Coordination failures—an N-player game with binary defection incentives and weak enforcement—explain why productivity gains haven’t translated into shorter workweeks and why AI development races on despite existential warnings; a proposed 'Tragedy Index' suggests AI governance is orders of magnitude more difficult than nuclear arms control or climate coordination.

Main Finding

Dasdan (2025) argues that persistent long working hours and the accelerating race to develop transformative AI share an identical game-theoretic structure: they are structural coordination failures (“structural tragedies”) in which individually rational choices systematically produce collectively harmful outcomes. The paper synthesizes five necessary-and-sufficient conditions that generate these tragedies, defines a Tragedy Index that quantifies severity via condition intensities, and shows that AI governance scores far higher on that index than historical coordination problems (e.g., nuclear arms control), making coordination around transformative AI dramatically more difficult. The framework is applied formally to firm productivity competition (proving a coordination failure that explains why productivity gains need not translate into worker welfare) and diagnostically to AI development, with empirical validation via European labor data and the Russia–Ukraine drone escalation.

Key Points

• Unified diagnostic framework: Five structural conditions (necessary & sufficient) that create a “tragedy”:
• N-player structure (many independent actors).
• Binary choice with negative externalities (cooperate/defect format).
• Dominance property (defection is individually dominant).
• Pareto-inefficiency (mutual cooperation Pareto-dominates mutual defection).
• Enforcement difficulty (structural barriers preventing credible enforcement).
• Condition intensities: Each condition varies in real cases; the paper introduces a Tragedy Index that aggregates these intensities to compare coordination difficulty across domains.
• Validation: Framework demonstrated across canonical cases (commons, security dilemmas, bank runs, space debris) and empirically for:
  • Productivity: Formal N-firm model shows firms have a dominant incentive to expand output/use productivity to cut costs or increase output, yielding labor-welfare decoupling despite aggregate productivity gains. European evidence shows decoupling persists even under favorable institutions.
  • AI governance: Mapped to the five conditions and an 8-dimension comparison with arms control; AI scores worse on dimensions like dual-use inseparability, verification, low entry barriers, time scales, and economic–military convergence. The Russia–Ukraine drone war functions as an empirical case where competitive necessity overwhelmed governance dialogue.
• Diagnostic emphasis: The analysis is explicitly diagnostic rather than prescriptive; it catalogs why standard mechanisms (repeated play, reputational effects, voluntary pledges, nonbinding norms) are structurally insufficient and identifies what would need to change in the competitive architecture to succeed.
• Robustness & limitations: The framework is formalized, extended to multiple welfare dimensions, relaxed assumptions are discussed, and limitations/falsifiability criteria are provided.

Data & Methods

• Theoretical synthesis: Integrates established game-theoretic literature (Prisoner’s Dilemma, Dawes on social dilemmas, Jervis on security dilemmas, folk-theorem results) into a compact diagnostic set of five conditions plus intensity scaling.
• Formal models: Presents formal proofs, including an N-firm game modeling productivity competition, showing dominance of defection and Pareto-inefficiency of equilibria; provides extensions, robustness checks (heterogeneous firms, multiple welfare dimensions), and welfare monotonicity results.
• Tragedy Index: Constructs an intensity-based index mapping continuous measures of each condition to a scalar severity metric; offers alternative formalizations using game-theoretic measures.
• Empirical/illustrative evidence:
  • Productivity: Uses macro/historical facts (productivity growth, OECD annual hours), literature on labor share and hours (Huberman & Minns, Pencavel, Alesina et al.), and European case study to show persistence of hours despite productivity gains.
  • AI governance: Draws on surveys and literature on AI risk perceptions (Bengio et al., Grace et al.), investment trends (Federal Reserve Board), and case study evidence from the Russia–Ukraine drone conflict (drone deployment escalation) to illustrate amplified intensity and coordination collapse.
• Methodological stance: Emphasizes diagnosis over policy prescription; explicitly discusses where standard cooperative mechanisms fail due to structural features (imperfect observability, anonymity, high stakes, short time horizons, low entry costs).

Implications for AI Economics

• Coordination risk is a fundamental market failure: Transformative AI development embodies a high-severity social dilemma that private incentives alone will likely fail to resolve. Market competition will tend to favor capability acceleration even when agents privately acknowledge grave collective risks.
• Policy and governance realism: Many familiar tools (voluntary norms, reputational pressure, repeated interactions, nonbinding coordination) are unlikely to suffice because AI scores poorly on critical dimensions (verification, dual-use, low entry barriers, short/fast time horizons). Economists and policymakers should not assume standard arms-control analogies (nuclear) map cleanly onto AI.
• Valuation and investment: Competitive pressures and structural tragedy imply a persistent “race premium” in firm valuations and investment decisions favoring rapid capability development over slow, safety-first paths. This can distort R&D allocation and increase systemic risk premium demands (insurance, risk capital).
• Market design levers likely needed: To change outcomes, interventions must alter the underlying game structure—e.g., raise enforcement feasibility (verifiable standards, auditing infrastructure), change payoff dominance (subsidize safety, tax risky capability pushes), increase coordination bindingness (treaties with verification or credible sanctions), or raise entry costs (regulatory gatekeeping, licensing). The paper implies these are difficult and may require major institutional transformation.
• Labor–AI interaction: The productivity tragedy analysis implies that even if AI raises aggregate productivity, firms’ competitive incentives may channel gains into output/market share rather than worker welfare unless institutional arrangements (labor market rules, bargaining power, redistribution mechanisms) change the firms’ payoff structure.
• Empirical prediction and research agenda: The framework predicts continued capability acceleration absent structural reforms, and that coordination success will correlate negatively with Tragedy Index values. For AI economists, this suggests focusing empirical work on quantifying condition intensities (verification difficulty, entry elasticity, strategic value) and on designing institutional mechanisms that change those intensities.
• Caution for prescriptive claims: Because the paper is diagnostic, implications emphasize that feasible solutions require altering incentives at scale; modest or narrowly targeted interventions are unlikely to overcome the structural obstacles identified.

Reference: Dasdan, A. (2025). The Tragedy of Productivity: A Unified Framework for Diagnosing Coordination Failures in Labor Markets and AI Governance. Preliminary working paper (arXiv:2512.05995v3). (Author used AI assistance in drafting; arguments and conclusions are the author’s own.)