Kenneth Arrow
Kenneth Arrow was among the most influential economists of the twentieth century, with contributions to general equilibrium theory, the economics of information, and social choice. His 1951 doctoral dissertation, published as Social Choice and Individual Values, formalized and generalized Condorcet's paradox into a theorem demonstrating that no method of aggregating individual preference orderings into a collective ranking can simultaneously satisfy non-dictatorship, unanimity, independence of irrelevant alternatives, and unrestricted domain. The theorem won him (with John Hicks) the Nobel Prize in 1972 and has shaped every subsequent effort to design democratic procedures, voting systems, and — more recently — methods for aligning AI systems with diverse human values.
In the AI Story
Arrow's relation to Condorcet is not merely historical. The impossibility theorem is the mathematical culmination of the program Condorcet began in the Essai: the rigorous analysis of what collective decision-making can and cannot achieve. Condorcet discovered the paradox and documented its specific form. Arrow proved its generality: every procedure is subject to analogous failures.
The theorem's implications for AI alignment are direct and largely unacknowledged in the technical literature. Alignment requires aggregating diverse human preferences into a coherent specification of system behavior. The theorem proves no such aggregation satisfies all reasonable conditions simultaneously. This is not a technical obstacle to be overcome by better engineering — it is a mathematical feature of preference aggregation that every alignment effort must confront.
Arrow himself, late in his career, engaged with questions of information, uncertainty, and the economic implications of AI-like technologies. His work on the economics of information (published 1962) anticipated many of the structural problems now central to AI governance: asymmetric information between producers and users of knowledge, the public-goods character of information, and the difficulty of pricing goods whose value depends on their distribution.
Origin
Born 1921 in New York City, educated at CCNY and Columbia, where Harold Hotelling supervised his dissertation. Served on the faculty at Stanford from 1949 until his retirement, with visiting appointments at Harvard and the Cowles Foundation.
The impossibility theorem was developed during his time at the Cowles Foundation and published in 1951. His subsequent work on general equilibrium (with Gérard Debreu, 1954) and on the economics of information sustained his influence across multiple subfields of economics.
Key Ideas
Generalized Condorcet's paradox. Every aggregation procedure fails at least one reasonable condition.
Nobel Prize, 1972. Shared with John Hicks for work in general equilibrium and welfare economics.
Foundational for AI alignment. The theorem is the structural constraint every alignment effort confronts.
Economics of information. Anticipated structural problems now central to AI governance.
Appears in the Orange Pill Cycle
Further reading
- Kenneth Arrow, Social Choice and Individual Values (1951; 2nd ed. 1963)
- Kenneth Arrow, Essays in the Theory of Risk-Bearing (1971)
- Amartya Sen, Collective Choice and Social Welfare (1970; expanded 2017)