Generative AI and Data Quality: Implications for Productivity, Labor Displacement, and Policy

Abstract Generative artificial intelligence (AI) is increasingly consuming and producing huge amounts of data. We propose a social learning model of AI, emphasizing a data-AI feedback loop: data quality affects AI productivity, which influences AI adoption and, consequently, the composition (AI versus human-generated) and quality of future data. Calibrated to evidence on synthetic training loops, the model predicts hump-shaped labor dynamics—short-term displacement that partially reverses as data quality deteriorates. A Grossman–Stiglitz-style externality emerges: AI adopters free-ride on the human-generated actions that supply the novel information on which AI itself relies. In a competitive market, AI should be taxed to correct the data-quality externality; a concentrated AI industry overcorrects, making a subsidy optimal.