Spatial Learning for Functional Hybridisation - Critical methodology for generative architectural suggestions in concept design phase

The research aims to define a methodology to support the creative process in the early architectural design phase through Artificial Intelligence (AI), focusing on spatial learning and generative typology hybridization. It specifically investigates how AI can interpret and suggests internal spatial configurations of buildings by leveraging Generative Adversarial Networks (GANs). In recent years, the rise of AI-generated art has challenged traditional notions of creativity, and its impact is now increasingly visible in architectural design processes. The proposed approach integrates AI into design by exploring the latent space of StyleGAN, training AI on building sections' data and reconstructing 3D forms through point clouds. This offers a hybrid workflow between AI and computational modelling to generate three-dimensional architectural forms. As a design tool for architects, this workflow enhances spatial understanding of a building, analysing and exploring spaces of different typologies, looking for potential spatial hybridisation in design. The methodology is grounded in a critical reassessment of AI's role in design creativity, not as a replacement for human intuition but as a collaborator capable of producing unexpected spatial interpolations. The study highlights how hallucination, often considered a flaw in generative AI, can become a productive tool for design speculation, enabling new forms of typological hybridisation. This research contributes to redefining the epistemology of architectural form-making in the age of intelligent systems, opening a path toward collaborative creativity between human designers and artificial agents.