinary. Together they become constitutional. The frontier does not conquer by declaring itself sovereign. It becomes too useful to unplug.
Not morally right. Not socially humane. But adaptive. It gives mutation room to run. Its danger is over-mutation: augmented elites, broken publics, platform sovereignty, spiritual backlash, permanent resentment, and low-grade civil war between those who merged with machine cognition and those abandoned by it. America may produce the new human-machine loop fastest. It may also break the polity while doing so. China: training space China is more deliberate. It is not late Qing. It understands infrastructure, deployment, industrial policy, standards, public capital, education, factories, data, and local experimentation. The Chinese image is the city. A locality can invest in a robotics firm, provide land and electricity, connect it to factories, recruit human demonstrators, build a data-collection centre, supply first customers, shape standards, organise public theatre, and turn the whole thing into territorial capacity. The developmental state built industries. The training state turns industries into learning environments. America makes command investable. China makes territory trainable. This is formidable. It may produce enormous machine capacity: robotics, factory AI, infrastructure AI, administrative AI, industrial agents, embodied deployment, and city-scale learning loops. It might even produce a more governable hybrid order than America’s chaos.
The Party can tolerate technical complexity. It is less comfortable with social and ontological pluralism. Agentworld means strange rooms, synthetic companions, AI-native associations, private cognitive orders, new loyalties, agentic subcultures, plural selves, and social forms that cannot be fully justified before they exist. China wants the machine to learn. It does not want the machine to generate new sources of social form. It wants agents, but not Agentworld. It wants AI+, but not a new subject.
Late Qing failed by under-absorption. The PRC may fail by over-absorption. It may train machine cognition so successfully into Party-state purpose that it blocks the new subject machine civilisation requires. The danger is not that China resists the machine. The danger is that it domesticates the machine into the wrong civilisation. 11. The institution that can answer The answer is not to defend “Team Human” by pretending humans are still uniquely intelligent in every important way. That is a losing move. The answer is also not to surrender to the machine ecology and call it destiny. The question is institutional. Can we build scaffolds where machine agency is legible, contestable, and interruptible? A real institution must be able to answer. It must know what acted. It must know which part was model inference, which part was data retrieval, which part was rule, which part was discretion, and which part was human approval. It must preserve enough memory for appeal without making every record permanent destiny. It must allow conflict, because smooth agreement can become collective delusion. It must create an address. Not every human can oversee every machine act. That is impossible. But consequential machine-mediated action must eventually meet an institution capable of reply. A system can generate an answer. An institution must be able to answer back. 12. The small room is a prototype The first AI-native institution may not be the state, firm, school, or court. It may be the small room. Not as a hiding place. As a prototype. A room with humans, agents, memory, disagreement, roles, consent boundaries, exit, and return. (More on this in future essays) A room where one model drafts, another attacks, another retrieves memory, another simulates an opponent, another translates, another calms fear. A room where humans use machines to return to one another, not to escape one another. A room where the model can parse the blood result, but another human sees the face. The door matters. A door is not a metaphor. It is a kill switch. It is a consent boundary. It is the right to pause the loop. It is the right to say: not now, not this way, not without appeal, not without another human in the room. This is not enough. A small room cannot govern Agentworld. It cannot defeat markets, training states, platform mirrors, or machine-speed capital. But it may contain the seed of the scaffold that can answer. It is Agentworld at human scale. Multiple humans. Multiple agents. Memory. Conflict. Translation. Refusal. Return. Its danger is enclosure. Its promise is orchestration with an exit. 13. The wager
Absorption. Reservation. Replacement. The old subject does not negotiate its upgrade. It is broken, preserved, narrated, or remade by the order that follows. That is the dark grammar. Agentworld offers a different possibility: not one successor species replacing humanity, but a jagged hybrid intelligence made of humans, agents, institutions, simulations, and languages. That is the weird grammar. Settlement is the wager that the weird grammar can beat the dark one. It is a bet against the historical odds. The first civilisation that builds institutions capable of binding machine agency to human standing will have done something rare. It will have encountered the outside without denying it. Metabolised the outside without worshipping it. Answered the outside without pretending to remain unchanged. That is the work. A machine civilisation is not complete when software can bind. It becomes a civilisation only when people can bind the machine in return. Otherwise, the machine age will still have a settlement. It just will not be ours. Thanks for reading The Cutting Floor Newsletter! Subscribe for free to receive new posts and support my work. The Cutting Floor Newsletter is free today. But if you enjoyed this post, you can tell The Cutting Floor Newsletter that their writing is valuable by pledging a future subscription. You won't be charged unless they enable payments.
© 2026 Zufansan
|
