AI Efficiency Index : Identifying Regulatory and Policy Constraints for Resilient National AI Ecosystems

We develop efficiency estimates for the production, implementation and diffusion of artificial intelligence (AI) services. We use a variety of data measuring factors relating to AI input and output. We start by constructing a set of nine different AI efficiency measures using non-parametric technique of data envelopment analysis (DAE). We then proceed to analyse the cross-country time variation of these estimates and compare these with different policy measures and institutional indicators. In particular, we link our AI efficiency measures to general characteristics of a country's innovation system including indicators reflecting product market regulation, tax subsidies for R\&D investment and institutional factors linked to starting a business and (intellectual) property rights. We find that policy measures differ in their impact on our AI efficiency scores: Tax subsidies are important to enhance start-up investment activity, especially in countries with high barriers to entry and weak property rights. Product market frictions such as public ownership --- especially in the telecommunications industry --- is pernicious to patenting efficiency in the field of AI. Our results highlight the importance of differentiated policy interventions at different stages (and attributes) of the AI production chain