GANism: When AI Rethinks the Borders of Architecture. The Cases of ArchiGAN and Deep Himmelb(l)au.

Abstract

This paper aims to shed light on the challenges and limitations of GANism in contemporary architectural design. It focuses on two specific experiments carried out using different generative adversarial networks (GANs): ArchiGAN by Stanislas Chaillou, on the one hand, and Deep Himmelb(l)au by Coop Himmelb(l)au, on the other.This research is based on an in-depth literature review on the use of artificial intelligence in architecture today, at the dawn of the digital era. It also includes a visual analysis of photographic and videographic documents drawn from conferences and academic publications.ArchiGAN (2019), developed within the framework of a master’s program at the Harvard Graduate School of Design, is an AI tool based on the Pix2Pix model, enabling the generation of various interior design configurations. It proposes a dynamic distribution of rooms, partitions, and structural elements (columns, load-bearing walls, etc.), thus adapting the space to multiple dwelling scenarios. By contrast, Deep Himmelb(l)au (2019), an experimental initiative conducted within the Coop Himmelb(l)au office, explores the acceleration of the architectural design process by employing a variety of GANs to reinterpret some of the firm’s emblematic projects.How does GANism influence the creative process in these two projects? What kinds of spatiality, functionality, and aesthetics emerge from these computational approaches? What future awaits architecture as a discipline in the face of the growing power of generative tools? What lessons and perspectives do these experiments open up for contemporary architectural and artistic creation?.