Towards Agentic Urban Digital Twins (AUDiTs): advancing new urban science through Human-AI co-learning agents

Abstract While digital twins provide predictive and planning capabilities, they often underrepresent social complexity, ethical considerations, and stakeholder participation. Concurrently, advances in artificial intelligence offer new opportunities for real-time sensing, adaptive learning, and decision support; yet current applications remain narrowly focused on optimization or monitoring. This falls short of the broader mission of New Urban Science: to Explain, Discover, Understand, and Generalize (EDUG) complex socio-technical systems. This vision paper introduces Agentic Urban Digital Twins (AUDiTs), a research agenda for integrating large language model (LLM) and multimodal agents into digital twin environments to foster human and AI co-learning, context-aware reasoning, participatory scenario design, and ethical deliberation. We identify critical challenges such as bias and fairness in foundation models, limitations in data-scarce contexts, computational sustainability, and institutional alignment, and we outline pathways for addressing them. AUDiTs extend existing digital twin paradigms by shifting emphasis from purely technical prediction toward collaborative, explainable, and value-sensitive urban intelligence, positioning digital twins as platforms for responsible and adaptive human and AI partnership in shaping just and resilient urban futures.