Abstract
Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by deficits in social communication and restricted behaviors. Traditional assessment and intervention methods rely heavily on subjective and time-consuming approaches, which limit their clinical impact. Advances in artificial intelligence (AI) offer transformative opportunities for ASD research and practice. This narrative review proposes six AI-driven strategies that address six core research challenges: uncovering causal mechanisms, modeling dynamic neurodevelopment, integrating multimodal data, individualized computational modeling, collaborative learning across institutions, and enhancing social training. We highlight the potential of causal inference to clarify gene-environment interactions, spatio-temporal graph neural networks to capture neurodevelopmental heterogeneity, and multimodal fusion for unified representation learning. Digital twin technologies enable personalized brain modeling and neuromodulation optimization, while social brain reverse engineering and federated learning frameworks support computational hypothesis generation and privacy-preserving collaboration, respectively. Large language models further facilitate context-aware social interventions. We also discuss key challenges-including data heterogeneity, interpretability, ethics, and clinical translation-and outline directions for building a more precise, human-centered research paradigm. This review aims to move beyond incremental tool improvements toward reconstructing scientific paradigms, thereby accelerating the effective translation of AI innovations into clinical ASD applications.