How should the government respond to automation? We study this question in a heterogeneous agent model that takes worker displacement seriously. We recognize that displaced workers face two frictions in practice: reallocation is slow and borrowing is limited. We analyze a second best problem where the government can tax automation but lacks redistributive tools to fully alleviate borrowing frictions. The equilibrium is (constrained) inefficient and automation is excessive. Firms do not internalize that automation depresses the income of automated workers early on during the transition, precisely when they become borrowing constrained. The government finds it optimal to slow down automation on efficiency grounds, even when it does not value equity. Quantitatively, the optimal speed of automation is considerably lower than at the laissez-faire. The optimal policy improves efficiency and delivers meaningful welfare gains.

Developing artificial intelligence (AI) technology requires data. In many domains, government data far exceed in magnitude and scope data collected by the private sector, and AI firms often gain access to such data when providing services to the state. We argue that such access can stimulate commercial AI innovation in part because data and trained algorithms are shareable across government and commercial uses. We gather comprehensive information on firms and public security procurement contracts in China’s facial recognition AI industry. We quantify the data accessible through contracts by measuring public security agencies’ capacity to collect surveillance video. Using a triple-differences strategy, we find that data-rich contracts, compared to data-scarce ones, lead recipient firms to develop significantly and substantially more commercial AI software. Our analysis suggests a contribution of government data to the rise of China’s facial recognition AI firms, and that states’ data collection and provision policies could shape AI innovation.