This paper outlines a conceptual and architectural framework for integrating Human Digital Twins (HDTs) and Hybrid Human-Artificial Intelligence (HHAI) systems as part of trustworthy, adaptive, and human-centered learning systems. Based on the recent convergence of Artificial Intelligence, Extended Reality and Digital Twin technologies, this paper presents a modular architecture with three layers - data acquisition, processing and analysis, and visualization and interaction - to enable learning systems to be dynamically modelled to capture human behavioral interaction in real-time. We take that architecture one step further to create HDTs that could represent social-presence, emotional expressiveness and adaptive behavior in an educational context. The paper highlights two case studies where the three-tiered architecture was brought to life: (1) Vivian, a spatially aware HDT aiming to have a natural multimodal interaction experience with users in immersive environments, and (2) VirtualTutor, an intelligent virtual human that provides socio-emotional support and personalized instruction. While more work needs to be done to empirically validate this learning model with larger universes of students, the results of both experiments show that HDTs potentially can increase engagement, improve accessibility, and deliver predictive and empathetic educational interventions.