A major gap exists between which jobs are technically exposed to AI and which workers actually adopt the technology in practice. Researchers from the University of Chicago's Becker Friedman Institute found that traditional measures of AI exposure explain only 14% of the variation in actual AI adoption across occupations, while a new framework based on comparative advantage accounts for 60% of that variation.

What's the Difference Between AI Exposure and AI Adoption?

The study, which analyzed data from a representative German employee survey linked to worker and establishment information, reveals a critical blind spot in how economists and policymakers think about AI's impact on the labor market. Just because a job is technically exposed to AI doesn't mean workers will actually use it. The researchers compared worker-reported AI use to prominent exposure measures and found the relationship was surprisingly weak.

This distinction matters enormously for understanding AI's real-world impact. Exposure measures typically ask: "Could AI theoretically do this task?" Adoption asks: "Will this worker actually use AI for this task, and will it make economic sense to do so?" These are fundamentally different questions with very different answers.

How Does Comparative Advantage Predict AI Adoption Better?

The researchers developed a new framework that considers not just whether AI can do a job, but whether it makes financial sense for a specific worker to use it. This framework balances several factors simultaneously: AI's productivity gains, the costs of using AI, the worker's own productivity, and the worker's wage.

Think of it this way: a highly paid specialist might find it uneconomical to use AI for routine tasks because their time is too valuable to spend on setup and oversight. Meanwhile, a lower-wage worker doing similar work might find AI adoption highly profitable. The same technology, the same task, but opposite adoption decisions based on comparative advantage.

• Technical Feasibility: Whether AI can theoretically perform a task, measured by traditional exposure indices that assess AI's absolute advantage in specific job functions.
• Profitability Calculation: Whether using AI makes financial sense for a particular worker, balancing AI productivity gains against the costs of implementation and the worker's own productivity relative to their wages.
• Task-Level Economics: The researchers operationalized their framework by estimating worker productivity relative to pay, mapping exposure indices into AI productivity gains, and inferring task-specific AI user costs from revealed-preference adoption patterns.

The results were striking. The two approaches diverged substantially for approximately 30% of workers, highlighting that comparative advantage, not exposure alone, is crucial for assessing AI's labor-market impact. This means that roughly one-third of the workforce would be misclassified if policymakers relied solely on exposure measures to predict AI adoption.

Why This Matters for AI Risk and Labor Market Planning

This research carries significant implications for how we think about AI's broader economic and social risks. If we misunderstand which workers will actually adopt AI and which won't, we'll likely mispredicate the technology's labor market disruption. Policymakers designing retraining programs, social safety nets, or labor market interventions need accurate forecasts of where AI adoption will actually occur, not just where it's technically possible.

The study suggests that AI adoption will be far more uneven and economically driven than simple exposure measures would suggest. Workers in high-wage positions may resist AI adoption even when it's technically feasible, while workers in lower-wage positions may embrace it enthusiastically. This creates a more complex picture of AI's economic impact than the narrative of "AI will displace workers in exposed occupations" typically suggests.

Steps to Better Predict AI Adoption in Your Organization

• Assess Comparative Advantage: Don't just ask whether AI can do a task; calculate whether it makes economic sense for each worker or team, considering their wage, productivity, and the costs of implementation and oversight.
• Map Task-Level Economics: Break down adoption decisions at the task level rather than the occupation level, since the same worker might adopt AI for some tasks but not others based on profitability calculations.
• Monitor Revealed Preferences: Track which workers actually use AI tools and infer the real user costs from their adoption patterns, rather than relying on theoretical estimates of what adoption should look like.

The Becker Friedman Institute's research underscores a fundamental principle: technology adoption is not inevitable or uniform. It's driven by economic incentives that vary dramatically across workers and tasks. Understanding these incentives is essential for predicting where AI will actually reshape work, and where it will remain a theoretical possibility that workers rationally choose not to pursue.

As organizations and governments plan for an AI-transformed economy, this research suggests they should move beyond simple exposure measures and develop more sophisticated models of adoption based on the actual economics facing different workers and firms. The gap between what AI can do and what workers will actually do with it may be the most important variable in understanding AI's real-world impact on labor markets and economic inequality.