In May 2023 the MIT economist Daron Acemoglu, who went on to share the 2024 Nobel prize in economics, asked a question on Twitter. Friedrich Hayek argued in 1945 that central planning fails because no planner can collect and compute all the information an economy runs on. “What if computational power of central planners improved tremendously?” Acemoglu wrote. “Would Hayek then be happy with central planning?”
About thirty economists sent him the same reply, according to Shruti Rajagopalan, who hosts the Mercatus Center’s Ideas of India podcast: Hayek never made a computational argument. He made an argument about the nature of knowledge itself. Acemoglu reached the same conclusion in print that June. In an essay titled “Hayek vs. AI Socialism”, he quoted Hayek’s warning that the relevant knowledge “by its nature cannot enter into statistics” and concluded that “not even an all-powerful large language model” could deal with it. AI socialism, he wrote, “offers only a superficial critique of Hayek.” That distinction is the whole story, and it explains why a model trained on most of what has ever been written still cannot do what a well-trained human mind does.
Hayek’s argument
Hayek’s essay “The Use of Knowledge in Society” appeared in the American Economic Review in September 1945. Its central sentence is worth quoting in full:
“The peculiar character of the problem of a rational economic order is determined precisely by the fact that the knowledge of the circumstances of which we must make use never exists in concentrated or integrated form but solely as the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess.”
He then separates scientific knowledge, the knowledge of general rules that a panel of experts could hold, from “the knowledge of the particular circumstances of time and place.” That second kind, he says, gives “practically every individual” an advantage over everyone else, because each person knows things about their own situation that nobody else can. The shipper who knows a tramp steamer is sitting half empty. The machinist who knows which lathe has been running badly this week. Hayek’s phrase for that person is “the man on the spot.”
His point about prices follows from that. He says “it is more than a metaphor” to describe the price system as “a system of telecommunications” that lets people act on knowledge they do not hold. A price carries the results of millions of small judgments without anyone having to write a single judgment down.
That is the part Acemoglu’s question misses. Hayek did not say the knowledge was hard to gather. He said it never exists in gathered form at all. Peter Boettke of George Mason University, who directs the Hayek Program at Mercatus, said it more bluntly on the Ideas of India podcast in July 2023. Outside the context that produces it, he said, the knowledge “doesn’t exist. It’s not even there.”
Polanyi: we know more than we can tell
Hayek described where the knowledge lives. Michael Polanyi, a Hungarian physical chemist who turned to philosophy in mid-career, described what it is like from the inside.
The first chapter of his 1966 book The Tacit Dimension states its whole thesis in six words: “we can know more than we can tell.” His examples are ordinary. You can pick out a friend’s face in a crowd, yet you cannot say what it was you recognized. You can ride a bicycle, yet you cannot state the rule that keeps you upright. Polanyi actually worked out that rule in Personal Knowledge in 1958 (adjust the curve of your path in proportion to your unbalance divided by the square of your speed) and then observed that no cyclist has ever learned from it. A radiologist reads an X-ray that a first-year student sees as fog. A wine taster names the vintage and cannot tell you how.
Polanyi drew a hard conclusion about how such knowledge moves from one person to another: “An art which cannot be specified in detail cannot be transmitted by prescription, since no prescription for it exists. It can be passed on only by example from master to apprentice.”
A model learns from prescriptions. It learns from text, from whatever somebody managed to write down. Tacit knowledge is, by Polanyi’s definition, the part that never got written. A model can only learn from the shelf. The apprenticeship is the part that is missing.
Six properties of the knowledge that matters
Economists who study socialism have had to be precise about this, because the whole case against central planning rests on it. The debate began in 1920, when Ludwig von Mises argued that without private property there are no market prices, and without prices there is no way to tell a wasteful production plan from a good one. János Kornai, who lived and worked under Hungary’s planned economy for decades, confirmed it from the inside in The Socialist System (1992): the volume of information that bureaucratic coordination needs is more than any center can process. Jesús Huerta de Soto’s 1992 book Socialism, Economic Calculation and Entrepreneurship, the standard treatment in this tradition, opens its second chapter with a list of six properties of the knowledge an economy runs on.
It is practical and subjective. It is learned by doing, and it is a different thing from scientific knowledge of general rules. It is private: each person holds it exclusively, and nobody else can get at it without that person’s cooperation. It is dispersed: billions of people hold pieces of it, and no committee could question them all. It is mostly tacit and therefore cannot be articulated, which is Polanyi’s point restated, and Huerta de Soto credits Polanyi for it by name. It is created out of nothing, by the act of a person noticing an opportunity that did not exist until someone saw it. And it is transmissible, mostly without anyone being aware of it, through the social process that prices make possible.
Hayek supplies the property that ties all six to a moment. The knowledge is of “particular circumstances of time and place,” so a factory in Madrid cannot simply copy how one in Beijing runs. And because circumstances change, the problem is never solved once. In his 1940 reply to the market socialists, as Boettke summarized it on the podcast, Hayek argued that the least-cost way of producing anything has to be discovered anew each day, and that whoever fails to discover it is competed away by someone who did.
The creation-out-of-nothing point was Israel Kirzner’s great theme too, and I wrote about his idea of alertness last month.
An old idea with many names
Philosophers and economists of very different politics have described the same line, each with a different name for it.
Aristotle, in Book VI of the Nicomachean Ethics, separates episteme, knowledge of what is universal and can be taught, from phronesis, practical wisdom, which deals with particulars and is learned only through experience. Gilbert Ryle’s The Concept of Mind (1949) drew the line between “knowing that” and “knowing how,” and argued that no quantity of the first adds up to the second. Michael Oakeshott’s essay “Rationalism in Politics” (1947) named them technical knowledge, which can be written in rules, and practical knowledge, which “exists only in use.” James C. Scott’s Seeing Like a State (1998) borrowed the Greek word metis, the practical cunning Homer gave Odysseus, for the local knowledge that every grand plan of the twentieth century tried to override and could not replace. Thomas Sowell built his 1980 book Knowledge and Decisions on the same premise.
Adam Smith got there first, in the opening chapter of The Wealth of Nations (1776). The number of people whose work goes into a common woollen coat, he wrote, “exceeds all computation.” Leonard Read’s 1958 essay “I, Pencil” made the same point with a pencil. Not one person on earth knows how to make one.
I made the same distinction last December, in a post on leading in the age of intelligent agents, and put it in two sentences: “Knowledge can now be automated. Judgment cannot.” That post builds on Dan Shipper’s idea of an allocation economy, in which the valuable skill is directing intelligence, human and artificial, toward the right ends. In Hayek’s and Polanyi’s terms, the knowledge that can be automated is the written-down kind. What I called judgment there is the knowledge of time and place that never got written down.
Why the machine stays on the wrong side of the line
The strongest current version of the argument belongs to Boettke and Rosolino Candela, in a 2023 paper in the Journal of Economic Behavior and Organization called “On the Feasibility of Technosocialism”. Their claim is that today’s enthusiasm for AI-run economies is the market-socialist case of the 1930s with a bigger computer, and that it repeats the same mistake. It treats the economic problem as computing data that already exists, when the problem is discovering knowledge that does not exist yet.
On the podcast, Boettke offered a way to see the difference. Some environments are “kind”: the rules are fixed and the board is finite, so a computer will beat Garry Kasparov at chess. Others are “wicked”: the situation keeps changing, and nobody has built a robot that plays football like Cristiano Ronaldo, because the spin on every ball is new. An entrepreneur works in the second kind of environment. So does a manager. So does a nurse or a teacher. Boettke also recalled that his own graduate adviser, Don Lavoie, a professional programmer, assigned Hubert Dreyfus’s 1972 book What Computers Can’t Do before any economics, alongside John Searle’s 1980 “Chinese room” argument. Searle’s point was that shuffling symbols correctly is syntax, and syntax is not understanding.
Jesús Fernández-Villaverde of the University of Pennsylvania reached a similar conclusion for public policy in a 2020 paper. Even where a problem is well defined, the data an algorithm would need is mostly not there to be had.
Anyone who uses these tools has already seen the problem. You ask a model for a draft, and what comes back is fluent and half wrong, because it did not “know” the ten things about your situation that you never typed. Which client is touchy about price. What your boss means by “soon.” The model’s pattern recognition is real, and it is running on nearly everything ever written. What it cannot run on is the context in your head, because that context was never written anywhere. Every time a prompt needs a second and third round of “no, what I meant was,” you are watching Hayek’s problem in miniature. The missing knowledge was of “time and place,” and the man on the spot was you.
Boettke is careful about this. He says he is “not dogmatic” about it, and he uses AI tools every day. The claim is narrower and stronger than “AI is overrated.” A model is built from what people wrote down. The knowledge Polanyi described is the part they could not.
What this means for your education
If Polanyi is right, this kind of knowledge is acquired one way: by practice, alongside people who already have it. You do not get judgment from a summary. You get it by making a judgment in front of others and hearing it challenged. Then you revise it.
That is what Reliance College has done in its seminars for decades, and it is what AI Proof U™ is built to do in ten weeks. The program uses Collaborative Socratic Practice on great texts to train judgment, independent thinking, and clear communication, the capacities that stay decisive as AI takes over everything that can be written down. The ten weeks open with an in-person intensive in Chicago and close with an in-person capstone where you present your own work. Between them are six guided online seminars.
Apply for the October 2026 cohort
Applications are open now for AI Proof U™. The cohort starts October 2, 2026, with the three-day intensive in Chicago on October 2 to 4, and finishes ten weeks later with a professional certificate. Tuition is $2,500, and monthly payment plans are available. The program is built for students and young professionals, and students rate it 4.9 out of 5.
Apply today at reliancecollege.org/ai.
If you want to see where you stand first, the same page has a four-question quiz, about two minutes, that gives you a free custom report on how AI-proof your skills are. And if you have questions about the schedule or the format, the brochure on that page answers them.
