Artificial Intelligence as a Meta Good: Dynamic Properties and Economic Implications

ABSTRACT We introduce two categories, meta goods and dynamic goods, that extend the standard taxonomy of economic goods. A meta good generates downstream outputs (e.g., text, code, images), expanding the feasible consumption set; a dynamic good improves endogenously with use, feedback, and retraining. We develop a welfare‐theoretic model in which artificial intelligence (AI) exhibits both properties. The main result shows underprovision of adoption in decentralized equilibrium when learning‐by‐using and data externalities are present, relative to a social planner. Formal propositions establish the taxonomy and the welfare ranking, and a stylized calibration illustrates diffusion and quality dynamics consistent with theory. The framework generalizes beyond AI to other general‐purpose technologies. These findings speak directly to ongoing global debates on AI governance, open versus closed access, and the role of international institutions in shaping equitable and socially inclusive diffusion.