Incentives, Equilibria, and the Limits of Healthcare AI: A Game-Theoretic Perspective

A new paper from Ari Ercole at the University of Cambridge applies game theory to challenge the prevailing optimism around AI in healthcare. Published on arXiv, the research argues that deploying AI as a simple 'deus ex machina' solution ignores the entrenched incentive structures that govern hospital systems. The paper categorizes healthcare AI into three distinct types: AI for effort reduction (automating tasks), AI to increase observability (better data tracking), and mechanism-level incentive change AI (altering payment/risk models).

Using a stylized model of inpatient capacity signaling—where hospitals decide whether to admit patients—the analysis demonstrates a critical insight. Simply making processes more efficient with AI does not change the system's equilibrium if the underlying financial incentives for providers (like fee-for-service vs. value-based care) remain the same. The paper's game-theoretic reasoning shows that only the third type of AI, which actively reshapes how risk and reward are allocated among stakeholders, can break stable but suboptimal patterns of behavior. This has direct implications for healthcare leaders and procurement teams, urging them to evaluate AI not just on technical performance but on its potential to redesign economic incentives.