Abstract

In this project, we use high-frequency job advertisements linked to administrative establishment data to estimate the impact of investments into artificial intelligence (AI) technologies on labor demand. The fundamental identification problem in estimating causal effects of AI on labor demand is that establishments with AI investments exhibit structural differences compared to establishments without AI investments that are mostly unobservable (e.g., different productivity levels or being affected by demand shocks). To address this identification problem, the empirical literature on the identification of causal effects of investments into AI technologies has so far mainly used instrumental variable strategies which validity hinges on untestable exogeneity assumptions. To circumvent this, we use a dynamic difference-in-differences approach. This approach has the advantage that we can only consider firms with AI investments and can exploit differences in the timing of AI investments. We measure the timing of AI investment by using job postings related to AI that are complementary to the adoption of AI technologies.