<b>Dynamic </b><b>p</b><b>ricing and </b><b>d</b><b>ispatching in </b><b>c</b><b>ar-</b><b>s</b><b>haring </b><b>s</b><b>ystems</b>

Artificial intelligence (AI) is reshaping intelligent mobility and shared transportation systems. However, current car-sharing system (CSS) optimization frameworks rarely capture social learning dynamics among users, which significantly influence mobility demand and market responses. To address this gap, we develop an AI-enabled dynamic decision-making framework integrating a monopoly CSS operator and heterogeneous consumers with network-driven social learning (SL). The operator optimizes real-time pricing and fleet dispatching decisions through deep reinforcement learning (DRL), while consumers make bounded-rational service and membership choices influenced by SL. A three-layer mobility ecosystem is constructed, consisting of a CSS station network, urban travel network, and consumer social network. Behavioral parameters are calibrated through user surveys, and two scenarios—with and without SL—are simulated. Results indicate that SL encourages lower real-time prices and reduces fleet redistribution pressure by stabilizing spatial demand, improving consumer welfare but reducing operator profit. SL also introduces short-term decision volatility due to imperfect rationality. This study contributes to AI-driven transportation research by integrating human behavior learning, multi-network mobility environments, and DRL-based policy optimization, providing insights for socially adaptive shared mobility services in future urban transportation systems.