Artificial intelligence and productivity: global evidence from AI patent and bibliometric data

In this paper we analyse the relationship between technological innovation in the artificial intelligence (AI) domain and productivity. We embed recently released data on patents and publications related to AI in an augmented model of productivity growth, which we estimate for the OECD countries and compare to an extended sample including non-OECD countries. Our instrumental variable estimates, which account for AI endogeneity, provide evidence in favour of the modern productivity paradox. We show that the development of AI technologies remains a niche innovation phenomenon with a negligible role in the officially recorded productivity growth process. This general result, i.e. a lack of a strong relationship between AI and macroeconomic productivity growth, is robust to changes in the country sample, in the way we quantify labour productivity and technology (including AI stock), in the specification of the empirical model (control variables) and in estimation methods.