A machine designs new Swiss Alpine architecture at the Seoul Biennale
“Our design brain is an experiment about whether machines can design structures,” says Professor Jeffrey Huang, head of the LDM. “That is, whether they cannot just recommend music or drive cars, but also create meaningful cultural artifacts, such as architecture with distinct Swiss characteristics.” Artificial Swissness will be on display at the Seoul Biennale for Architecture and Urbanism until 31 October. It was developed by LDM in association with Convergeo, a design agency co-founded by Prof. Huang, and Seoul-based architecture firm SPOA.
Infinite architectural images
The exhibit is intended to be a continually shifting spatial interface that reveals the inner thoughts of an AI machine trained on 10,000 images of Swiss chalets and Alpine architecture. It takes the form of a large, highly active, artificial brain that captivates viewers through a palimpsest of architectural imagery. The radial form of the artificial brain gives viewers a sense of the visual cortex of the deep convolutional neural networks as the machine generates a virtually infinite number of architectural images of Swiss Alpine cabins.
Frederick Kim and Mikhael Johanes, two PhD students at LDM – who had to quarantine for 14 days in order to be on site and set up the exhibit – describe how the design brain works: “We make explicit the visual inferences in these convolutional neural network layers, progressing from a very low resolution to the final resolution. This provides a glimpse into the inner workings of our generative artificial network, or GAN, as it creates architectural images.”
A constantly evolving learning process
The exhibit consists of multiple layers of displays arranged in a circular format and composed of 30 high-resolution, high-luminosity digital screens powered by Raspberry Pi mini-computers. The screens’ angles, height and depth are designed to give viewers a varied and personalized field of vision. The neural network’s thinking process is depicted through a programmed LED projection of the GAN’s convolutional layers. The digital screens and LED projection run simultaneously so as to engage viewers emotionally (viscerally) and intellectually (cerebrally). The digital screens show the machine-generated images of typically Swiss architecture (the essence of Alpine architectural culture), and the LED projection reveals the constantly evolving learning process that the machines go through as they sort through thousands of images of Alpine architecture in an effort to distill the essence of “Swissness.”
LDM is run jointly by EPFL’s School of Architecture, Civil and Environmental Engineering and School of Computer and Communication Sciences. It studies the effects of digitalization on architecture and modern cities. Artificial Swissness aims to expand the epistemology of computational science into the cultural domain, demonstrating specific computational-based perspectives of architectural and artistic culture.