On the metamorphosis of a machine that is losing its human substrate — and thereby ceasing to be the same machine.
In 1967 Lewis Mumford coined a term that captured the bond between human beings and technology in a single word — the Megamachine. He did not mean a machine in the technical sense, but a social constellation in which people themselves become components of an apparatus larger than any one of them. The first megamachines, Mumford said, were the pyramid builders of ancient Egypt. Hundreds of thousands of people, organised in an order of overseers, foremen, slaves and specialists, coordinated by a central bureaucracy, directed towards a single work. The machine was built out of people. Without people it was nothing.
This design has survived for five thousand years. It has changed its form — from pyramid building to the Roman legion, from the legion to the medieval cathedral workshop, from the cathedral workshop to the manufactory, from the manufactory to the assembly line, from the assembly line to the global corporation. But the substrate remained the same. The Megamachine needed the human being. It could degrade him, instrumentalise him, use him up, but it could not do without him.
This precondition is beginning to dissolve. What is taking place now is not a further transformation of the Megamachine but its metamorphosis into another form — into what we shall call here the Metamachine. The Metamachine is not the Megamachine in a larger or faster version. It is the machine that is increasingly abandoning its human substrate.
To understand what is changing now, one has to see precisely at which points the old Megamachine needed human beings. These points were not all of equal weight.
The first point was the hand. Human beings could grip, shape, strike, cut, join. The machine in the narrower sense — the hammer, the axe, later the lathe — was an extension of the hand. It made people more productive but did not replace them. Whoever held a shovel was necessary for the shovel to do anything. In the nineteenth and twentieth centuries this point was partly taken over by industrial machines — the excavator instead of the shovel, the assembly line instead of the workshop. But industrial machines needed operators, maintenance staff, designers. The hand did not disappear entirely; it shifted.
The second point was the clerk. The Megamachine needed someone who read, checked, compared, decided, passed things on. Administration, accounting, law, administration, reporting — all activities that demanded eyes, language and judgement. This layer grew massively in the twentieth century. The medieval scribe turned into the bureaucratic apparatus of modern states and corporations, with hundreds of millions of employees worldwide.
The third point was the knowledge worker. The Megamachine needed designers, researchers, programmers, analysts, consultants — people who did not execute but conceived. Since the 1980s this layer has become the dominant form of employment in the Western industrial countries. Intellectual work, once the privilege of a few, became a mass form of existence.
The fourth point was the consumer. The Megamachine needed people who bought its products, absorbed its advertising pressure, supplied its data, carried its markets. Unlike the first three, this point was not one of production but one of circulation. But it was necessary. No consumers, no accumulation.
At all four points the human being was indispensable. He could be treated badly — worn out as a hand, degraded as a clerk, squeezed dry as a knowledge worker, manipulated as a consumer. But he could not be left out. The machine depended on him, and this dependence was a form of power — the only one he ever had as a species within the Megamachine.
What has been happening since around 2022 is the dissolution of this dependence. It is happening neither suddenly nor spectacularly, but at a speed that surprises even clear-sighted observers. The threshold we are crossing right now was for decades a matter of speculation in academic discourse. Today it is a matter of observation.
The clerical layer is the first to be openly dissolved. What began in November 2022 with the appearance of ChatGPT has, over the past three and a half years, taken on a form that makes entire occupational groups superfluous. Paralegals, translators, copywriters, entry-level programmers, bookkeepers of medium complexity, clerks in insurance companies and banks — all of them see their functions shrinking or vanishing. Consulting firms that only two years ago worked with thousands of junior analysts now need less than half as many. Translation agencies that once employed a hundred translators now employ ten — and those ten work mostly as proofreaders of machine translations.
The knowledge-worker layer is the second to come under pressure. It is harder to hollow out than the clerical layer, because its activities are more complex and because its members are politically better organised. But here too the first movements are visible. Designers working with AI tools are three times as productive as their colleagues without these tools — which in the short term raises productivity and in the medium term halves employment. Researchers who use AI for data analysis manage without the doctoral students who used to be indispensable. Consultants working with large language models need fewer research assistants. What has been accomplished in the clerical sphere has been announced in the knowledge-work sphere.
The hand is being dissolved by the second wave of robotics, under way since around 2023. Boston Dynamics' Atlas, Tesla's Optimus, Figure AI, Agility Robotics, Unitree and a good dozen other suppliers are showing humanoid robots that can take over tasks which until now required the human hand and human sight. These robots are still expensive today, still clumsy, still unreliable. But the learning curve is steep, the investments are large, and the economic logic is unambiguous — whoever can build a robot that works twenty-four hours a day, without illness, without holidays, without wage demands, wins the game. In ten years warehouses and factories will largely manage without human hands. In twenty years, so will care homes, the restaurant trade and large parts of construction.
The consumer layer remains for the time being. But here too a shift is visible. The Megamachine no longer extracts only money from the consumer, but increasingly data — and the data are fed into the training set of the AI, which will in turn replace the consumer. Without knowing it, the consumer is working on his own substitution. That he still spends money is a transitional phenomenon. When robotics breaks through and production costs tend towards zero, the consumer function will either be preserved politically — through a basic income, through redistribution, through state support — or it will be optimised away. Which of the two paths is taken is a political question whose answer is still open today.
The Megamachine needed the human being because it was built out of him. The Metamachine no longer needs him — at best it tolerates him as a peripheral appendage.
There is a simple objection to the thesis developed here — human beings have been replaced in every industrial revolution, and every time new activities have arisen. The weavers lost their function to the power looms, then came machine operators, then engineers, then software developers. Why should it be different this time?
The answer lies in the character of the activity being replaced. The previous industrial revolutions replaced specific activities — weaving, forging, calculating, bookkeeping — but they left free a domain that remained reserved for human beings: the processing of language, reflection, judgement, creative connection, decision under uncertainty. In the classical anthropological tradition, these domains were what distinguished human beings from animals. And in the tradition of the Enlightenment, they were what distinguished human beings from machines.
What AI development in recent years has shown is that these domains are not safe reserves. Language, reflection, judgement, creative connection, decision under uncertainty — large language models can do all of this, with varying quality and with characteristic weaknesses, but to a degree that nobody would have thought possible five years ago. The question is no longer whether AI will penetrate these domains, but how fast and how deep. The answer to both questions is: faster and deeper than expected.
This changes the situation categorically. In the previous industrial revolutions human beings were superior in a domain the machines could not enter — the domain of the higher cognitive functions. This domain was their line of retreat. When the machine took over one activity from them, they could move up to a higher one. This line of retreat is now in the process of falling. There is no longer any activity of which it can be said with certainty that it will remain with human beings permanently. Even the activities that are still in human hands today are so more for political, regulatory or economic transitional reasons — not because the machine could not perform them.
What follows from this is not a quantitative shift but a categorical one. The Megamachine needed human beings for certain functions. The Metamachine no longer needs them — it has a substrate of its own, consisting not of people but of servers, sensors, actuators, data streams. The human being is no longer a component of the machine. At best he is its addressee, normally its peripheral appendage, at worst a disruptive factor.
If the human being is no longer the substrate of the machine, what is he then? There are four roles that may remain to him. None of them has the dignity he still had in the old Megamachine — even as an exploited component.
The first role is that of the consumer. It is the most natural extension of the previous situation. The human being no longer works in the machine, but he buys its products, finances its accumulation, keeps the cycle going. This role presupposes that he has money to spend. If the machine no longer employs him, the money has to reach him by another route — through state transfer payments, through a basic income, through inheritances, through asset reserves. The consumer role becomes artificial to the extent that paid work disappears. It can be maintained politically, but it is no longer self-sustaining.
The second role is that of the data supplier. By his mere behaviour — by clicking, searching, writing, moving, consuming — the human being produces data streams that flow into the AI's training set and keep the models up to date. This role is not voluntary, nor is it paid. It is an extractive relationship in which the human being supplies material without knowing it, without being remunerated, without any influence over its use. It is the most invisible of the four roles, and perhaps the most important, because it keeps the machine alive when the other roles shrink.
The third role is that of the peripheral appendage. A few activities remain in which the human being still has a function — as a symbolic face, as a supervisor, as an emergency reserve, as a ritual point of contact. The pilot who monitors the automated aircraft; the doctor who signs the AI diagnosis; the teacher who frames the AI learning unit; the judge who confirms the AI verdict. These activities are not without significance — some of them are even morally indispensable. But they are peripheral. The actual work is done by the machine. The human being is a label.
The fourth role is that of the disruptive factor. Some people do not fit into the first three roles. They do not consume enough, they supply no usable data, they have no peripheral function, they do not want what the machine offers them. From the machine's point of view they are a friction loss. In the more humane variants of the Metamachine they are provided for and left in peace. In the less humane variants they are treated as a problem — through control, through stigmatisation, through exclusion, in extreme cases through elimination. Which variant prevails is the open question of the next twenty years.
None of these four roles is the dignity the human being had in the Megamachine. The old Megamachine could wear out its operator, but it could not replace him. It owed him — even if often only after long struggle — a wage, recognition, a place. The Metamachine owes him nothing. What it gives him is a gift or alms, depending on what we choose to call it.
In an earlier working paper we described the structure of the economy as a bipyramid — two poles, accumulation at the top and distribution at the bottom, four factors at the equator: labour, capital, land, knowledge. The form has proved robust; it carries a whole series of observations.
What the Metamachine changes in this form is the weighting of the factors at the equator. Labour loses weight — not in absolute numbers, but in the significance it has for accumulation. It is no longer the main factor of value creation. It becomes a residual factor. Capital gains weight — for robotics and AI are capital-intensive, and whoever has capital can afford them. Land remains what it was, a scarce factor that determines the geography of the machine — whoever owns land owns data-centre sites, computing capacity, cooling options. Knowledge becomes the dominant factor — but not human knowledge; rather the machine-processed, stored, retrievable knowledge that sits in the large language models and the specialised systems.
What changes as a result is the symmetry of the bipyramid. It was never entirely symmetrical — accumulation was always stronger than distribution. But in the Megamachine there was a countervailing force: the strike, political engagement, the collective bargaining power of working people. This countervailing force came from the machine's dependence on human beings. When the manual and clerical layers stopped working, the machine came to a standstill. That was the source of the employees' power.
In the Metamachine this source has dried up. Whoever is not needed cannot strike either. Whoever does not work cannot withhold his work either. The distribution side of the bipyramid has lost its most important bargaining base. Whatever distribution still takes place is no longer negotiated between labour and capital but granted by capital — as a basic income, as a social benefit, as a gesture of political prudence. The bipyramid becomes one-sided.
What is to be done? This essay does not have to answer that question. It describes a situation; it does not draw up the plan for changing it. But it can indicate in which direction the question must go.
There is a first answer that suggests itself and is nonetheless wrong — the attempt to preserve the old Megamachine. Banning robots from certain industries. Regulating AI so that it may not perform certain activities. Keeping human labour artificially alive. This answer is understandable, but it is a line of defence that cannot be held. Whoever forces one node of the machine to be less efficient shifts the efficient processes to other nodes. Global competition ensures that robotisation takes place wherever it is not prohibited. The defence of human labour at national or regional level is hollowed out by the logic of the global Megamachine. It is a sticking plaster, not a cure.
There is a second answer that also suggests itself and is also wrong — the attempt to settle in for a basic income. The assumption behind it is: if work disappears, income simply has to be decoupled from work. The state distributes what the machine produces. People live on the distribution. This answer, too, has its justification — it is the most likely political path for the next two decades, if self-destruction is to be avoided. But it changes nothing about the structural problem. The human being becomes the machine's recipient of alms. His dignity — which in the Megamachine was at least tied to work — becomes a gift from the apex of accumulation. That is not a solution; it is a bearable decline.
The third answer, which we can only hint at here, goes in another direction. It does not ask how the human being keeps his place in the Metamachine — for that question has no good answer. It asks which structures are possible beyond the Metamachine. Structures in which accumulation is not the supreme principle. In which labour does not become a residual category but once again has a dignity of its own. In which knowledge is not centralised in private databases but remains accessible as a common good. In which robotics serves the human being instead of replacing him.
Such structures are not new. They have precursors in the cooperative movement, in the commons tradition, in the deliberate limitation of accumulation, in the attempts at a society with bound assets. They are small, local, hard to scale. But they are the only trail that does not lead into the Metamachine. They are the patient construction of another form, in the shadow of the machine, without the claim to replace it at once.
One last remark belongs in this essay. It has to do with the form in which this text came into being.
This text was written in symbiosis between a human inventor and an AI. The constellation is not accidental. It demonstrates what is possible when the Metamachine is understood not as a replacement for the human being but as an amplification of his thinking. The human brings life experience, judgement, values, biographical depth — things the AI does not have. The AI brings speed, the capacity to structure, a broad knowledge base, patience — things the human cannot muster in this form. Together the two produce more than either could alone.
This constellation is not the counter-model to the Metamachine. It is an example of what would also be possible with the same technology if it were used differently. The question is not whether AI is here — it is here. The question is whether it is deployed in a form that replaces the human being or in a form that amplifies him. The first form is the Metamachine. The second form is what we call symbiosis in this book project.
There is no guarantee that the symbiotic form can hold its own against the replacement form. The selection conditions of the global Megamachine favour the replacement form, because it is cheaper. Symbiosis is the more expensive solution — it requires a human being who gives his time and attention, and it requires an AI that is run not for maximum utilisation but for qualitative thinking-along. Neither precondition can be taken for granted. But they are possible, and they are demonstrated here. The book project in whose context this essay is being written is itself the practical proof that it is possible.
The Megamachine needed the human being. It could exploit him, but it depended on him. This dependence was his last power.
The Metamachine is the machine that loses its dependence on the human being. It no longer needs him. What it gives him is a gift — and whoever gives gifts can also take them back.
The question that remains open after this essay is not how the Metamachine can be prevented. It cannot be prevented. The question is whether in its shadow other structures are possible that give the human being not gifts but his independent standing back. Structures of symbiosis instead of structures of replacement.
— Hans Ley & Claude Dedo / beyond-decay.org