Measuring artificial intelligence: a systematic assessment and implications for governance

Abstract

Governing artificial intelligence (AI) inventions is a major policy concern, yet definitions and measurement remain contested. We compare four patent-based approaches reflecting distinct understandings of AI. Using US patents (1990-2019), we assess the degree to which each approach classifies AI as a general-purpose technology (GPT) and examine patent concentration--two central policy-relevant dimensions. The approaches overlap in just 1.37% of patents, defining between 3-17% of all US patents in 2019 as AI. All approaches confirm AI's GPT characteristics, with the smallest keyword-based set exhibiting the highest growth and generality. High GPTness indicates public good characteristics, justifying public support. Across methods, AI patents concentrate among a few firms, highlighting market power and regulatory challenges. Policy implementation, thus, requires careful consideration of multiple classification methods to ensure robust, inclusive, and effective AI governance.