Governing AI innovation diffusion through policy mixes: strategic interactions in an innovation ecosystem

The diffusion of AI technologies depends on both technological development and coordinated governance among heterogeneous stakeholders. This study develops a three-agent evolutionary game model to examine how policy mixes influence AI innovation diffusion among research institutions, enterprises and users under incentive, regulatory and informational policies. The results reveal heterogeneous and non-linear responses across stakeholder groups. Informational and subsidy policies generate strong responses from enterprises and users, whereas regulatory intervention exhibits a threshold effect. Moderate regulation enhances trust and coordination by reducing uncertainty, while excessive regulation increases compliance costs, weakens innovation incentives and shifts the system from a high-innovation to a low-innovation equilibrium. Policy interactions also exhibit threshold-dependent complementarities. Subsidies mitigate regulatory crowding-out effects, while information dissemination enhances regulatory effectiveness. Together, incentives and information promote adoption by reducing uncertainty and aligning expectations. These findings highlight the importance of adaptive governance and suggest that policymakers should adopt differentiated, stage-sensitive policy mixes combining appropriate regulation, incentives and information to support stable diffusion and ecosystem-wide participation.