AI-mediated socio-emotional support for children with autism:a human-supervised framework

Artificial Intelligence (AI) is progressively reshaping the field of inclusive educa-tion by offering new tools to support structured and personalized learning processes for students with autism spectrum disorder (ASD). This contribution introduces EMORI, a conceptual and architectural framework for the use of artificial intelligence as a mediated educational support in socio-emotional interventions for children with autism. The project is implemented through EmoRobot, an embodied robotic agent, and supported by Socrate, an AI orchestration platform designed to manage conversational logic through pedagogically guided retrieval-augmented generation techniques.The system is grounded in the so-cio-emotional intervention program proposed by Patricia Howlin and adopts a state-basedconversational model that structures interaction into progressive and predictable stages. Within each state, the language model operates under constrained generative conditions, providing adaptive prompts and feedback while preserving instructional control. A central feature of the framework is the adoption of a human-in-the-loop approach, ensuring con-tinuous pedagogical supervision.The contribution of this work lies in the definition of a theoretical and methodological model for the pedagogically sustainable integration of artifi-cial intelligence into socio-emotional interventions for children with autism.