tl;dr: Pandemics, supply chains, labor markets. Every population-scale crisis has an iceberg below the waterline. Feedback loops create it: individual responses compound across millions faster than measurement can follow

Crises Cascade Before We Act

AI is changing how work gets done by automating tasks, restructuring workflows, and shifting which skills matter. Adoption is uneven and difficult to measure in real time. What is already visible in the labor market is harder to ignore: hiring slowdowns, preemptive layoffs, and roles being restructured quietly before any formal policy response exists. The disruption is not waiting for clear measurement. It is already spreading. But what is visible is only the surface.

The rest is hidden because disruption does not spread evenly or legibly. It moves through feedback loops: individual responses that compound across workers, firms, and industries faster than any official metric can track. By the time the cascade registers, structural adjustments below the surface have already occurred. The question is whether we can see what is forming underneath before it becomes crisis. 

Feedback Loops Shape the Iceberg

Population-scale disruptions unfold through feedback loops. Individual behavior shapes system outcomes, and system outcomes reshape individual behavior. Because these loops operate through interconnected behavior across millions of people, small shifts can cascade into outbreaks, supply shocks, or labor market disruptions that no single indicator predicted. Miss the feedback loop and you miss how fast things escalate and what the right intervention actually is. 

 Post-pandemic vaccine hesitancy led to measles resurgences in communities that had previously contained the disease. Bird flu on poultry farms triggered consumer panic that moved through dairy supply chains far from any infected flock. In both cases the cascade was driven by behavioral response interacting with system structure, not by the original trigger alone. The right intervention targeted behavior, not just the source. 

 COVID-19 demonstrated this at national scale. Policymakers prioritized test speed because faster information altered behavior sooner. Public spending concentrated on behavioral incentives such as stimulus payments, mobility support, and communication campaigns because behavior determined transmission dynamics. Getting the science right changed real decisions. 

 Intervening early requires modeling the population itself. 

Large Population Models

Large Population Models (LPMs) construct digital societies composed of millions of autonomous agents. Each agent represents an individual situated in context: occupation, skills, geography, household constraints, and industry exposure. Agents make decisions, interact through networks, and respond to shocks. 

 LPMs explicitly represent individual behavior and the interactions between individuals. System-level outcomes emerge from these networked interactions over time. When a shock is introduced, whether a virus, a supply chain disruption, a policy shift, or a new AI capability, the system traces how responses propagate across connected individuals, firms, and regions. The system captures variation in adaptation speed, emerging bottlenecks, and regional amplification or stabilization effects. 

 These models are calibrated to real-world data and executed on scalable infrastructure. Policy changes, technological shifts, or training programs can be introduced and their effects evaluated before implementation. 

 During COVID-19, this infrastructure supported epidemic modeling, vaccination strategy evaluation, and behavioral intervention analysis at scale. The same foundation now supports a national workforce model, capturing how AI capability overlaps with human skills and propagates through labor markets at scale. That model represents over 150 million worker agents, each mapped to detailed skill profiles, occupations, industries, and counties.

Workforce Security

AI capability is advancing rapidly across administrative, financial, professional, and technical work. Some sectors are seeing genuine productivity growth. Others are navigating workflow restructuring. Beneath both, skills are being repriced before workers receive signals to retrain. These effects are not contained to the technology sector. They ripple across industries and regions through the roles that support manufacturing, logistics, healthcare, and public services. 

 Most of this shift remains below the surface. Visible AI adoption, concentrated in computing and technology roles, represents about 2.2% of labor market wage value. But AI technical capability already overlaps with skills across finance, healthcare, and professional services, reaching 11.7% of the labor market, roughly 1.2 trillion dollars in wage value. That represents a fivefold gap between what surface indicators detect and what is already structurally exposed. Unlike technology-sector adoption clustered in coastal hubs, this broader exposure is distributed across every state. 

Traditional metrics such as GDP and unemployment measure outcomes after adjustment has already occurred. By that stage, communities are already absorbing the impact. 

 Understanding the full picture requires measuring exposure patterns across workers situated in their specific contexts: occupation, industry, geography, and household circumstances. An auto worker in rural Tennessee facing AI-driven task changes is in a different position from one in Seattle. Those differences shape how disruption spreads and where ripple effects land. Modeling the population requires capturing that heterogeneity and tracing how individual responses compound across millions of workers. 

 The Iceberg Index applies LPM infrastructure to this challenge. It maps where human and AI capability overlap across the economy and measures how exposure patterns compound across workers, firms, and regions before they appear in official statistics. The State of Tennessee formally adopted it in its 2026 AI Action Plan as the framework for monitoring AI's impact on its workforce [1].

What Is Next

States are adopting the Iceberg Index to monitor labor market exposure, measure how disruption moves across industries, and test workforce interventions before committing to them. We are expanding to full national coverage at ZIP-code granularity so that institutions can identify where exposure exists, track how it is unfolding, and assess risk at the community level. Workforce security requires early measurement, population-scale simulation, and proactive intervention.

References:

[1]: TN's AI Action Plan

[2]: Project Iceberg Index Report