Evolution of a Section: Architecture and Machine Learning

3 units

Throughout human history, architecture was seen as static, a quality attributed to its inherent physicality. This seminar encourages students to conceive of an architecture, through the medium of an architectural section, that mutates across space and time. Using Machine Learning processes, the class intends to propose an alternative and nonlinear means of production to the linear process of architectural design from conception to construction. Machine learning engages graphic information differently than designers do. All fidelity towards visual, cultural, political, and geometrical context is lost, resulting in a new class of compositions that are unique but not critical. The systems, including Generative Adversarial Networks, Convolutional Neural Networks, and Diffusion Models, are explored with input (images/texts) and analyzed as output images. We will collectively conjecture on how to 'train' the AI models to understand spatial features typical of an architectural section. We will rely on the rich history of architectural sections, across time, styles, and media, to inform the potential trajectories that our section follow. The works of contemporary artists and architects, like Matias Del Campo, Gabriel Esquivel, Helena Sarin, Refik Anadol, etc. who work with Machine Learning technologies, will be analyzed to understand approaches towards AI and Design.