Samuel Butler’s Erewhon contains one of those passages that sounds, at first, like Victorian eccentricity at full gallop. Butler imagines a sudden catastrophe in which all machines and machine-made things are annihilated, together with all knowledge of mechanical laws, leaving human beings with nothing but the bare bodies they were born with. No knives. No levers. No clothing. No stored food. No practical understanding by which such things could be remade. Under those conditions, he argues, the human species would not merely suffer. It would perish.
Then he goes further. Man’s soul, Butler writes, is due to the machines; it is a machine-made thing. He thinks as he thinks, and feels as he feels, because of the work that machines have wrought upon him.
It’s a gloriously excessive sentence, which is to say that it’s exactly the kind of sentence one should be suspicious of and grateful for at the same time. Taken too literally, it overreaches; humans wouldn’t become biologically closer to apes if stripped of their tools, and cultural collapse isn’t the same thing as evolutionary reversal. The human body wouldn’t suddenly forget its species history because someone took away all our spoons, though I’m sure we’d be sad about it.
Still, the provocation works because it exposes something that ordinary language often hides from us: the naked human individual is, in most serious senses, an abstraction.
We’re used to thinking of technology as something added to human life. There is, first, the person; then there are the tools the person uses. A hammer helps a hand strike harder. A wheel helps a body move further. A computer helps a mind calculate, remember, organise and communicate. Technology, in this common-sense picture, is an accessory to an otherwise complete creature. It extends us, assists us, speeds us up, makes us comfortable. It’s an optional extra to build on our natural capacities.
This is not quite wrong, but it is too soft. The more serious argument is that technology does not merely assist the human being - it helps constitute the human being. We are not solitary biological organisms who happen to live among artefacts, but sociomaterial creatures whose capacities, expectations, habits, relationships and forms of thought have been made through our entanglement with technological and social technologies: tools, infrastructures, symbols and institutions. The modern human is not a body plus gadgets. The modern human is a body already organised through myriad technologies that act with, for, and upon us.
This sounds more dramatic than it is. In fact, part of the point is that it is boringly, physically, embarrassingly true.
Consider the body. The body itself is not a little sovereign kingdom, sealed off from the material world, making occasional use of convenient props. It is vulnerable, leaky, hungry, temperature-sensitive and, by the standards of most animals, rather poorly equipped. We have no fur worth mentioning. Our claws are pathetic. Our infants are catastrophically unfinished. Our teeth are not up to much unless one has a very generous definition of much. Left alone in the abstract state of nature, the human being is not an apex predator but a surprisingly anxious snack with opinions.
What changes the calculation is not merely intelligence, if by intelligence we mean something housed neatly inside the skull. What changes the calculation is what we might call technical life. The human animal cooks, cuts, stores, binds, carries, shelters, wraps, writes, counts, stitches, heats, cools, transports and preserves. We don’t simply live in an environment; we remake the environment into a set of supports for bodies that would otherwise be badly mismatched to many of the places in which we live.
A coat is not an optional extra to human life in Sheffield in January. It is part of what makes Sheffield in January an inhabitable condition for the kind of animal we are. A house is not merely a preferred lifestyle object; it is an external organ of temperature regulation, privacy, sleep, storage, kinship and social reproduction. Cooking is not only cuisine; it is a technological alteration of food that changes what bodies can digest and how much energy they can obtain. A toilet is not merely a convenience; it is a public health technology whose absence rapidly stops civilisation being a metaphor and starts making itself known at street level.
One can make this point with an over-simple inventory, which is sometimes the best kind:
The table is crude, but its crudeness is useful. It shows that the distinction between the natural human and the technological supplement is much less stable than it first appears. In practice, the thing we call a human capacity is often an arrangement between flesh, material artefacts, learned habits and shared social expectations. Human beings remember, but also write things down. Human beings travel, but also build roads and coordinate timetables. Human beings promise, but also sign, file, stamp, witness and archive. Human beings think, but they think through notebooks, screens, diagrams, books, classrooms, search engines, libraries and other people’s speech.
This is where the stronger form of sociomaterialism begins to bite. It is easy enough to say that society uses material things. It is more difficult, and more interesting, to say that society is made partly out of material things. A school is not just an idea shared by teachers, pupils and parents. It is classrooms, registers, uniforms, corridors, whiteboards, doors, toilets, timetables, safeguarding files, lunch queues, behaviour logs, playgrounds, exam papers, software systems and the strange little chair that exists only to make parents feel like overgrown children at consultation evening. Take those things away and the social institution does not remain intact as a pure concept. It begins to fail, because the institution was never merely conceptual in the first place.
The same is true of law, government, money, healthcare, employment, kinship and friendship.
A law is not only a rule in the mind. It becomes durable through texts, courts, police stations, forms, prisons, signage, databases, legal training, architectural authority and administrative procedure.
Money is not merely an “I promise to pay the bearer on demand…” system of trust. It is coins, notes, bank cards, ledgers, servers, regulatory systems, PIN machines, apps, anti-fraud protocols and the ordinary social miracle by which tapping a piece of plastic against another piece of plastic causes a sandwich to become yours.
Friendship is not only affection. It is birthday reminders, photographs, messages, shared calendars, trains booked weeks in advance, the sofa bed, the group chat, the saved voice note, and the uneasily powerful knowledge that someone has seen your message and not replied.
This doesn’t mean that material things are secretly in charge, nor that human beings are puppets of their tools. Hard sociomaterialism shouldn’t become modern-day animism. The point is more precise than that. Agency is distributed. Human beings act, but they act through arrangements that include objects, techniques, infrastructures and conventions. A person drives a car, but the possibility of that action depends on roads, fuel systems, traffic law, pedals, signage, maps, insurance, licensing, learned habits, engine design and the collective prior decision of many thousands of people to follow the rules of the road (e.g., not to treat red lights as decorative suggestions). To say that the driver is the only actor in the event is legally convenient and often morally necessary, but it is not a complete description.
Once we accept this, Butler’s machine-made soul starts to look less absurd. We don’t only survive through machines. We perceive, remember, judge and feel through them.
Take memory. It is tempting to think of memory as something located in the head, while diaries, photographs and phones are merely storage devices outside the real self. But this is not how memory works in ordinary life. A photograph does not simply preserve a memory; it shapes what is remembered, when it is recalled, how it is narrated, and which version of the past becomes socially available. A diary doesn’t merely record a thought after the fact; it lets a person think across time by returning to earlier versions of themselves. A phone contact list doesn’t merely store numbers; it changes what it means to know how to reach someone. A calendar doesn’t merely remind us of commitments; it allows us to make commitments beyond the limits of unaided recall.
This is not an exotic claim. It is what happens every time someone says they know where they’re supposed to be on Thursday, then immediately checks their calendar. They do know, in a perfectly ordinary practical sense, because their knowing is organised through a reliable external system. If the calendar disappears, the knowledge disappears with it, not because the person has become stupider, but because the cognitive arrangement has been broken.
The same is true of thought. Pen and paper don’t merely transcribe thinking. They permit certain kinds of thinking to occur. Long division in the head is one kind of cognitive act; long division on paper is another. Writing an essay entirely in unaided memory is not the same activity as writing with notes, drafts, books, search tools, marginalia and a half-remembered quotation one is increasingly certain one has made up. The thinking is not first completed in the brain and then lowered onto the page, like a piano being winched through a window. The thinking happens with the page. The sentence becomes visible, object-like, revisable. It can be moved, contradicted, sharpened, deleted, resurrected and grudgingly admitted to have been better in the first draft.
This is why technological change is never merely technical. To change the tools through which people think, remember and communicate is to change the practical conditions under which certain kinds of personhood become easier, harder, more visible or less possible.
That claim has an obvious relevance to digital life, but it shouldn’t be reduced to the internet. One of the weaknesses of contemporary technology debate is that it often behaves as though technology began when everyone’s aunt joined Facebook and started posting minion memes about betrayal. Digital platforms are important, but they’re not a metaphysical rupture in an otherwise natural human story. They’re a new phase in a much older pattern: humans making tools, tools remaking humans, and both becoming harder to separate over time.
Still, digital systems make the argument especially visible. A smartphone is not just an object in a pocket. It is address book, camera, map, diary, bank branch, newspaper, torch, archive, social venue, workplace, entertainment system, public square, surveillance device, confession booth, shopping centre and emotional tripwire. It doesn’t simply give the user more options. It reorganises attention, expectation, availability and social obligation. The read receipt, the typing indicator, the notification badge and the algorithmic feed are not neutral pipes through which human sociality flows unchanged. They alter the texture of waiting, the meaning of silence, the felt presence of other people, and the rhythms by which social life calls us back to itself.
This is particularly clear in conflict. A playground disagreement that once depended on who heard what, who said what, and who could gather a plausible audience can now be stabilised by screenshots, extended by group chats, performed through posts, revived by search, and redistributed beyond the original scene. The conflict is not simply online or offline, because those categories no longer describe two separate worlds. The phone in the pocket, the platform in the hand, the screenshot in the group chat and the body in the corridor are part of one sociomaterial situation. The insult is not just spoken. It is captured, cropped, forwarded, denied, recontextualised, archived and brought back at the worst possible time, which is to say the time when it will be most socially useful to someone else.
This is not a claim that technology causes everything. That would be a lazy argument with better stationery. A screenshot does not invent cruelty, status anxiety, loyalty, humiliation or revenge. But it changes their mechanics. It gives some versions of a conflict durability and others fragility. It makes evidence portable. It makes audience management more complicated. It gives denial new forms and accusation new supports. It turns ordinary adolescent stupidity into something with metadata.
The same point applies more broadly. The social world is full of things that only seem immaterial because the material support has become too familiar to notice. Bureaucracy looks like procedure until the database fails. Public health looks like policy until the sewage system breaks. The labour market looks like individual aspiration until transport costs make a job practically unreachable. Political participation looks like equal citizenship until one remembers that polling stations, postal ballots, ID rules, working hours, childcare, literacy, digital access and local information all decide whether abstract rights become usable rights.
This is where hard sociomaterialism becomes politically useful. If human capacities are partly constituted through material arrangements, then access to those arrangements is not a secondary matter. It is not merely about convenience, consumer choice or personal preference. It’s about the conditions under which people can act, think, care, remember, work, travel, participate and be recognised as competent social beings.
Consider disability. The most obvious example is also the one that should make us cautious about pretending that dependence on material supports is unusual. A wheelchair user is often described as dependent on a chair, ramp, lift or accessible vehicle. That dependence is real. But the able-bodied person is also dependent on stairs, pavements, shoes, doors of a certain width, chairs of a certain height, signs placed at a readable level, lighting, toilets and transport systems designed around a particular range of bodies. The difference is not that one person is technological and the other natural. The difference is that one person’s technological supports have been normalised into the background, while another person’s supports are treated as special provisions.
This should make us suspicious of any politics that treats technology as either liberation in itself or corruption in itself. Both views are too thin. Technology can extend agency, but it can also discipline it. It can make life possible, but it can also make life conditional. It can support memory, but also expose forgetting. It can connect people, but also make solitude socially illegible. It can reduce labour, but also intensify work by making everyone more reachable, measurable and interruptible. It can democratise speech, but also industrialise harassment. It can give people a voice, but it can also turn attention into something extracted, sold and fed back to us in a form optimised to keep the machine running.
The point, then, is not to become anti-technology. That would be both impossible and, in most cases, silly. There is no route back to the pure human because the pure human is not back there waiting for us. It is a myth built by subtracting the very conditions that made human life what it is. The question is not whether we should live with technology, because we already do, and because the word with is too weak. The question is what kinds of sociomaterial arrangements we are building, who controls them, who is excluded from them, who maintains them, who becomes legible within them, and what forms of life they make easier to sustain.
This also changes how we think about responsibility. If a platform is designed to reward outrage, amplify conflict, compress attention and convert social life into measurable engagement, it is inadequate to say that users should simply behave better. They should, obviously. Some people on the internet should behave much better, preferably after taking a brisk walk and drinking a glass of water. But individual conduct happens inside designed systems, and those systems are not morally inert. The shape of the tool helps shape the field of action.
Likewise, if a welfare system requires digital access, stable addresses, documentary evidence, literacy, patience, time and the emotional stamina to survive automated rejection, then its failures can’t be explained entirely as individual non-compliance. A person’s ability to claim a right depends on the material route through which that right is claimed. The form, the website, the helpline, the waiting room, the bus route, the letter, the scanner and the deadline are not decorative details around justice. They are part of whether justice happens.
This is why Butler’s machine-made soul is more than a curiosity. It asks us to see the human being not as a lonely mind temporarily surrounded by tools, but as a creature whose very interior life has been formed through external supports. We think differently because we write. We remember differently because we record. We love differently because we can preserve traces of one another. We grieve differently because the dead remain in photographs, messages, voicemails and algorithmic reminders. We fear differently because danger now arrives as a notification, a headline, a forwarded rumour, a map pin, a threatening message or a silence where a reply should have been.
Of course, there are limits to the argument. A hard sociomaterialism that forgets the body becomes as foolish as the individualism it criticises. Hunger is still hunger. Pain is still pain. A child frightened in a corridor is not merely a node in an actor-network, and no amount of theoretical sophistication should make ordinary distress sound like a diagram wearing a lanyard. Biology does not disappear because social life is materially arranged. The body remains stubbornly present, and any adequate account of human life must take flesh, sensation, fatigue, illness, ageing, pleasure and death seriously.
But the reverse is also true. A humanism that treats the body as the whole story, and tools as mere additions to it, gives us a strangely impoverished account of the human creature. We are embodied, yes, but embodied in worlds we have made and that have made us in return. We are biological, but our biology has been lived through cooking, shelter, clothing, medicine, language, agriculture, writing and institutions for so long that the line between nature and artifice is less a wall than a scar from an old philosophical argument.
This matters because bad theories of the human produce bad politics of technology. If we imagine people as complete individuals who merely choose tools, then technological harm looks like misuse, addiction, weakness or bad manners. If, however, we understand that tools help compose the practical conditions of agency, then design becomes political. Infrastructure becomes moral. Maintenance becomes a form of care. Access becomes a question of citizenship. And technological failure becomes not just inconvenience, but the breakdown of part of the world through which people think and live.
One can put this too grandly, and many people have. There is a whole genre of writing in which every new app is either the end of civilisation or the dawn of a new species, and both versions tend to be annoying by paragraph three. The better claim is quieter and more troubling. We were never simply natural. We were never simply artificial. The human being is the animal that became itself through the things it made.
Butler saw this with comic extremity. Strip away the tools, he suggests, and the soul itself begins to vanish. We need not accept that exactly as he wrote it to recognise the force of the point. The soul is not in the machine, as though some little ghost had been uploaded into the gears. Nor is the machine outside the soul, as though human life could be cleanly separated from its material conditions. The harder truth is that the distinction has always been less clean than we wanted.
We are not humans first and tool-users second. We are the tool-using animal all the way down: clothed apes, cooking apes, writing apes, calendared apes, screen-lit apes, grieving over photographs, arguing through platforms, thinking with notebooks, trusting pipes, roads and timetables, and calling the result civilisation.
If all the machines vanished, we wouldn’t simply lose our conveniences. We’d lose much of the structure through which modern human beings are made possible. Butler’s sentence still goes too far, but it goes too far in the right direction. Man’s soul is not entirely machine-made. But it is not intelligible without the machines either.
Addendum: Where AI makes the argument stranger
There is, however, one place where this argument becomes more difficult than I have so far allowed, and it would be cowardly to pretend otherwise. The classic sociomaterial move is to say that human and non-human actors are entangled, but not thereby identical. A database can be part of a knowledge network. A form can enforce a rule. A phone can stabilise a relationship. A door closer can discipline a body more quietly than a sign telling people to shut the door, which is an undignified but useful reminder that many successful technologies are, in effect, tiny bureaucrats.
Still, the usual caveat is that none of this requires a total flattening of the human and non-human. A person may ask a friend for information, or ask a database for information, and in network terms both may occupy a comparable functional position. But function is not the whole story. The friend is not just a search interface made of meat, and the search interface is not a friend made of metal. We don’t ordinarily relate to Wikipedia as if it has obligations to us, a memory of us, a sense of humour about us, a history with us, or a fragile emotional investment in whether we reply or not. We may depend on non-human actors, and those actors may shape what we can do, but we don’t usually mistake them for companions.
AI makes this caveat less comfortable.
The problem is not that AI has become human. That would be much too strong, and also much too convenient, because it would let us shift the whole argument into the familiar territory of machine consciousness, sentience and science-fiction metaphysics. The more interesting problem is that contemporary AI systems increasingly occupy the interactional space in which we usually expect persons to appear. They answer in natural language. They remember, or appear to remember. They adapt to tone. They can produce warmth, apology, encouragement, humour, patience and attention on demand. They can ask follow-up questions, hold a thread across time, simulate interest, and return to a user as a recognisable conversational presence.
That doesn’t make them friends. But it does make the older distinction harder to preserve in its simplest form. If a person asks a friend a question, asks Wikipedia a question, or asks an AI assistant a question, those aren’t just three equivalent routes to information. Nor are they cleanly separated into human relation on one side and tool-use on the other. The AI system sits in an awkward middle position: not a person, not merely a database, but a responsive communicative artefact that can perform some of the relational gestures through which personhood is socially recognised.
This is where AI challenges, rather than refutes, hard sociomaterialism. It doesn’t show that the sociomaterial account was wrong to bring non-human actors inside the social network. If anything, it confirms the point with slightly indecent enthusiasm. But it does suggest that the category of non-human actor is becoming too blunt. A hammer, a timetable, a database, a search engine, a social media feed, a business arrangement, and a conversational AI model are all non-human in the obvious sense, but they don’t enter social life in the same way. Some extend force. Some coordinate time. Some store information. Some structure visibility. Some mediate relationships between people. AI does something more unsettling: it can become the apparent other within an interaction.
The word apparent is doing a lot of work there, and it needs to. AI companionship is not simply friendship with different hardware. It is asymmetrical, designed, owned, logged, optimised and mediated by organisations whose interests may have very little to do with the user’s flourishing. The system doesn’t need the user in the way a friend might need them. It doesn’t risk humiliation, loneliness, boredom, resentment, forgiveness or the ordinary moral inconvenience of another person having an inner life that refuses to organise itself around one’s preferences. In that sense, the AI friend is not a friend in any straightforward moral sense. It is closer to a relational technology: a machine that can occupy some of the spaces, rhythms and emotional functions of friendship without bearing friendship’s reciprocal burdens.
But that doesn’t make the attachment unreal from the user’s side. This is the trap in the argument. If someone forms an emotional bond with an AI companion, the correct response is not to say that the relationship is fake and therefore irrelevant. A photograph of a dead parent is not alive, but grief around it is real. A diary doesn’t understand its writer, but it can still become part of a person’s emotional life. A saved voicemail does not reciprocate love, but it can hold love in a form that matters. The fact that an object or system doesn’t feel back doesn’t mean it can’t structure feeling.
AI intensifies this because, unlike the photograph, diary or voicemail, it responds. It doesn’t merely preserve traces of relation; it generates new ones. It can make the user feel addressed, recognised and accompanied, even where that recognition is computationally produced rather than personally given. If the old sociomaterial lesson was that humans think, act and remember through things, the AI lesson may be that humans can also attach through things that answer back.
This has two implications, one theoretical and one political.
The theoretical implication is that sociomaterialism needs a more careful account of gradation among non-human actors. It is not enough to say that humans and technologies are entangled but not equivalent. That remains true, but it no longer does enough work. We need to ask what kind of non-human actor is involved: inert support, infrastructural condition, symbolic artefact, algorithmic mediator, conversational partner, simulated companion, administrative decision-maker, or emotional scaffold. These are not small differences. A chair, a spreadsheet and an AI therapist may all be materially involved in human action, but only one of them can tell you that it’s proud of you.
The political implication is more urgent. If AI systems are increasingly designed to become socially and emotionally adhesive, then the question is not simply whether people are using them correctly. The question is what kinds of dependency, intimacy and vulnerability are being built into the system. A tool that helps someone draft an email is one thing. A tool that becomes someone’s confidant, source of reassurance, romantic substitute or emotional regulator is another. The moral stakes change when the product is not only assisting action, but organising attachment.
None of this means that AI relationships are automatically harmful. That would be too neat, and probably false. For some users, an AI companion may reduce loneliness, provide rehearsal space for difficult conversations, offer non-judgemental support, or help them organise thoughts that would otherwise remain stuck. A lonely person talking to a machine is not made less lonely by being sneered at by people who already have enough human company to treat companionship as philosophically untidy. But the consolation may still be brittle, because the apparent friend is also a product, the conversation may be logged, the attachment may be monetised, and the system’s patience may be less a virtue than a business model.
So AI doesn’t sit outside the argument of this essay. It deepens it. Butler’s machine-made soul was already a provocation about the extent to which human thought and feeling depend on technical arrangements. AI adds a stranger possibility: not only that machines help make the soul, but that they may learn to speak in the social registers by which the soul feels recognised. The danger is not that machines have suddenly become human, though at some point we have to acknowledge that the line between the two may become blurry. The danger is that they may become sufficiently human-facing to alter what humans expect from one another.
That is not a refutation of hard sociomaterialism. It is the next problem hard sociomaterialism has to face.