With recent innovations in artificial intelligence (AI), machines can now recognize architectural patterns and generate new forms in ways that far exceed human capacity. The integration of machines in the design process has ushered in a substantial shift in architecture discourse and design, whereby traditional â form-givingâ and â form-findingâ have given way to â form-searchingâ . State-of-the-art technologies, particularly deep machine learning, now offer architects more advanced tools and methods for analyzing and generating architectural forms. This research explores this ground-breaking â transition of form-makingâ within the discipline of architecture, examining how AI can understand the innate complexities of architectural form and is fast becoming a powerful co-design agent in its own right.

The reconceptualization of form as a spatial token embedded with a multitude of data and information can speak to any such entity's tangible and intangible aspects: inherent formal qualities, intrinsic building information, deep-rooted cultural and environmental knowledge, unexplored semantic meanings, and unfamiliar morphological characteristics. Achieving this multidimensional comprehension of form as a data-rich spatial entity requires a framework that can integrate the computational capabilities of machines with the creative mindset of architects. This symbiosis in turn, requires the purposeful merger of machine learning techniques with conventional architectural methods and knowledge. To this end, this research introduces a comprehensive machine-learning framework encompassing the multimodal stages of form collecting, form learning, form decoding, and form generating, which together promote a dynamic interaction between human architects and intelligent machines.

The thesis investigates four dimensions of architectural formâ figurative, perspectival, semantic, and diagrammatic. The figurative dimension expands the tangible and geometric properties, emphasizing the formal and volumetric attributes of architectural entities. The perspectival dimension addresses the transformation of two-dimensional representations into three-dimensional architectural forms, highlighting the intersection of human and computational perceptual capabilities. The semantic dimension utilizes formal language-based generative AI to develop a nuanced formal vocabulary and embed semantic richness into architecture. Lastly, the diagrammatic dimension explores the use of parametric computational tools in combination with graph-learning to model and visualize complex flows within the built environment as a novel means to innovate architecture design.

Advanced machine learning techniques are leveraged to advance our understanding of architectural forms by uncovering intricate formal patterns and relationships that can inform innovative and contextually responsive design solutions. By critically and creatively embracing AI and deep learning, architects can go beyond inherited form-making techniques to engage more complex contemporary environmental, cultural, and social challenges with exceptional data-driven insights. The convergence of human creativity and machine intelligence paves the way for a novel design process, where human-driven methodologies are integrated with advanced computational tools to expand future possibilities of adaptive, future-viable architecture.