Who Owns the Deserted Hall?

I. The Calculation

The sentence in the motto comes from an interview that also contains the other sentence that is never missing from such conversations: the new machines are meant to relieve employees of physically strenuous work, not to replace them. The two sentences stand side by side as if they were compatible. They are not. An automation has not paid off so far because a human being was cheaper at that spot. When the calculation turns, the work disappears. That is what the calculation is for.

What is new about so-called physical AI is not the robot. What is new is that it no longer has to be built for a single movement. Today it picks parcels, tomorrow cable harnesses, the day after strawberries. This removes the boundary at which classical automation had to stop: where the work was too varied, too untidy, too small for a dedicated installation. Until now that was the last refuge of manual work, in the warehouse, in the field, on the assembly line. Nobody knows how fast this boundary will fall. That it will fall is no longer disputed by anyone in industry.

At the end there is not an empty hall but a deserted one, or a largely deserted one: the machines keep running, only people hardly stand in it any more. The question of this essay is simple: who owns it?

II. The Lever That Disappears

As long as a factory needed thousands of people, it depended on them. The workers could strike, they voted, they lived in the town around the plant. Everything employees have won in two hundred years, from the eight-hour day to co-determination, rested on this dependence. The trade unions never had any power other than this one: that without the work nothing runs.

This power is already weaker today than it looks. In 2025 just under half of all employees in Germany worked in a company covered by a collective agreement; at the end of the 1990s it was around three quarters in the West. The shortage of skilled workers masks this for the moment. It gives power to the individual who can choose an employer. It gives none to the organisation. Whoever is in demand negotiates alone.

Physical AI now removes the lever itself. A strike in a hall that hardly needs people shuts down little. That is the real shift of power, and it is larger than any shift in market shares. Whoever owns the installations no longer needs people to the same degree as before, neither as labour nor as neighbours in the town.

It also affects small and medium-sized firms, though not inevitably. When the CNC machine arrived, people feared the end of small workshops. It helped them, because it made small batch sizes economical. Physical AI promises above all flexibility, and flexibility actually helps the small. What matters is the form in which it reaches them: as a tool they buy and own, or as a service they rent while handing over their process data. In the first case they become stronger. In the second they deliver their knowledge piece by piece to the platform, until the platform no longer needs them. This is not a technical question. It is a question of ownership.

III. Ownership through Work

There are enterprises in which this question has already been answered. In the cooperatives of Mondragón the business belongs to those who work in it. When a machine replaces a person there, the yield goes to the members, who are at the same time the workers. Whoever is no longer needed at their post is not dismissed but transferred to another cooperative of the group. For Mondragón automation is therefore no threat. It is a gain for the owners, and the owners are the workforce.

In our essay “The Experiment That Did Not Want to Become a Model” (German) we described where this ends. Less than half of those employed in the group are members today. Mondragón has grown mainly through subsidiaries abroad whose employees have remained wage earners. We explained this with Franz Oppenheimer’s law of transformation: successful cooperatives prefer to hire wage labourers rather than admit new members who would share the yield with them. Our conclusion was the question of how the cooperative principle could be extended to the entire workforce.

That question presupposes that there is a workforce. Physical AI gives Oppenheimer’s law a new form. The cooperative of the future no longer has to hire wage labourers to keep the circle of owners small. It simply hires no one any more. It remains fair towards those already inside and becomes, towards the outside, a closed circle of owners of automated plants. That is better than a corporation, because the yield is spread over more heads. But ownership through work ends where work ends. Whoever is no longer needed no longer gets in.

IV. Ownership through Purpose

There is a second form in which ownership does not hang on shareholders: the foundation. The company belongs to it, and it itself belongs to no one. It depends not on persons but on a purpose. For our question that makes it more interesting than the cooperative, because a purpose needs no work in order to exist.

Ernst Abbe demonstrated this. In 1891 he transferred his shares in Zeiss to the Carl Zeiss Foundation and in 1896 gave it a statute whose purpose went far beyond preserving the company. The employees received enforceable rights, and the foundation was to promote science and the city of Jena. The yield of the company thus belonged to a community, not to an owner and not only to the workforce. Applied to the deserted hall this would mean: its yield belongs to the purpose for which it was founded, even when nobody works in it any more. And that purpose can be a region, a town, or the people who used to work in it.

In our essay “The Long Game, Lost” we showed what ails this form. At Bosch the supervisory board essentially supervises itself, because the same people sit on the trust and on the board. Ownership without owners also means control without liability. On top of that, most foundations today have a more modest purpose than Abbe’s statute: the preservation of the company. Such a foundation preserves the hall, but not the people. The foundation is therefore an answer only if its purpose reaches beyond the company and if someone from outside checks whether the administrators fulfil it.

V. Ownership through Belonging

There is a third form, and it comes from an unexpected direction, from oil production. Since 1982 the state of Alaska has paid every resident a dividend each year from a fund into which part of the oil revenue flows. In 2025 it was 1,000 dollars per head, for children as for pensioners. What decides the claim is not work but belonging alone: whoever lives in Alaska is a co-owner. Norway has built the largest sovereign wealth fund in the world on the same idea.

With oil the justification is simple. It lies in the ground, nobody created it, so it belongs to everyone. With the automated hall this seems not to fit. Somebody built and paid for the installations.

But one has to look more closely at what a learning machine actually is. It learns from data, and these data are nothing other than the recorded experience of those who did the work before. The robot that lays a cable harness tomorrow has copied it from people who did it for years. The data from a hundred plants mentioned in the interview are the accumulated skill of several generations of workers. Whoever uses it did not create it alone. In this sense the knowledge inside the machines is more a natural resource than a product. It was deposited over decades and is now being extracted.

This form, too, has its weakness, and Alaska shows it. Since 2016 governors and legislature have repeatedly cut the dividend to plug budget holes. What belongs to everyone is administered by politics, and politics helps itself when money is short.

VI. Three Forms, One Gap

So we have three forms in which ownership is distributed differently than through the stock market, and each fails at a different point:

Ownership through work holds as long as work is needed. Ownership through purpose holds as long as someone controls the administrators. Ownership through belonging holds as long as politics keeps its hands off the fund.

Physical AI weakens the first form and makes the other two more urgent. We have no ready-made solution. But one can see what a solution would have to achieve. It would have to bind ownership to a purpose that reaches beyond the company, as with Abbe. It would have to subject the administration to a control that is not part of what it controls, which is where German foundations have failed so far. And it would have to define the people entitled to the yield in such a way that no parliament can expropriate them in the next budget crisis.

For small and medium-sized firms there is a model that already works. In 1966 tax advisers who could not afford a mainframe of their own founded DATEV as a cooperative. They did not rent their data processing from a corporation; they built it together and owned it themselves. Today DATEV is one of the largest software houses in Germany, and it still belongs to its members. That is Mondragón’s cooperation among cooperatives, transferred to the platform. What the tax advisers did with data processing, small and medium-sized firms could do with physical AI: own the platform instead of surrendering their data to it.

VII. How This Essay Could Be Refuted

The first objection is the oldest: every wave of automation has in the end created more work than it destroyed. The machine breakers were wrong. If that holds this time as well, the lever of work survives, and the question of the deserted hall is a transitional problem. That is possible. But one should remember that wages in England stagnated for decades during early industrialisation although productivity rose sharply. A transition, too, can last a lifetime.

The second objection: a dividend turns people into recipients. Dignity and belonging come from work, not from a bank transfer. This objection is valid. But it speaks not against the question of who owns the hall, but against an answer that is content with the transfer. Mondragón has shown that ownership can also mean having a say.

The third objection concerns the pace. Humanoid robots in untidy environments are still far from the reliability that series production demands. True. But systems of ownership change more slowly than machines. Whoever starts to think about it only when the hall is deserted comes too late.

The essay would be refuted if it could be shown that in the most highly automated industries and regions the number of employees and their share of the yield do not fall.

VIII. The Deserted Hall

The question of who owns work occupied the nineteenth and twentieth centuries. The workers asked it because they were needed. The question of who owns the machine that takes over the work must be asked before nobody is needed any more who could ask it.

Who owns a hall in which nobody works any more: those who hold its shares, or those whose skill is inside its machines?