Sensorial Space in the Age of A.I.

Mary Greenert, Fa Zhe Ren, Anh-Thu Vuong

A{AI} - Architecture and Artificial Intelligence


Matias del Campo + Sandra Manninger


Our relationship to nature, though ever-changing, has always existed. Humans, though a product of nature, redefined the natural environment into artificiality by creating the built world. Starting with the replication of natural space, we curated and produced our own interpretations of the natural world through the mediation of material objects. We created our own language of spatial relationships, building upon it over thousands of years and reformulating our relationship to what we consider to be nature.

According to theorist Quatremère de Quincy, architecture is interested in “not nature's concrete phenomena but the uniformity (Gesetzlichkeit) and rules by which she exists and creates.” Are these rules the beauty that make nature so appealing to human eyes that our built environment is a response to it? This project is interested in how Artificial Intelligence takes part within this relationship and how it can help us better understand our sensorial surroundings. Embedded within Chicago's underground infrastructure, the project reoccupies the abandoned freight tunnel system and connects with the Chicago Pedway network. Using the results of a cyclic Generative Adversarial Network (CycleGAN) based on a 4,000 image database of elements found in nature and architectural spaces, the tunnels become occupied with patterns and textures aiming to provide for a more dynamic, and perhaps intrinsic, experience.

It is our assertation that we can make use of artificial intelligence by asking the following questions; If we were to think of Artificial Intelligence as a pure creative mind, knowing only the natural world, what type of environment would it design? How would it interpret these new environments into architectural space? Neural Networks have the unique ability to interpret and create data without external bias, allowing us to visualize artificial environments, facilitating the creation of a new sensorial context in which we can exist.

Dataset of 4,000 images (partially shown)

CycleGAN Input and Output Matrix

View of Tunnel Interior with Patterns taken from CycleGAN Testing Results