Cities with dense skill networks absorb AI shocks by reallocating workers into skill-adjacent jobs, but the process raises inequality; sparse-network rural areas and compute-poor developing regions lack local alternatives and risk prolonged job and participation losses unless policy invests in infrastructure, skill connectivity and portable protections.

Main Finding

AI’s labor-market effects are highly spatially heterogeneous. Beyond occupational exposure, a region’s resilience to AI depends on (1) the density of its local occupational-skill network (which governs the availability of skill-adjacent reallocation paths) and (2) labor mobility patterns (which determine whether shocks are absorbed locally or through migration). Cities with dense, connected skill networks tend to absorb automation shocks but experience rising within-city wage inequality; rural and low-resilience regions face deeper, persistent job loss and labor-force withdrawal. Developing countries face a different threat—structural dependence on the global AI value chain (compute deserts, ghost work, weak data sovereignty) that creates high systemic risk despite lower direct automation exposure.

Key Points

• Mechanisms integrated:
  • Task-based substitution/augmentation (AI replaces some tasks, creates others).
  • Occupational network resilience: occupations embedded in dense, skill-similar networks allow easier occupational switching.
  • Labor mobility: migration and occupational switching mediate whether adjustment is temporary or persistent; mobility can either equalize or exacerbate regional divergence.
• Spatial heterogeneity:
  • Cities: skill upgrading and inflows of high-skill workers preserve labor-force participation but increase wage inequality.
  • Rural/low-resilience areas: loss of low-skill jobs often leads to exit from the labor force and long-term scarring.
  • Within-country pockets: AI-related job growth concentrates in innovation-rich, digitally connected regions.
• Global South: lower formal exposure but high vulnerability due to:
  • Concentration of high-value AI activities in few countries.
  • Dependence on low-paid data labeling / moderation work (“ghost work”).
  • Compute deserts (limited electricity, broadband, affordable compute) and weak data sovereignty.
• Trade-offs and dynamics:
  • Regions that are most “resilient” (dense skill networks) can also be most exposed to AI, creating a resilience–risk paradox.
  • Migration can turn brain drain into brain circulation with policy, but without it, flows amplify core–periphery divergence.
• Evidence gaps:
  • Most occupational-network studies are from high-income countries.
  • Poor coverage of Africa, South Asia, informal sectors, and long-term longitudinal outcomes.

Data & Methods

• Nature of the paper: literature review/synthesis that integrates three strands—task-based models, occupational network theory, and labor mobility research—into a spatial framework.
• Empirical evidence cited (types and examples):
  • Task-exposure and cross-country aggregates (e.g., IMF exposure estimates by income group).
  • Regional case studies: Italian NUTS-3 panel (Capello & Lenzi) showing divergent urban/rural adjustments; US commuting-zone and county-level analyses (Moro et al., Andreadis et al.) linking occupational networks or local characteristics to resilience and AI job shares.
  • Occupational network construction: nodes = occupations, edges = skill similarity; used to measure connectivity and shock absorption (Moro et al., network-based resilience metrics).
  • Mobility and knowledge diffusion: linked worker-migration and patent records (Giorgi et al.) showing how migrating skilled workers transmit capabilities.
  • Job-posting and administrative data: county-level AI job shares, skill composition, STEM degrees, patents, labor-market tightness (Andreadis et al.).
  • Development studies and qualitative/quantitative analyses on compute infrastructure, informal work, and “ghost work” (Lehdonvirta, Chigbu, Gray & Suri).
• Methods highlighted or recommended:
  • Task-based exposure measures extended by local task composition (instead of applying U.S. scores universally).
  • Occupational-skill network analysis (skill proximity, centrality, connectivity).
  • Linked employer–employee and migration–patent linkage designs to trace knowledge diffusion and reallocation.
  • Use of administrative panels, job-posting data, and region-level infrastructure measures to capture heterogeneity.
• Limitations noted in the literature:
  • Limited longitudinal career-level data on reallocation after AI shocks.
  • Underrepresentation of developing countries, informal sectors, and compute-access metrics.
  • Existing exposure indices risk mismeasurement when applied across geographies without adjusting for local task mixes.

Implications for AI Economics

Policy, measurement, and research implications for the economics of AI:

• Measurement & empirical strategy

  • Move beyond occupational-exposure indices: measure local task compositions and build region-specific exposure scores.
  • Incorporate occupational-skill networks into empirical models to predict reallocation capacity and distributional outcomes.
  • Collect/use longitudinal linked employer–employee data, migration flows, job postings, and patent-worker links to observe dynamics of reallocation, wages, and scarring.
  • Include informal-sector and compute-infrastructure variables (electricity, broadband, local cloud/GPUs) in cross-country and within-country analyses.

• Modeling & theory

  • Integrate network topology and mobility costs into task-based models to generate spatially explicit counterfactuals (cascades of displacement vs. reallocation).
  • Model trade-offs between regional resilience and within-region inequality (how resilience via in-migration and skill upgrading may raise local inequality).

• Policy design (targeted, spatially differentiated)

  • Strengthen local occupational networks: fund cross-occupational training, create regional skill maps, and attract anchor occupations to thicken skill adjacency.
  • Foster brain circulation, not one-way brain drain: portable credentials, remote-work facilitation tied to local retention, regional innovation branches.
  • Close compute and infrastructure gaps: invest in green data centers, public AI compute platforms, broadband and power in low-connectivity regions; promote open-source models and data-benefit-sharing rules.
  • Tailor social protection by local conditions: active labor-market programs and relocation support in low-resilience regions; redistribution and algorithmic protections (transparency, appeal rights) in high-resilience metros.
  • Protect informal workers: portable benefits and certification systems to include informal laborers in upskilling and safety nets.

• Research priorities for AI economics

  • Empirically quantify how occupational-network connectivity moderates AI impacts on employment, wages, and participation across many countries and over longer horizons.
  • Study the causal effects of compute investments and local cloud access on local AI adoption and labor outcomes.
  • Examine policy experiments that convert migration into circulation (e.g., incentives for return migration, remote-work hubs) and their effects on regional inequality.
  • Incorporate climate and infrastructure shocks into models of AI-driven regional change, and study interactions with informal-sector dynamics.

Takeaway: AI’s labor impacts are fundamentally spatial. Effective economic analysis and policy must combine task-based exposure measures with occupational-network metrics and mobility constraints, and must recognize distinct policy needs across cities, peripheries, and countries.