Abstract

How does AI think about economic policy? While the use of large languagemodels (LLMs) in economics is growing exponentially, their assumptions oneconomic issues remain a black box. This paper uses a conjoint experiment totease out the main factors influencing LLMs' evaluation of economic policy. Itfinds that LLMs are most sensitive to unemployment, inequality, financialstability, and environmental harm and less sensitive to traditionalmacroeconomic concerns such as economic growth, inflation, and government debt.The results are remarkably consistent across scenarios and across models.

 

Quick Read (beta)