Paper

A Global Task-Skill Atlas of Automation Exposure

Automation exposure is often measured as if it were a context-free property of occupations or tasks, even though automation arrives through concrete work activities, depends on complementary skills, and varies across countries. We build a global task–skill atlas of automation exposure designed around that problem. The atlas combines validated many-to-many links between tasks and skills with a country-conditioned automation layer and a product/service–task bridge to technology-bearing goods and services. The atlas links 6,275 ESCO skills to 15,586 O*NET tasks through 68,824 retained edges. We then classify the full 23,851-task universe under a common four-part measurement framework covering exposure level, technology channel, labor margin, and implementation dependency, and apply that framework to 142 countries. This yields 3,386,842 country-task labels with zero residual normalization failures and supports a GDP-weighted synthetic global benchmark built from country-conditioned task labels, which we compare directly to a matched context-free benchmark. That replacement changes the measured global benchmark materially. Across tasks, channel concordance between the matched context-free benchmark and the GDP-weighted synthetic global is 0.788, while margin concordance is 0.860 and dependency concordance is 0.855. The synthetic global is nonetheless robust to weighting choices, with task-level Spearman concordance between 0.970 and 0.988 across GDP-weighted, population-weighted, and unweighted country means. Cross-country heterogeneity is also substantive: average task exposure ranges from 0.315 to 2.101 across countries, and tasks with larger cross-country dispersion also tend to move more when the benchmark becomes country-conditioned. We further connect the atlas to 23,827 retained product-task links, of which 17,212 carry explicit role labels, and to goods-side country-task projections covering 11.8 million rows across 141 goods-trade reporters. The contribution is therefore not a single index but a reusable work-technology measurement system linking tasks, skills, countries, products, and trade-facing technology channels. The Automation Atlas measures feasibility, mechanism, and implementation constraints; it does not estimate realized adoption, displacement, wages, productivity, or causal labor-market effects.