Knowledge Problem
The knowledge problem is F. A. Hayek’s argument that the hard part of economic organization is using knowledge that never exists in one place. In “The Use of Knowledge in Society” (American Economic Review, September 1945) he states the constraint directly: the knowledge of the circumstances “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.” The economic problem of society is accordingly “a problem of the utilization of knowledge not given to anyone in its totality” — of securing the best use of resources known to individual members of society, for ends whose relative importance only those individuals know.1
Knowledge of particular circumstances
Two kinds of knowledge matter in this argument. Scientific knowledge — the kind that can be written down, aggregated, and taught — is the kind planners and economists habitually count. The other kind is knowledge of the particular circumstances of time and place: the idle machine, the half-empty ship, the local demand shift that only the person on the spot sees in time to act on. It is ephemeral, frequently inarticulate, and mostly useless to anyone not situated to exploit it. Hayek’s claim is that this second kind dominates the practical problem of organization, and that no central body can collect it, because it cannot be conveyed to a central authority in statistical form and it expires before it travels.
The price system as a communication channel
Given dispersed knowledge, Hayek asks what mechanism lets people coordinate as if a central mind existed. His answer is the price system read as a communications network. His standard example is tin: when new demand for tin arises somewhere, the price rise propagates just enough information for tin users everywhere to economize, and none of them need learn why the price moved. The system’s notable property is its economy of knowledge — “only the most essential information is passed on, and passed on only to those concerned” — and each participant needs to know very little to take the right action. A single price carries a compressed summary of conditions no participant could describe. Rigid prices weaken the channel: Hayek notes the function is fulfilled less perfectly as prices grow less responsive.2
The given-data assumption
The paper’s target is the formal setup in which all relevant information is treated as given, after which the remaining problem “is purely one of logic.” Hayek argues that assuming the knowledge to be given to a single mind “is to assume the problem away,” since the practical problem arises exactly because the facts are never so given. Statistical equilibrium analysis keeps a useful function as a preliminary; it does not deal with the social process that actually solves the problem. He presses the same point against Schumpeter, whose treatment of the data as objectively given he attributes to “the ambiguity of the term ‘datum’.”3
Modern threads
Cowen’s two-factor model of AI-era growth gives the knowledge problem a modern name: Polanyi knowledge, the tacit, inarticulable knowledge of time and place, custom and habit, that Hayek argued could never be centralized. In that model intelligence is the abundant factor and the tacit complement is the bottleneck, so returns flow to holders of local knowledge. See ai-economic-growth-complementarity.
The knowledge problem also appears in multi-agent AI systems. Krishnan reports that a market of models bidding for tasks beat an orchestrating hub on cost and tied a strong solo model on quality, and explains the result in Hayek’s terms: “markets work because we have local information that a price signal can elicit.” Even badly calibrated bidders help when the value of an independent retry is high, and MarketBench measures the friction the price channel must tolerate — agents badly miscalibrated on their own success probability and token use, some overconfident, some underconfident. See theory-of-the-firm.4
Prediction markets are the wiki’s other designed instrument for this aggregation problem, with calibration measured at scale across four platforms — see prediction-markets.
Rent control supplies a concrete case of a legally rigid price channel. A capped rent stops transmitting local scarcity to prospective tenants and owners, so allocation shifts toward queues, tenure, connections, and other non-price signals.
Interstate-energy-permitting frames the complementary boundary problem: state review preserves knowledge of local costs, while federal review can register interstate benefits that a transit state’s process excludes. Moving authority changes whose knowledge can affect the decision; it does not make the underlying conflict disappear.
The wiki’s older threads on coordination without a coordinating mind run through invisible-hand (Smith’s own limited uses of the phrase) and the emergent-order case of gongkai. Scitovsky’s asymmetric-markets approach the same knowledge unevenness from the opposite flank: not dispersal across individuals but inequality across the two sides of a market, where the better-informed side converts its knowledge into monopoly power and competition harnesses that power through nonprice services.
Open questions
- Whether model-internal state (memories, scaffolds, prompt drift) creates genuinely private knowledge among AI agents over time, as Krishnan speculates — the premise on which a market advantage would grow with continual learning.
- Whether prediction-market machinery transfers to agent markets given measured self-miscalibration, or whether the elicitation instruments need redesign first.
- How much of the tacit-knowledge bottleneck is permanent and how much is slowly automatable as deployed systems accumulate firm-specific context.