Human creators may find a balance with generative AI

A competition model presented at ICML 2024 examines whether human creators and generative AI can coexist when they compete for audience attention. AI learns from human content, improving with diminishing returns, while additional content attracts users until demand saturates. AI-generated output does not contribute to further training in the model.

When generative AI operates as an outside competitor, the market reaches a unique stable outcome under specified conditions. More efficient creators produce and earn more, while AI changes total human output by at most a constant factor regardless of its learning speed. That market-level stability does not protect everyone: simulations show stronger AI reducing total human effort and earnings.

When every creator can adopt AI at almost no cost, stability is no longer guaranteed. Choices can cycle indefinitely when creators work across topics with linked costs. For a single topic, an equilibrium exists under a technical condition, with the highest-cost creators switching to AI. Simulations suggest greater AI capability leaves fewer human creators and lowers their combined output and earnings, while those who remain produce and earn more on average. Human activity also shifts toward niche topics as AI becomes stronger.