The Calculation Debate, Reopened

EDO SEGAL: Friedrich, a hundred years ago you and your teacher Ludwig von Mises argued, against Oskar Lange and the market socialists, that a planned economy could never calculate its way to rational allocation — that without real prices generated by real exchange, the planner is blind. Lange thought it was a math problem that better equations would solve. You thought it was a knowledge problem that no equations could solve. For most of a century your side was taken to have won, and then the machines arrived, and a generation of people looked at them and said: now we can finally do what Lange wanted. So I have to ask you the blunt version. Did AI just reopen the calculation debate — and did it reopen it on Lange's side?

HAYEK: It reopened it. I will not pretend it did not, and the people who say my argument is simply vindicated by history have not understood how serious this challenge is. Lange's reply to me, all those years ago, was that the planner could simulate the market — could adjust prices by trial and error until supply matched demand, solving the equations the market solves, only consciously. My answer was that the equations require data the planner cannot have, because the data does not exist independently of the market process that generates it. Now along comes a machine that can, in principle, run Lange's trial and error at a speed and scale he could never have imagined, ingesting signals from everywhere at once. The honest position is that this is the most serious test my argument has ever faced. The dishonest position is to wave the machine away. I do not wave it away. I say: run the test carefully, and watch where it breaks.

MARX: And I say it has already half-broken in my favor, and you are too good a thinker not to feel it. Look at what these systems actually do. A logistics network the size of a continent, routed by a single model, allocating millions of decisions a second — that is planning, Friedrich, conscious central allocation, and it works. The warehouse does not use internal prices. It uses an optimizer. A firm the size of a nation coordinates its entire internal economy not through markets but through plan — and it has done so for a century, and the AI only deepens it. You told us planning was impossible. The largest and most successful enterprises on earth are planned internally, and they grow more planned every year as the machine gets better. The island of plan inside the sea of market is expanding. Why should the shore hold at the edge of the firm?

HAYEK: Because the firm exists inside prices, Karl, and that is the whole answer to the puzzle that has fooled the planners for a century. Yes, a firm plans internally — I wrote about exactly why firms exist, why some coordination is cheaper inside a hierarchy than across a market. But the firm knows whether its plan is working because it is surrounded by prices it did not set: the wages it must pay, the inputs it must buy, the price its product fetches against rivals, the profit or loss that tells it, ruthlessly and from outside, whether its internal plan was wise. The optimizer in the warehouse is optimizing toward targets that are ultimately set by a market it sits inside. Remove the surrounding market — plan the whole thing — and the optimizer loses its compass. It can still compute. It can no longer tell whether the answer is good, because the only test of a good allocation is whether people freely chose it at a price, and you have abolished the choosing and the price. The internal plan is a child held up by the market's hands. Karl wants to plan the floor itself, and then there are no hands.

EDO SEGAL: Let me make this vivid, because there is a real case in the history, and it haunts this exact argument. In Chile, in the early 1970s, Stafford Beer built Project Cybersyn — an actual attempt to run a national economy through a network and a control room, telexes feeding a central model, in real time, decades before the compute existed to do it properly. It was the calculation debate made flesh, and it ended with a coup. Karl, was Cybersyn premature truth — the right idea waiting for the machine? Or, Friedrich, was it the fatal conceit caught in the act, lucky to fail before it could do real harm?

MARX: It was premature truth, and the tragedy is that it was strangled in its cradle by exactly the powers that profit from calling it impossible. Beer understood something neither the old planners nor my friend will grant: that you do not plan by commanding every detail from the center, but by building a system of signals and feedback — a nervous system for the economy — that lets the whole adapt. He was reaching, with telexes, for what your data centers now have in abundance. And he was reaching for it in the name of the people who actually run the factories, with the workers in the loop, not over them. That it failed tells you nothing about whether it could work, any more than the first airplane crashing tells you flight is impossible. It tells you who holds the power to strangle the alternative before it can prove itself.

HAYEK: And I say Cybersyn is the clearest warning we have, precisely because its designers were brilliant and sincere. Beer was no fool and no tyrant. He genuinely believed feedback could substitute for prices, that the control room could see what the market sees. And the very ambition is the disease — the conviction that a room full of cybernetic genius could hold what a whole society of free choosers generates. It was the pretence of knowledge in its most seductive form, technocracy with a human face, and it would have failed even without the coup, slowly, as the model drifted from the moving reality it could not keep pace with, while real shortages appeared in the gap. Your machines do not abolish that gap, Karl. They make it harder to see, because the dashboard is so much more convincing now. The control room that lies to you confidently is more dangerous than the one that admits it is blind.

EDO SEGAL: Let me offer you a version of Karl's machine that isn't a control room at all, Friedrich, because I think it's the one that actually tests you. Imagine not one central model but millions of AI agents, each holding one person's or one firm's local knowledge, each acting on it, all coordinating through signals they pass between one another — buying, selling, bidding, adjusting in real time, faster than any human market ever could. No control room. No central planner. Just an invisible hand made of machines. Is that Karl's dream finally realized — or is it just your market, sped up?

HAYEK: It is my market, sped up — and I am delighted you asked, because it is the cleanest thing I can say all night. The moment you accept that the knowledge must stay distributed, held by many agents, and that coordination must happen through compressed signals passing between autonomous actors rather than being assembled in one place — you have accepted my entire architecture and abandoned Karl's dream, whatever you call the agents. A genuinely distributed system of agents trading on local knowledge is not an alternative to the market. It is a market, or something so like one that the distinction is a word game. AI agents transacting with one another might well form markets of a new and faster kind, and I would welcome it. But that is a Hayekian future, not a post-Hayekian one. The thing my argument explodes was never the market. It was the planner — the single mind over the whole. Build a million minds passing prices and you have built me a cathedral, not a refutation.

MARX: And I will take the cathedral, Friedrich, because you have just conceded more than you think. You say a million agents passing signals is a market. Fine. But who owns the million agents? Who wrote their objectives? Who keeps the surplus when they transact? Your "invisible hand of machines" is invisible only about its labor and its ownership, never about its profit. The agents are not free choosers. They are instruments, owned, optimizing for whoever deployed them. You have described a market of slaves and called it liberty because the slaves move fast. The distribution you celebrate is a distribution of execution, not of power. The power — the question of whose ends the whole machine serves — is exactly as concentrated as it was, and your speeding it up has only made the concentration harder to see.

HAYEK: Then we agree the ownership of the agents is the question, which is a remarkable place for the two of us to arrive — and we disagree, as ever, on whether the answer is one owner or none. But I will not let you call it a market of slaves, Karl, because a slave cannot quit and the person behind each agent can switch it off, retrain it, sell it, point it elsewhere. The agent is owned, yes. The human behind it is not. That distinction is the whole of liberty, and you keep collapsing it because your system needs it collapsed.

EDO SEGAL: I want to hold us here one more beat, because something important is happening underneath the argument. Karl, you keep saying the gap between the model and the moving present is shrinking. Friedrich, you keep saying the gap is permanent — a feature of reality, not of weak machines. That is the actual disagreement. Everything else is downstream of it. And neither of you can fully prove it yet, because the test — a machine genuinely fast enough to attempt the whole computation — is only now becoming possible. You are arguing about an experiment that is finally about to run. That is what makes this floor different from every floor below it. Hold the gap. We come back to it from a different door — not whether the machine can plan, but what it does to the human being inside the plan. Because Karl has a word for that, and he wrote it as a young man, with a fury he never quite recovered.