Hayek vs Mises: The Calculation Problem Distinction

Two different critiques of central planning that most people conflate. AI addresses Hayek’s problem but not Mises’s.

Source: Bob Murphy — Can AI Solve the Socialist Calculation Problem? Links: Market Efficiency, Value and Profit, Business Cycles, Equation of Exchange

The Four Problems With Central Planning

Problem	Description	Can AI help?
Incentive	Without differential compensation, who does the hard jobs?	No
Corruption/power	Central planning concentrates power → oppression	No
Hayek’s knowledge	Dispersed, tacit, local knowledge can’t be centralized	Partially yes
Mises’s calculation	Without property + exchange, no prices → no common denominator for comparing heterogeneous inputs	No

Most people (including prominent economists like Acemoglu) treat #3 and #4 as the same thing. They’re not.

Hayek’s Knowledge Problem

Claim: Knowledge is dispersed across millions of individuals — local conditions, momentary preferences, tacit know-how. No central planner can aggregate it.

AI’s challenge to this: Recommendation engines, demand forecasting, pattern recognition. Amazon’s algorithms already aggregate dispersed preference data at massive scale. Murphy honestly concedes: “AI could help mitigate the problem and make your forecasts more accurate.”

Status: Genuinely weakened by AI. Not dead — tacit knowledge still resists formalization in some domains — but the gap is narrowing.

Mises’s Calculation Problem

Claim: Even with perfect knowledge of all preferences, all technology, all resources — and even with perfectly benevolent, motivated planners — you STILL can’t do rational economic calculation without market prices. Because prices aren’t data. They’re emergent outputs of a specific institutional process (private property → exchange → bidding → price).

Murphy’s oven analogy: A thermometer measures temperature — an objective physical property that exists independently of measurement. But “what is this oven worth?” isn’t measuring a property embedded in the oven. It incorporates everyone’s subjective preferences, expectations, and alternatives. The value only exists as an output of a market process. More data doesn’t help because the value isn’t IN the oven.

Salerno’s formulation: “Appraisement is neither knowledge nor arithmetic but something new under the sun, introduced into the world only when the institutional prerequisites of a market economy are fulfilled.”

Status: Untouched by AI. The problem isn’t computational power — it’s that the thing you need (prices) only exists within the thing you’re trying to replace (markets).

Why “Just Simulate a Market” Doesn’t Work

The obvious response: have the AI simulate an entire market economy — agents, property, exchange — and read off the emergent prices.

Counter-arguments:

The model is always a model. It will have approximations, gaps, and edges. In a competitive system, any gap is an arbitrage opportunity someone will exploit. The model breaks precisely where it matters most. (Same principle as computational irreducibility — you can’t shortcut the economy, you have to run it.)
Opportunity cost: The computing resources used to simulate the market could, in an actual market, be used by private actors to enhance production directly. “Socialism + AI” should be compared to “capitalism + AI,” not “capitalism circa 2000.”
If you’re simulating a market… just have a market. The simulation requires all the institutional features (property, exchange, competition) that define a market economy. At that point you’re not replacing the market — you’re running one inside a computer and calling it planning.

Connection to the Vault’s Framework

This is the same map/territory distinction that runs through everything:

Gödel: The formal system (map) can’t capture all truths about the territory
Numbers: Agent-imposed abstractions that approximate reality but aren’t reality
Deontology/consequentialism: You can’t calculate consequences across a computationally irreducible system
Prices: Emergent outputs of a process, not data points to be collected. The map (simulation) is always less than the territory (actual market)

The decimal precision question applies: how accurate does the simulation need to be? Answer: any gap is exploitable. And in economics, exploitation of model gaps isn’t a minor error — it’s where fortunes are made and crises are born.

Usage Note

Too technical for the YouTube economics series as a main topic. But essential reference material for engaging with economists who understand the nuance. The Hayek/Mises distinction is the thing that separates “AI solves planning” (plausible against Hayek) from “AI solves planning” (impossible against Mises). Most AI-optimist arguments only address Hayek and don’t know Mises’s argument exists.