ARCHITECT AS CURATOR: NAVIGATING INTELLIGENCE DATA, AND DESIGN

This thesis investigates the integration of generative artificial intelligence (AI) into architectural design, specifically exploring its potential to augment the conceptual phase of the design process. Through a combination of historical contextualization, critical analysis, and design experimentation, the research examines the evolving role of the architect in an AI-mediated workflow, shifting from sole creator to strategic curator. The study begins by situating AI's emergence within the broader genealogy of automation in architecture, tracing key theoretical discourses from early cybernetics to contemporary Generative Adversarial Networks (GANs) and diffusion models. It critically examines the affordances and constraints of image-based generative models, evaluating their relevance to established architectural workflows. A central tenet of the research is the acknowledgment of AI as an active participant in the design process, characterized by inherent biases, aesthetic tendencies, and limitations. To evaluate the architectural relevance of AI-generated imagery, the thesis undertakes a design experiment centered on a competition proposal for a crafts museum in Akita, Japan. Employing Midjourney, a diffusion-based image generation platform, the research establishes a detailed workflow that balances algorithmic autonomy with human design intent. This workflow encompasses iterative prompting, selective image evaluation, and comparative assessment, emphasizing spatial consistency and material logic. The research reveals that generative AI, while capable of producing visually compelling imagery, necessitates a critical curatorial approach from the architect. The translation of AI-generated images into coherent architectural drawings and 3D models requires significant interpretive labor to address issues of scale ambiguity, image fidelity, and spatial reasoning. The thesis argues that the architect's role is to strategically navigate the generative process, transforming AI-generated provocations into contextually relevant architectural designs. Ultimately, this research identifies the procedural and epistemological gaps in the current generative AI design workflow. It proposes that with more structured prompt systems, better 3D integration, and improved dataset curation, generative AI could evolve from a visual provocation tool into a more robust design assistant—potentially enabling the simultaneous generation of spatial plans, sections, elevations, and 3D massing. The thesis concludes by emphasizing the need for critical reflection on the ethical and social implications of AI in architecture, urging the field to prioritize meaningful integration over superficial automation.