Polanyi Knowledge and AI: Why Intelligence Alone Is Not Enough

Summary

Arnold Kling discusses Tyler Cowen’s argument that the Solow model is insufficient for understanding major economic changes caused by AI. The proposed alternative distinguishes between formal intelligence and “Polanyi knowledge”: tacit, context-specific knowledge of time and place, custom, habit, physical properties, and social interaction. Kling argues that AI models can be trained relatively well for the digital world because text and code provide large data stores, but available data is much poorer for the physical and social worlds. Examples include the practical knowledge of dental assistants and personal trainers, and the ability of human guides at Alpha School to motivate students in ways its AI tutors cannot. He also argues that productivity differences between workers and countries reflect Polanyi knowledge embedded in supervision, firms, legal systems, and social institutions, which forms part of the Solow residual and helps explain differences in living standards. On this view, claims that current models are near artificial general intelligence, fears that AI will soon eliminate all jobs, and calls for broad AI regulation may all underestimate tacit knowledge. Kling concludes that sound judgment about AI requires thinking explicitly about where this knowledge resides and how difficult it is to digitize.