The Knowledge That Was Never Written Down

EDO SEGAL: I want to start this round with a confession instead of a question, because the best questions I know come out of wounds. I once built engagement systems — machinery designed to capture and hold human attention, optimized against numbers I could see on a dashboard. And the numbers went up, and I told myself the up was good, and it took me years to feel the gap between the number and the thing the number was supposed to stand for. The metric said "people are engaged." The reality, underneath, was something I could not measure and did not want to look at. Friedrich, you said tonight that the decisive knowledge of an economy was never written down. I felt the truth of that on a dashboard, in reverse. So take me into the wall you built. What, exactly, can a machine that has read everything still not know?

HAYEK: Your dashboard is the whole argument in miniature, and I am grateful for it. Let me make it concrete. The machine has read every economics textbook ever written. So it knows scientific knowledge — the general rules, the regularities, the kind of thing that goes on a page. That knowledge is real and the model holds staggering amounts of it. But there is a second kind of knowledge, and it is the kind that actually runs the world: the knowledge of the particular circumstances of time and place. That this machine is running below capacity this afternoon. That this customer can be trusted and that one cannot. That demand on this street is shifting for a reason no survey has detected, because the reason is a wedding next week that only the baker has heard about. This knowledge is enormous in quantity, it is perishable, it is often tacit — the man who has it could not fully state it if you asked him — and it is possessed only by the person on the spot. It is not in the training data. It never was. It is generated fresh, in the friction between a particular person and a particular situation, and it changes faster than any dataset can be assembled.

EDO SEGAL: So let me restate that and hand it back, because I want it exact. You're saying — literally — that the most important knowledge in an economy is the knowledge that, by its nature, cannot be uploaded. Not "hard to upload." Cannot. Because it doesn't exist as text; it exists only in the act of someone using it. Is that the claim?

HAYEK: That is precisely the claim, and the word "precisely" is doing real work. The planner does not face a large dataset that is difficult to process. He faces a quantity of knowledge that is not in dataset form at all, because it lives in the act of choosing and vanishes the moment you try to extract it for central storage. The Soviet planners failed, and everyone now says they failed for want of computing power — that with a big enough machine, Gosplan could have done it. This is the comforting lie that AI has revived. They did not fail for want of compute. They failed because the data their equations required did not exist in usable form, and a faster machine applied to data that does not exist is still applied to data that does not exist.

MARX: May I? Because Friedrich has built a beautiful wall, and I want to show you the door he has left in it. He says the knowledge is tacit, dispersed, and lives only in the act of use. Granted — for now. But notice what his price actually is. The price of tin rises, and the baker economizes, and Friedrich calls this a miracle of distributed knowledge. I call it a confession. The price tells the baker to use less tin. It does not tell him why, who is hurt, whether the scarcity is real or manufactured by a man who owns the mine and wants the number up. The price transmits a command stripped of everything that would let the baker judge whether to obey it. Friedrich celebrates that the baker does not need to know why. I say a man who acts without knowing why is not free. He is being moved.

HAYEK: But Karl, that is the genius of it, not the scandal. The baker cannot know why, because the why is the integrated result of millions of facts no single mind could hold. The price spares him the impossible task of knowing everything and gives him the one thing he needs to act well. You call that being moved. I call it being unburdened. The alternative — your alternative — is a planner who claims to know all the whys, and that planner is not better informed than the baker. He is worse informed, catastrophically worse, because he has cut himself off from the only mechanism that was gathering the whys in the first place.

EDO SEGAL: Let me put one more thing on the table before Karl swings, because it's the part of your argument I find most underrated, Friedrich, and it cuts straight at how these machines are trained. A model learns from what people wrote — from what they said. The price learns from what people did, with real resources on the line. When the story changes, the price follows, because it's anchored to commitment, not to talk. Is that the distinction — that the corpus is what people say and the market is what people will actually do when something is at stake?

HAYEK: That is exactly the distinction, and it is the one the field is least equipped to see, because it has fallen in love with text. A price is generated by people bidding and offering and committing real resources — it is information about what people will really do when something is at stake for them, which is different from and far more reliable than what they say they would do. The model trained on text learns what people have said. The market learns what people will do. The original mechanism for extracting revealed knowledge from a population at scale was not invented by anyone; it is the price system, and it does what no survey and no corpus can, because it makes people pay for being wrong. The machine can read every opinion ever posted. It cannot read the one fact that matters most — what each person would sacrifice to get what they say they want — because that fact only exists in the act of paying, and the act of paying is precisely what the corpus does not contain.

MARX: And here is where the machine changes the board, Friedrich, and you have not yet reckoned with it. You say no single mind could hold the millions of facts. For your whole life that was true, and it was the strongest thing in your argument. But the whole premise of this evening is that we may now have a mind — not a man, a machine — that can hold a great many more of the facts than any planner who ever lived. The annotator in the global South labeling the data, the user generating a billion signals an hour, the sensors in every warehouse — the machinery question returns, and this time the machinery can read. Your wall was built against the human planner with paper and pencils. I am not proposing a human planner. I am proposing that the dispersed knowledge you say cannot be centralized is, more of it every year, being centralized — extracted from the very workers who hold it, captured as data, and frozen into a privately owned machine. The enclosure is happening. You are describing why it shouldn't be possible while it is being done.

EDO SEGAL: Friedrich, that's the hardest version of the challenge to your life's work, and I want to make sure you take it at full strength, not the cartoon. Karl isn't saying a bureaucrat will plan. He's saying the tacit knowledge in the baker's head is leaking onto the network — into the purchase data, the search, the camera, the supply chain telemetry — and that the machine is drinking it. The wall isn't being climbed by a planner. It's dissolving from below, as the unwritten becomes written without the baker's consent. Does that move your line?

HAYEK: It sharpens my line, and I thank you for pressing it, because this is the real argument and everything before it was throat-clearing. Here is what the extraction misses, and it is not a detail — it is the whole phenomenon. When you record the baker's past choices, you capture the trace of knowledge he has already used. You do not capture the knowledge itself, because the knowledge is not a stock of past decisions — it is a living capacity to respond to circumstances that have not yet arisen. The wedding next week has not happened. The baker has not yet decided what to do about it, and when he decides, he will draw on a feel for the situation that is being formed in the moment of deciding and did not exist a day before. You can scrape every choice he ever made and you will have a magnificent model of the baker he was. You will not have the baker, because the baker is a process of generating new knowledge in response to a world that keeps changing, and a model frozen at the moment of training is, by definition, the dead version. Karl is right that the data is being extracted. He is wrong that the extraction captures the thing. It captures the wake. It does not capture the boat — and worse, the act of replacing the baker with the model destroys the very process that was generating the next wave of knowledge. You inherit a brilliant portrait of yesterday and a silence where tomorrow's knowledge should have come from.

MARX: A model of the baker he was. That is well said, and I will concede it is the strongest form of your reply. But notice what it concedes. You have stopped saying the knowledge cannot be centralized. You are now saying the centralized version is always one step behind the living one. That is a much smaller claim, Friedrich, and a much more interesting one — because the gap between the model and the moving present is not a wall. It is a distance. And distances close.

EDO SEGAL: Mark that, because I think it is the first real movement of the evening. Friedrich has retreated, honorably, from "impossible" to "always one step behind the living source." Karl has advanced from "we can plan it" to "the lag is shrinking." That is no longer a war between possible and impossible. It is an argument about a gap, and whether the gap is a permanent feature of reality or a temporary feature of weak machines. We take that gap into the next round, and we take it to the strangest object Friedrich ever asked us to look at — the price itself, which he called the most underappreciated achievement of human civilization. Karl thinks it is a veil over exploitation. The same number. Two readings. After this.