AI often acts as cheap, scalable prediction that can boost output where firms can rewire work and data flows, yet organizational frictions, data dynamics and implementation costs frequently blunt its economic impact; crucially, social‑science research so far provides limited, rigorous guidance for policymakers preparing for very rapid or systemic AI‑driven transitions.

Main Finding

Social-science treatments of AI over the past dozen years yield valuable, practical insights—AI functions largely as cheap prediction; adoption is constrained by organizational architecture and implementation costs; data behaves as an economic asset; and redistributional effects depend on firm- and task-level heterogeneity. However, these books and the essay that reviews them stop short of providing rigorous guidance for policymakers facing fast, transformative scenarios emphasized by many AI researchers (e.g., rapid labor churn, accelerated scientific progress, abrupt labor–capital shifts, or low‑probability high‑impact/existential risks). (JEL C45, C80, D83, O31, O36)

Key Points

• AI ≈ cheap prediction: A useful analytic lens is viewing many AI capabilities as dramatically cheaper or higher‑quality prediction, which reshapes task allocation and complements certain inputs while substituting others.
• Architectural and organizational barriers: Technical capability alone does not guarantee economic impact; adoption depends on firms’ internal architecture, processes, and ability to integrate AI into production and decision workflows.
• Data as an asset: Data has productive value, network effects, and potential market failures (spillovers, concentration), which affect firm incentives and distributional outcomes.
• Implementation frictions matter: Measurement, management, retraining, legal/regulatory constraints, and coordination problems blunt and shape AI’s realized effects on productivity and employment.
• Distributional and equilibrium analyses are well developed for incremental/transitional scenarios: Economists have made progress mapping winners/losers, task‑level reallocation, and short‑to‑medium‑run labor market effects under gradual adoption.
• Gap on transformative scenarios: Existing social‑science treatments provide limited tools for policymakers worried about very rapid, economy‑wide change or extreme outcomes (e.g., abrupt unemployment shocks, step‑change in R&D productivity, or existential risks from advanced AI).

Data & Methods

• Evidence base: The essay synthesizes seven books by social scientists, drawing on a mix of empirical and conceptual work across firm-level studies, labor-market analyses, and theoretical frameworks.
• Common methods in the reviewed literature:
  • Case studies and qualitative organizational analysis to explain adoption and implementation barriers.
  • Reduced‑form empirical work (e.g., difference‑in‑differences, instrumental variables) using firm, establishment, and worker datasets to estimate effects on productivity, tasks, and wages.
  • Structural and conceptual models that embed AI as a technology reducing the cost of prediction or automating tasks within standard production functions.
  • Policy and institutional analysis addressing data governance, market structure, and labor policies.
• Limitations highlighted:
  • Little direct modeling or empirical work on rapid, non‑incremental transitions or on low‑probability high‑impact scenarios.
  • Sparse integration between social‑science empirical methods and forward‑looking scenario analysis used by some AI researchers.
• Recommended methods to fill gaps:
  • Use of matched employer–employee and administrative datasets, firm experiments, and natural experiments to identify causal effects of fast adoptions.
  • Structural models and macroeconomic simulations that allow for endogenous technological change, non‑linear adoption dynamics, and sectoral spillovers.
  • Interdisciplinary collaboration to translate technical progress assumptions from AI research into economic scenarios amenable to quantitative analysis.

Implications for AI Economics

• Research priorities for economists:
  • Build models that capture prediction as an explicit input and allow for fast, non‑linear adoption, endogenous R&D acceleration, and capital–labor substitution dynamics.
  • Quantify data externalities and concentration risks to inform antitrust and data‑governance policy.
  • Develop credible scenario analyses and stress tests for rapid labor churn and macro instability, using structural models and stochastic simulations.
  • Study policy levers (social insurance, retraining, wage subsidies, spatial/sectoral mobility supports) under plausible fast‑change scenarios and unequal bargaining power.
  • Integrate risk assessment for catastrophic or existential outcomes into economic policy design—e.g., incentive‑compatible governance of powerful AI systems, insurance markets for systemic risks, and mechanisms for global coordination.
• Empirical and institutional needs:
  • Better real‑time indicators of AI capability diffusion (productivity signals, hiring/skill postings, API consumption, firm‑level AI investment).
  • Expanded access to high‑granularity administrative and firm data to observe rapid reallocations and adoption patterns.
  • Pilot interventions (RCTs) at the firm and regional level to test retraining, matching, and institutional responses to rapid automation shocks.
• Policy guidance:
  • Short/medium term: focus on lowering adoption frictions for socially beneficial uses, strengthening data governance, and expanding active labor‑market policies targeted to likely losers.
  • Preparedness for rapid transitions: invest in scenario modeling, shock‑absorbing institutions (income support, rapid retraining), and regulatory regimes that can respond to concentrated power or systemic risks from very capable AI systems.
• Call for interdisciplinary work: economists should work with AI scientists to translate technical forecasts into economic variables and with political scientists and legal scholars to design feasible governance architectures for extreme but plausible outcomes.