The Concave Relationship Between AI Exposure and Unemployment: Reframing the Supervisory Economy as an Exploratory Moderation Test

Background Rapid advances in general-purpose artificial intelligence are compressing automation timelines and renewing concern about technological unemployment. This article examines whether aggregate AI exposure is associated with unemployment in a cross-country panel, and whether a broad managerial-share proxy provides any evidence for the proposed “supervisory economy” mechanism. Methods Using a balanced panel of 12 economies observed annually from 2014 to 2023, we construct a sector-weighted AI-exposure index and match it to labour-force data on unemployment, senior- and middle-management employment, public transfers, R&D, and GDP per capita. Two-way fixed-effects regressions are estimated linearly and with a quadratic AI term to test non-linearity within the observed support. Results The preferred quadratic specification reveals an inverted-U association between aggregate AI exposure and unemployment: joblessness rises at low-to-moderate exposure but falls once exposure reaches the upper end of the sample distribution. The managerial-share proxy has no significant standalone effect and does not significantly moderate the AI-unemployment association. Conclusions The most robust empirical contribution is the concave AI-unemployment relationship. The supervisory-economy argument should therefore be read as a conceptual and policy-research agenda rather than as a mechanism directly identified by the present proxy. Future work requires vacancy-level or occupation-level measures of AI governance, algorithmic-risk, model-monitoring and prompt-engineering roles to test the mechanism directly.