Mapping and Comparing AI Policy Portfolios: A Pipeline for Classifying Targets and Instruments

This paper uses a policy portfolio approach to analyse how artificial intelligence (AI) is governed across a diverse set of countries. We collect and classify AI-related policies into two key dimensions: targets, denoting the specific objectives pursued by each policy, and instruments, referring to the regulatory or programmatic tools employed. In addition to building the dataset, we present a comparative description of how policy portfolios vary across countries along these two dimensions. This data collection and classification method is based on a pipeline that combines tools grounded in text analysis and generative AI (Large Language Models, LLMs). The goal is to provide a general method of policy mapping that is both scalable (to other policy sectors) and not restricted to specific constituencies (usable for countries, regions, or local entities). We detail the steps of the pipeline and the classification scheme that enable systematic cross-country comparison while handling heterogeneity in policy formats and language. While our focus is on collecting and classifying these policies, future research will analyse how these policy portfolios might influence broader trajectories of technological development and innovation. By highlighting patterns of both convergence and divergence in AI regulation, this study offers an empirical foundation for discussions about the politics of AI and lays the groundwork for subsequent analyses of broader policy effects. For panel: P1