Bridging the Gap: AI-Driven Generative Design for Sustainable Learning Environments and Behavioral Transformation

Abstract

Urban buildings account for 40% of global energy consumption, yet conventional sustainable design neglects both occupant behavior and the educational potential of the built environment. This study operationalizes the convergence of Artificial Intelligence (AI), Human-Computer Interaction (HCI), and Educational Engineering to address this dual gap. We present EcoDesign-GAN, a generative adversarial network integrated with reinforcement learning, to optimize residential and educational spatial layouts for simultaneous energy efficiency and the facilitation of informal sustainability learning. Using 5,000 urban unit layouts across five global cities for training, and high-fidelity building simulation for evaluation, we examine how intelligent spatial configurations can function as a “hidden curriculum,” influencing simulated energy consumption and fostering environmental literacy through experiential feedback. Simulation results demonstrate that AI-optimized designs have the potential to reduce simulated energy use by up to 25% (95% CI: 22.3%-27.7%, Cohen’s d = 2.8, p < 0.001 ) and improve the adoption of sustainable behaviors and environmental awareness by 40% in controlled virtual environments, compared to matched traditional designs (n = 100 pairs). The Herfindahl-Hirschman Index (HHI = 0.15) confirms high design diversity, suggesting the GAN’s capacity to generate varied learning environments adaptable to diverse cultural and pedagogical contexts. These findings offer a proof-of-concept framework contributing to theoretical advances in Technology-enhanced Informal Learning and Design for Sustainable Behavior, and offer methodological insights for educators, architects, and policymakers.