A Multi-Scale AI Framework for Informal STEM Learning: Paramorphic Digital Twins for Underserved Communities

Abstract

This research presents a novel multi-scale paramorphic kernel learning framework (MPKLA) that is designed to enable autonomous, context-adaptive STEM learning for electrical and renewable energy topics. By employing distributed multi-agent kernel cores, recursive kernel reweighting, and entropy-guided abstraction modulation, this system dynamically adapts instructional content and sequencing based on the specific cognitive state of individual learners. Concept learning history is maintained in persistent memory buffers to facilitate individualized reinforcement and remediation in asynchronous, informal environments. Grounded in a physics-informed knowledge graph, the system offers epistemic coherence and domain consistency at multiple levels of abstraction. Evaluated across multiple, underrepresented student groups in community and laboratory implementations, MPKLA demonstrated a 45% rate increase in concept recall, a 3.2× improvement in student-led project completion, and sustained 68% learner interest over 12 weeks. These results emphasize the effectiveness of this architecture in delivering scalable, culturally sensitive, and high-fidelity STEM education without human interaction. The paper also discusses system deployment, statistical validation, and longitudinal deployment settings, informal education problem-solving, cultural adaptation, and learning assessment. MPKLA provides an extensible blueprint for inclusive, technology-driven workforce development in clean energy sectors, fueling inclusive participation and expertise in advanced technical fields.