A skills-focused simulation suggests AI’s technical reach in the U.S. labor market is far larger and more geographically widespread than visible adoption indicates: current AI capabilities overlap with about 11.7% of U.S. wages (~$1.2tn), compared with roughly 2.2% (~$211bn) reflected in present adoption concentrated in tech hubs.

Main Finding

Project Iceberg introduces a skills-centered measure (the Iceberg Index) and a large-scale agent-based simulation showing that AI technical capability in the U.S. labor market is substantially larger and far more geographically diffuse than visible adoption suggests. Visible AI adoption concentrated in computing/technology represents about 2.2% of U.S. wage value (~$211 billion), while cognitive automation potential across administrative, financial, and professional services amounts to 11.7% (~$1.2 trillion). Traditional macro indicators explain <5% of the resulting geographic variation, so skill-level exposure metrics and simulations are needed to anticipate and plan for AI’s broader labor-market effects.

Key Points

• Problem: Conventional workforce metrics capture outcomes after disruption (employment, wages) but not where AI capabilities overlap with human skills before adoption.
• Iceberg Index: A skills-centered metric measuring the wage value of the specific skills AI systems can perform within each occupation. It quantifies technical exposure (capability overlap), not displacement, adoption timelines, or realized outcomes.
• Scale & resolution: Model represents 151 million U.S. workers as autonomous agents, covers 32,000+ skills across ~3,000 counties, and links agents to thousands of AI tools.
• Key quantitative results:
  • Visible AI adoption in tech/computing: ~2.2% of wage value (~$211B).
  • Broader cognitive automation potential in admin, financial, professional services: ~11.7% (~$1.2T) — ~5× larger and spread across all states, not just coastal hubs.
  • Traditional indicators (GDP, income, unemployment) explain <5% of state- and county-level variation in skills-based exposure.
• Purpose: The framework lets policymakers and firms simulate alternative scenarios to locate exposure hotspots, prioritize investments (training, infrastructure), and test interventions prior to large deployments.
• Implementation: Model and simulations built with the AgentTorch framework.

Data & Methods

• Modeling approach: Large Population Models / agent-based simulation (agents = 151M workers) representing individual workers’ occupations, skills portfolios, wages, and county locations. Agents interact with a catalog of thousands of AI tools mapped to task/skill capabilities.
• Skill taxonomy: ~32,000 distinct skills mapped to occupations and tasks; each skill assigned wage value contribution within occupations to compute the Iceberg Index.
• Iceberg Index computation: For each occupation/geography, sum the wage value associated with skills that current AI capabilities can perform. The index is a measure of technical exposure (potential substitutability/complementarity), not an adoption or displacement probability.
• Geographic granularity: Analysis across ~3,000 U.S. counties and all states, enabling spatial dispersion analysis.
• Scenarios & simulations: Counterfactual scenarios that vary capability diffusion, adoption rates, policy interventions, and firm behavior to explore how exposures might translate into outcomes.
• Validation & limitations noted: The Index captures capability overlap but not firm adoption choices, regulatory constraints, social acceptance, complementarity effects, or reallocation dynamics. Simulations depend on mapping of AI tools to skills and assumptions embedded in AgentTorch agent behaviors.

Implications for AI Economics

• Measurement: Skills-centered exposure metrics (like the Iceberg Index) are essential complements to traditional employment and macro indicators for anticipating AI impacts before outcomes materialize.
• Policy targeting: Because exposure is geographically widespread and concentrated in service and administrative work as well as tech, policies should be spatially and sectorally granular (county- or state-level interventions, not only coastal/hub strategies).
• Labor-market interventions: Use the model to prioritize retraining, upskilling, and credentialing where wage-value exposure is high; align training investments to skills where AI is unlikely to fully substitute and where complementarities can raise productivity and wages.
• Infrastructure & economic development: Identify local infrastructure needs (broadband, compute, AI adoption support for SMEs) in exposure hotspots to avoid uneven benefits and displacement.
• Scenario testing for costly decisions: Firms and governments can simulate policy levers (subsidies, tax incentives, public-training programs, adoption incentives) to estimate likely labor-market trajectories before committing large budgets.
• Research agenda: Encourage more granular skill-to-AI-capability mappings, longitudinal tracking of adoption vs. exposure, and integration of firm behavior and regulatory dynamics into agent-based models to move from exposure assessment toward outcome prediction.
• Caution: The Iceberg Index signals where capability exists, not where job losses will occur or when. Policy responses should combine exposure analysis with empirical monitoring of adoption, complementarities, and local labor reallocation.