Social‑science research on AI and public policy is tilted toward regulation, not use: roughly three‑quarters of articles treat AI as a problem to be constrained rather than as a tool states and firms deploy, leaving questions of power, redistribution and real‑world governance underexplored.

Main Finding

Social-science research on AI and public policy is epistemically skewed: most studies treat AI as a neutral, problem-solving tool and focus on regulating AI rather than empirically studying how AI is actually used as an instrument of governance. A systematic review of 229 English-language social-science journal articles through 2025 reveals entrenched methodological and ontological biases that limit critical engagement with power asymmetries, techno-solutionism, and distributional consequences in the AI–policy complex.

Key Points

• Three core contributions:
• A critical interrogation of the value and limits of systematic reviews in policy research: systematic syntheses can reproduce epistemic hierarchies and mask normative choices if not reflexively applied.
• A two‑axis heuristic for mapping AI–policy scholarship:
  • Ontology A: AI as a neutral, problem‑solving technology vs AI as a socio‑technical system that shapes power relations.
  • Ontology B: Technology as an instrument of governance (how states/organizations use AI) vs technology as a regulatory target (what regulation should constrain). This framework exposes where scholarship clusters and where critical social science has more or less space to influence debates.
• Empirical result from the review of 229 articles: about 75% focus on AI regulation; empirical studies of AI’s use in governance (AI-as-instrument) are marginal. The dominant framing is problem-solving: AI as optimizer of policymaking or as an objectively manageable risk.
• The literature’s skew toward regulation and problem-solving ontologies tends to de-politicize AI, downplay power dynamics, and underexamine social inequalities produced or amplified by AI-enabled governance.
• The article calls for greater reflexivity in social-science knowledge production, more debate about the political stakes of research choices, and more scholarship that centers power, redistribution, and socio-technical analysis.

Data & Methods

• Method: Systematic literature review of English-language social-science journal articles published through 2025.
• Corpus: 229 articles.
• Analytic tools:
  • A reflexive critique of systematic-review methods applied to critical policy scholarship.
  • A heuristic mapping approach that classifies each article along two ontological dimensions (AI as neutral tool ↔ socio-technical system; technology as governance instrument ↔ regulatory target).
• Limitations (as identified or implied):
  • Language restriction (English) and restriction to social‑science journals may omit relevant interdisciplinary or non‑English scholarship.
  • Systematic-review methods themselves are interrogated as potentially reproducing epistemic biases; classification depends on author coding/interpretation.

Implications for AI Economics

• Research agendas:
  • Move beyond regulation-focused, problem-solving narratives. Prioritize empirical studies of how AI is actually deployed in governance (e.g., tax administration, welfare delivery, regulatory monitoring) and how deployments reshape incentives, market structure, and distributional outcomes.
  • Integrate socio-technical and political‑economy perspectives into economic models of AI adoption, diffusion, and regulation to capture power asymmetries, path dependence, and strategic behavior (e.g., regulatory capture, incumbent entrenchment).
• Modelling and measurement:
  • Incorporate endogenous institutional responses and governance uses of AI into equilibrium frameworks rather than treating regulation as an exogenous constraint.
  • Measure distributional impacts of AI-enabled governance (labor, firms, regions, demographic groups) and feedback effects on demand, taxation, and public-good provision.
• Policy evaluation:
  • Evaluate regulatory interventions in terms of how they alter incentive structures and power relations, not only in terms of efficiency or risk mitigation.
  • Study alternative governance instruments (procurement rules, auditability, public-sector design choices) that shape AI’s effects on markets and inequality.
• Methods and interdisciplinarity:
  • Use mixed methods: qualitative institutional case studies, process tracing, field experiments, and administrative data analysis to capture mechanisms and contextual heterogeneity.
  • Maintain reflexivity about epistemic choices (data sources, framing, metrics) to avoid reproducing techno-solutionist biases.
• Normative attention:
  • Explicitly analyze distributional and welfare tradeoffs of AI policies and deployments; consider justice, accountability, and democratic oversight as economic outcomes, not only ethical afterthoughts.

Taken together, the paper suggests that AI economics should broaden its empirical and theoretical lenses to study AI as a socio-technical, governance-shaping force, adopt reflexive methodologies, and foreground power and distributional consequences when designing and evaluating policy.