Knowledge and Decisions

In the AI Story

Sowell built Knowledge and Decisions on Hayek's 1945 essay "The Use of Knowledge in Society," which argued that the knowledge required to coordinate a complex economy exists in fragments scattered across millions of minds, each knowing details about their particular circumstances that no central planner can access. Hayek showed that the price system aggregates this dispersed knowledge into actionable signals without requiring anyone to understand the whole. Sowell extended this insight beyond economics into law, education, family structure, and bureaucracy—demonstrating that the knowledge problem is universal. Every institution faces the choice between centralizing decisions (gaining coordination, losing context) and decentralizing them (preserving context, losing coordination). The question is which matters more for the decision at hand.

The book's application to AI is direct and consequential. Large language models are knowledge centralization engines—they aggregate the expertise of millions of practitioners into systems controlled by a small number of companies. The Sowell-Hayek framework asks: does this aggregation preserve the character of knowledge or destroy it? The optimist says patterns learned from dispersed expertise remain useful when re-deployed. The Hayekian skeptic says situated knowledge loses value when stripped of context—the programmer's knowledge of this system's edge cases, the doctor's knowledge of this patient's particular history, the lawyer's knowledge of this judge's interpretive tendencies. AI can generate outputs matching the form of situated knowledge without possessing its substance—producing confident wrongness that users lacking situated expertise cannot detect.

Knowledge and Decisions provides the framework for understanding why AI productivity multipliers benefit experienced practitioners disproportionately. The senior engineer using Claude Code gets superior results not because the tool works differently for her but because she supplies the contextual knowledge the tool cannot generate—which questions to ask, which edge cases matter, which architectural choices are appropriate for this particular system. The junior developer receives syntactically correct code lacking the situated judgment that determines whether code is merely adequate or genuinely good. This is the knowledge problem at the individual level—the gap between access to tools and capacity to use them wisely, a gap that correlates with the expertise AI is simultaneously democratizing and devaluing.

Origin

Sowell wrote Knowledge and Decisions as a theoretical foundation for arguments he had been making empirically across prior books on race, education, and housing policy. His observation was that policy debates consistently ignored the question of who possessed the knowledge required to make decisions wisely. Minimum wage advocates assumed legislators knew workers' productivity better than employers who hired them. Rent control advocates assumed regulators knew housing markets better than landlords and tenants navigating them. The pattern was universal: centralized decision-makers confidently overriding the dispersed knowledge of people in particular circumstances, producing predictable failures the advocates then blamed on inadequate enforcement rather than flawed design. The book built the theoretical case against this pattern, earning comparison to Hayek's The Road to Serfdom as one of the most rigorous defenses of decentralized decision-making in twentieth-century economics.

Key Ideas

Decisive knowledge is dispersed. The most important information—particular circumstances of time and place—resides with actors in those circumstances; centralization loses context.

Institutions succeed by aligning authority with knowledge. Decision-makers need situated understanding of consequences; distance from circumstances produces predictable failures through knowledge loss.

Prices transmit knowledge without centralizing it. Market prices aggregate dispersed information into actionable signals; participants coordinate without understanding the whole system.

Expertise claims require scrutiny. Specialists possess deep knowledge in narrow domains but lack the cross-domain and contextual knowledge their authority often requires; deference is dangerous.

Unintended consequences follow knowledge gaps. Policies designed without situated knowledge reliably produce outcomes designers did not anticipate; the surprise is structural, not accidental.