Workers facing higher AI substitution risk — and those who feel personally threatened — push for stricter AI rules; optimism about AI's benefits does not reduce demand for regulation.

Main Finding

Workers whose occupations are objectively more exposed to AI-driven task automation — and workers who personally perceive greater risk to their jobs from AI — are significantly more likely to support stronger government regulation of AI. This relationship holds across multiple specifications and within-person panel analysis. There is a clear asymmetry: pessimism and exposure to substitution risks increase regulatory support, but optimism or exposure to complementary opportunities do not reduce it.

Key Points

• Distinction between objective and subjective exposure:
  • Objective exposure: occupation-level measures of AI’s task capabilities (i.e., how much tasks in an occupation can be automated or augmented by AI).
  • Subjective exposure: workers’ own beliefs about how AI will affect their job prospects (threats or opportunities).
• Both higher objective exposure to job-substituting AI and greater subjective perceived risk predict stronger support for government regulation of AI.
• Panel (within-person) evidence shows that increases in an individual’s concern about AI’s effects on their job are followed by increases in their support for regulation — supporting a causal interpretation at the individual level.
• Asymmetry in effects:
  • Negative effects (pessimism, substitution risk) increase regulatory support.
  • Positive effects (optimism, complementarities/opportunities) do not produce a corresponding decrease in regulatory support.
• Results are robust across multiple model specifications.

Data & Methods

• Data:
  • Original nationally-representative survey data from Britain measuring political attitudes, perceived job threats/opportunities from AI, and other covariates.
  • Occupation-level measures of AI exposure constructed from task-capability mappings (i.e., how AI can perform or augment occupational tasks).
  • A panel component of the survey enabling within-person longitudinal analysis.
• Empirical strategy:
  • Multivariate regression analyses linking objective occupation-level AI exposure and subjective beliefs to support for AI regulation.
  • Robustness checks across multiple specifications.
  • Panel fixed-effects or within-person models to identify changes in regulatory support as individual perceptions change over time.
• Identification strengths and limits:
  • Combining objective occupation measures with subjective perceptions improves inference about occupational self-interest vs. attitudes.
  • Panel analysis strengthens causal claims about changing perceptions driving attitude change.
  • No single-country, survey-based study can fully rule out all confounders (see limitations).

Implications for AI Economics

• Occupational self-interest matters: labor-market exposure to AI helps shape political demand for regulation. This links micro-level labor economics (task-based exposure) to macro-level political economy of technology.
• Regulation dynamics likely driven more by perceived downside risks than by expected gains from complementarities. Policymakers should expect pressure for restrictive or protective regulation from workers facing substitution risk even if many stand to gain from AI augmentation.
• Political mobilization & policy design:
  • Areas and occupations with high substitution exposure may form constituencies favoring tighter AI rules, influencing national regulatory agendas.
  • Policies that only highlight positive complementarities may not be sufficient to reduce demand for regulation; addressing downside risks (safety nets, retraining, transitional assistance) may be necessary to shift political preferences.
• Labor-market policy relevance:
  • Findings support targeting worker protections, re-skilling programs, and active labor-market policies in high-exposure occupations to mitigate political backlash and smooth adoption.
• Research and policy caveats:
  • Evidence is from Britain; generalizability to other political and labor-market contexts requires further study.
  • Measurement of “objective exposure” depends on the task-capability mapping used; improvements in mapping AI capabilities may refine estimates.
  • Future work should examine heterogeneity by industry, firm size, political ideology, and cross-country institutional differences, and investigate long-term effects as AI adoption evolves.