From black box to transparency: Advances in ai-drivenco-creation in architectural design

This thesis examines the integration of artificial intelligence (AI) into architectural design processes and explores the impact of these technologies on creative practices, while also addressing the limitations that arise from the "black box" nature of AI tools. AI technologies, particularly those using machine learning and deep learning algorithms, are increasingly being utilized in architecture, providing new tools for design generation, analysis, and optimization. However, the opaque nature of these technologies can make it challenging for users to understand and control these tools, creating barriers to their effective use as collaborative partners in the design process. As an alternative to this, a "gray box" approach is suggested. The research explores various AI-supported methods for both 2D and 3D production, including tools such as Stable Diffusion, Midjourney, ZoeDepth, GhPython, Blender, ChatGPT, Tripo AI, and Comfy UI. Experimental studies using these tools highlight their strengths, such as visualization and modeling capabilities, while also revealing weaknesses like lack of transparency and limited user interaction. The study provides an assessment of the current state of co-creation between AI and human designers in practice. It emphasizes the need for more interactive and explainable AI systems (XAI) to give architects a better understanding and influence over the outputs of AI tools, thus enabling a more effective co-creation process. x This work advocates for the development of AI tools specifically designed for architectural purposes and suggests that integrating methodologies like XAI could enhance transparency and user engagement, making AI more creatively useful within design workflows. XAI could increase trust by making AI systems' decision-making processes more understandable to architects and could enable AI to serve as a more effective co-creator in design processes. Additionally, the thesis explores the potential for AI to evolve from a passive tool to an active co-creator, where AI not only generates design options but also interacts dynamically with human designers in a feedback-driven process.