(iPoster) Uncovering Policy Demand Responses to AI-Driven Automation

• In Event: Poster Session: Public Opinion 2

Abstract

Why do citizens often fail to update their policy preferences when presented with objective evidence of economic threats? We address this "information inertia" puzzle through the lens of AI-driven job automation. While standard accounts suggest automation risk information should increase demand for government protection, recent experiments frequently report null effects. We propose the Conditional Activation Model (CAM), a framework positing that information activates policy demand only when it passes two sequential gates: (1) source trust and (2) anxiety triggered by a belief gap—a positive shock revealing that objective risk exceeds prior expectations. Leveraging a pre-registered experiment with 901 U.S. workers who received personalized automation-risk scores derived from validated occupational indices, and a Bayesian finite mixture model to recover latent trust profiles, we demonstrate that standard Average Treatment Effects (ATE) mask meaningful heterogeneity. While the aggregate effect is modest, information increases support for compensatory policies—such as expanded unemployment benefits and worker retraining—by 0.61 standard deviations among high-anxiety individuals who trust the source and discover they face greater risk than expected. These results reconcile conflicting findings in the persuasion literature and suggest that public communication strategies must account for trust heterogeneity to effectively mobilize support for policies addressing technological disruption.

Author

• Ignacio Urbina, Stony Brook University