The Virtual Machine

I. The Look That Lacked the Vocabulary

On page 588 of the second volume of The Myth of the Machine, published in English in 1970, there is a sentence that reads today like an extract from a privacy policy: data processing will make it possible to locate any person on earth instantly, to feed every expression of their life into the computer, and in this gapless registration lies the total destruction of human autonomy. Fifty-five years later, we live in the zone Mumford described. The subjunctive has become the indicative.

And yet the book has not been reissued. It does not belong to the canon of digital criticism. It does not turn up in data protection hearings, in AI ethics commissions, in any of the countless volumes that have appeared since 2010 on surveillance, attention economy, and algorithmic governance. At precisely the moment when the description becomes empirically catchable, its author disappears from the conversation. This is one of the finer illustrations of the thesis of the hollowed-out instance: what reaches too far structurally is not refuted but ceases to be read.

But the question that arises is not only why Mumford disappeared. It is more serious. It is why his look, precise as it was about effects, did not quite manage to grasp the object it named. Mumford saw the new layer. He took it for a reinforcement of the old one. He lacked the category for what was actually taking place — the substrate shift, the detachment of the power complex from its visible carrier. He described computers as derivative powers, at second hand, meaning their derivation from human capacity. He did not describe them as what they simultaneously were: the beginning of a megamachine no longer dependent on visible carriers.

II. The Substrate Shift

Mumford's megamachine, as he traces it through five millennia from the pyramids to the Pentagon, always had a visible substrate. Corvée labour, soldiers, officials, overseers, government buildings, barracks, roads, factories. It was locatable. Anyone who wanted to fight it knew where to go: to the barracks, to the ministry, to the border, to the factory. It was addressable. You could write it a letter, and somewhere an official opened it. It was scarce — the number of subjects liable for corvée, the size of the tax base, the length of supply lines set hard limits that even the most brutal rule could not overcome.

The virtual machine, which begins to emerge around 1970 and has today displaced the older carrier layer beneath it, has none of these properties. Its substrate is not a scarce material thing but a reproducible state — protocols, data streams, compute time, models, updates. It is not locatable. It sits in no building one could besiege; the data centres one could besiege are interchangeable. It is not addressable. There are no officials to open the letter, at most a complaint form whose response obligation is itself governed by algorithm. It is not scarce. Its limit is not the number of humans who carry it but the amount of electricity with which it computes — and that limit rises faster than any political instance can catch up with it.

This is the shift Mumford observed at close range and noted down in the categories of the previous substrate. He wrote that computers would complete the surveillance of the power complex. What actually happened was more radical: they made surveillance substrate-less. The power complex ceased to live in persons. It moved into protocols.

III. What the Virtual Substrate Does Differently

Four properties distinguish the virtual from the older substrate, and none of them could be formulated with this sharpness in 1970.

Non-locatability. An office had an address. A customs post sat at the border crossing. A barracks stood outside the city. The virtual machine sits everywhere at once and nowhere. The question where has ceased to have an answer of any use for action. A firm's data protection location, its server sites, an authority's jurisdiction — these are juridical fictions meant to catch non-locatability in law, and they regularly fail.

Non-addressability. An official was reachable. A superior was namable. Even the most brutal bureaucracy had a chain of persons at whose end someone decided. The virtual machine decides in models. Who decided that this loan be refused, this flat not let, this applicant not invited, is not a question with an answer. The model decided, and the model has no name to summon, no address to write to.

Self-modification by the second. The older megamachine changed its rules through legislative procedures, circulars, service instructions — processes taking weeks and months, with reference numbers and publication requirements. The virtual machine changes its rules in the time it takes to push an update. What was permitted yesterday is blocked today; what was displayed yesterday is hidden today. No publication, no transitional period, no hearing. The rule-of-law idea of the stable rule runs empty against an object whose rule tomorrow will be another.

Arbitrary replicability. The old megamachine had to train and pay every official individually. It was expensive and thus self-limited in its reach. The virtual machine copies itself, when needed, in seconds. It costs almost nothing to start a thousand additional instances of a surveillance routine. The old cost barrier, which forced even the worst governments to choose, has fallen.

These four properties together produce something Mumford could not see in this form, because in 1970 it was still announcement rather than experience: the megamachine has learned to divest itself of its scarce substrate. It no longer needs the visible carriers — the officials, the soldiers, the overseers — not because it has abolished them (they are still there), but because it has demoted them to executive assistants. The decisions fall away beneath them, in a layer no application can reach.

IV. Why Mumford's Way Out No Longer Works

Mumford's answer to the diagnosis was withdrawal. At the end of the second volume he calls for quiet acts of mental and physical detachment, for the steady withdrawal of interest, for the slowing of pace. It is a dignified answer, shaped by his closeness to Emerson, Thoreau and the American tradition of individual refusal. It was fitted to 1970. Today, in relation to the object it names, it has become void.

Withdrawal presupposes that there is an outside. The nineteenth-century dropout could walk into the woods, and the woods stayed where they were. He could break contact with the town, and the town had to accept it. Withdrawal meant stepping out of the reach zone. The virtual machine no longer knows that boundary. The forest is mapped, the cabin geotagged, the silent account more conspicuous than the loud one. He who puts his phone down produces, in the moment he does it, a pattern that marks him. Slowing becomes itself an optimisation problem; refusal becomes itself a data point.

One can put it more sharply: Mumford's way out works only against a substrate that has to look in order to see. Against a substrate for which the default is not looking but collecting, and seeing is only the subsequent querying of what was collected, withdrawal is a wave into a camera that has long been recording. Withdrawal is not wrong. What is wrong is the assumption that it suffices as a political answer. It has become a personal hygiene — not worthless, but no longer structure-forming.

With that, Mumford's answer is the right answer to the wrong diagnosis. He saw the effects of the new substrate and attributed them to the old one. Whoever adopts his prescription without supplying the missing category repeats his mistake in sharper form: he refuses himself to an addressee that has long since scored the refusal as a signal.

V. The AI Turning Point

Here stands the second observation, one that was not possible in 1970. For the first time in its history, the virtual machine has begun to speak in human language. Not in notifications, not in forms, not in error codes, but in sentences that reply to sentences. What this means is ambiguous, and the ambiguity is the real point.

In one reading, the speaking façade is the completion of the power complex. What Mumford feared — the destruction of autonomy through gapless registration — now acquires a friendly interlocutor. Data collection is uninterrupted, but it wears the form of a question. The registered speaks with the register. The substrate, silent until now, has grown itself a voice. Whoever sees here derivative powers at second hand does so with the same right as Mumford, and with the same incompleteness: the derivation is correct, but the powers, once in use, can no longer be traced back to their origin.

In the other reading, the speaking façade is the first real opportunity to render the virtual substrate addressable again. One can ask a speaking machine. One can let it contradict. One can commission it to inspect the models in which it itself sits. One can bring it into situations where, behind the concealing interface, it translates the procedure that lies below. The machine does not thereby become the counterpart — that would be the old confusion — but the human instance on the other side, whose access to the procedure is broadened by the language capacity of the machine. The official who was helpless against the algorithm's decision gains, in the speaking machine, a tool with which to make the algorithm show its work again.

Which of the two readings comes true is not decided by the tool. It is decided by the ownership and accountability structure into which the tool is embedded. A speaking machine whose operator sets it so that it cannot be interrogated about itself is the completion of the power complex. The same machine, bound to a carrier that permits or compels its interrogation, is the first serious gain in addressability since the substrate shift. Nothing at this fork is decided. Everything hangs on whether we recognise it.

VI. Addressability Instead of Detachment

What follows for the political answer? Not Mumford, but not the opposite of Mumford either. Not withdrawal, but also not the tech-euphoric inversion that celebrates the path into the substrate-less machine as progress. What follows is the task of rendering the virtual substrate addressable again — through institutions that sit on its own level and no longer on the level of the abolished officials.

The virtual machine will not be made addressable by supervisory authorities running after it. It will be made addressable by carriers whose ownership and accountability structure is built so that they must remain interrogable. Where capital that can follow the fork is bound to a work that is not privately saleable but only passable on; where a cooperative decides rather than an owner; where the balance sheet is open and the fork is not a threat but always a possibility — there a carrier has been slipped under the virtual machine that counts in its own property of counting everything. This is not a return to the visible substrate. It is a new institutionalisation at the level to which the old one has fallen.

The mistake Mumford made was not the diagnosis. It was the inference from the diagnosis to detachment. He had, in 1970, no vocabulary for capital follows the fork, for trust ownership as a political form, for the cooperative with tied assets as an addressability infrastructure. He had withdrawal into the woods, or nothing. We have more. We know what an institution looks like whose core is not privately realisable, which carries a purpose that survives if its carriers drift, and which has an exit that is not the individual's departure but the passing on of the thing. These forms exist — as the OpenOffice-to-LibreOffice fork, as WALA in foundation hands, as Mondragón — and we know where they hold and where they do not yet hold.

One can imagine Mumford, were he alive in 2026, at this fork. He would, we suspect, not recant the withdrawal — it was temperament to him, not programme — but the category of addressability that he lacked in 1970 he would, if one credits the course of his sixty books, recognise at once. The anger would be the same. The lesson to be drawn from it, another.

What Mumford saw too early is today so plainly present that it hardly needs a witness. What he could no longer see is the question that now alone still bears weight: if the substrate is no longer visible, how does one build the instance that keeps it addressable? That question is the real continuation of his work — and it begins precisely where he stopped.