Transaction Costs, Extended Phenotype, and Liability for Artificial Intelligence: A Common Conceptual Framework

This working draft proposes a common conceptual framework for three problems usually treated separately in the literature on liability for harm caused by artificial intelligence: the optimal design of liability rules, the prediction of their success when transplanted across jurisdictions, and the conceptual limit of the framework if AI systems begin to behave as agents with their own interest in persistence. The paper rereads Coase, Calabresi and Melamed, and Shavell through transaction-cost economics, Extended Phenotype Theory, and heterogeneous intentionality, arguing that AI liability must be evaluated not only for static efficiency but also for evolutionary stability under non-compliance pressure, institutional transplantation, and changing agent architecture. This record is part of the Law as Extended Phenotype research program.