Market Design for AI: Beyond the Copyright Binary

Abstract

How can we design a market of human-generated content for use by generative artificial intelligence (AI) that both enables technological progress and preserves individuals’ incentive to create high-quality content? Existing answers to this important question take two strong forms: Some argue that the technological-progress imperative requires that the use of copyrighted content in model training qualify as a noninfringing fair use (the “free for all” model) while others argue that strong intellectual property rights are necessary to preserve human creative incentives (the “strong property rights” model). We show, using a game-theoretic model, that both answers fall short. One of our central findings is that not only the free-for-all model but also, counterintuitively, the strong property rights regime lead to underpowered creative incentives. This result inverts common wisdom on the incentive-boosting effect of intellectual property rights. We show, further, that a market characterized by strong individual property rights will privilege homogenized content over creative contributions. Dynamically, the degradation and homogenization of human content eventually undermines AI performance. We draw on the static and dynamic market failures we uncover to propose a novel design of a data market for AI, incorporating a data intermediary who can help perserve creative incentives while enabling technological progress. (NOTE: I selected Law as the Principal Discipline, but the paper would be equally at home in Economics.)