On February 3, the University of Texas at Austin School of Architecture hosted a symposium exploring the potential for artificial intelligence (AI) to revolutionize the field of architecture. The “Architecture After AI” symposium complemented an exhibition of the same name, curated by Associate Professor Kory Bieg, Assistant Professor Daniel Koehler, and Associate Professor Clay Odom. The exhibition ran from January 30 to February 24 and featured more than 2,000 images created by more than 60 designers. UTSOA welcomed six participants from across the country to speak at the one-day symposium: Andrew Kudless, Jenny Sabin, Ed Keller, Kyle Steinfeld, Marta Nowak, AIA, and Sofia Crespo. Each guest gave an individual presentation, then participated in one of two panels moderated by UTSOA faculty members.
After some opening remarks from the curators of the exhibition, Andrew Kudless — founder of Matsys Design and professor at the University of Houston — presented a lecture titled “Hierarchies of Bias: The Weight of Our Tools.” Kudless began by examining biases towards certain resolutions, languages, and geographic stereotypes in multiple versions of Stable Diffusion and Midjourney. He also experimented with the ability of CLIP (Contrastive Language-Image Pre-Training) to understand and interpret a description of Confluence Park in San Antonio, finding that the program quickly drifted back toward the stock images on which it was trained. However, Kudless believes that there are benefits to using AI in architecture, and he employs these programs for sketching in his own work. “I think the value of generative AI images is in their ability to produce ambiguity,” he says, explaining that he views AI as a valuable conceptual tool.
Jenny Sabin, principal and founder of Jenny Sabin Studio and associate professor at Cornell, opened her talk by recognizing the power of AI to radically change existing workflows. Sabin also acknowledged the potential ethical issues raised by this new technology’s lack of diverse users and data. Her lecture, “Biosynthetic Architecture: Driven by Humans, Powered by AI,” highlighted the importance of transdisciplinary design. Sabin’s project “Ada,” named for mathematician Ada Lovelace, is a hybrid of analog and digital design processes created in partnership with Microsoft Research. Sabin explained how Ada’s network of 3D-printed nodes and Raspberry Pi microcontrollers allows the project to “smile back at you” as it anonymously reads facial expressions and adjusts the color of embedded LEDs accordingly. Direct engagement with Ada “augments emotion through aesthetic experience,” says Sabin. Her work emphasizes the potential for AI to increase user well-being through responsive design, while continuing to acknowledge the cultural biases present in our current technologies.
In his talk “Distributed Cognition: AI & Design as a Geo/Cosmopolitical Project,” AUM Studio and Spec.AE co-founder and principal Ed Keller took a more rhapsodic approach to AI. Keller discussed his recent work in exploring concepts of non-human cognition and communication, attempting to synthesize feelings of loss and love in collaboration with AI. He indicated that there is a disconnect between how humans and AI communicate because there is no computational equivalent for physical gestures or speech tones. To bridge this gap in symbolic understanding, Keller believes that “we need to look at models of speaking with the alien.” In his presentation, he cited the film “Arrival” as an example of communicating with the alien mind through pseudo-language. Keller views AI as an intelligent, information-rich partner in the design process — almost like having another person in the room with you.
In a panel moderated by Assistant Professor Ria Bravo, Associate Professor Kory Bieg, and Lecturer Rasa Navasaityte, the first three speakers addressed larger questions about the impacts of AI on current design methods. Parsing out truth from fiction in AI-generated products can be a challenge. “Truth, at least in my opinion, is present in what is absent,” states Sabin, referring to the ways in which AI reflects our cultural biases back at us. The panel discussed the need for truthful reporting and accountability regarding the energy used by these AI programs, but that information is not yet forthcoming. Keller then initiated a conversation about the pros and cons of big data versus open-source AI programs, noting that both are problematic models with their own biases. The panel concluded with a look into the future of the architectural field. Kudless explained why he believes that AI will not result in the loss of traditional design practices: “The invention of photography obviously didn’t kill painting, it just changed what painters did.” The speakers agreed that there is an opportunity for interdisciplinary expertise to evolve in this field over time as processes of generation and selection become more complex.
Following the first panel discussion, UC Berkeley Associate Professor Kyle Steinfeld offered a retrospective on the relationship between humans and machine intelligence. His “Architecture After AI” lecture focused, not on the future of design, but on what has brought us to where we are now. Steinfeld positioned himself as an advocate for augmentation, not automation; he views AI as a cultural technology with opportunities for humanistic application. Steinfeld suggests that AI could make prescriptive CAD technologies more human; instead of teaching designers to think like machines, AI is attempting to teach machines to think like people. He pointed out that while new AI tools may feel alien, they are rooted in a technical history dating back nearly a century. Steinfeld closed by asking a series of open-ended questions addressing common AI anxieties: “Might AI wreck existing cultures in design? Might AI produce inequitable shifts in power? Might AI devalue certain aspects of the craft of design?”
In her “Intelligent Workplace” presentation, Marta Nowak, founding principal of AN.ONYMOUS and assistant professor at Ohio State University, discussed her attempts to rethink architectural components as intelligent, interactive objects. She shared a few projects that were realized through her work with Google R+D and centered around the creation of team-based collaborative workspaces using AI. Nowak also detailed the prototyping process for an autonomous floating partition that uses machine vision AI to navigate through occupied office spaces, developed in collaboration with Ohio State University. “What I’m interested in,” Nowak says, “is to give artificial intelligence to objects, architectural components, and buildings and allow them to continually interact and respond to their human users.” She is inspired by the potential to create dynamic, reactive interiors through AI.
Finally, Sofia Crespo, artist and co-founder of Entangled Others Studio, spoke about the importance of data in AI during her talk “Expressing Entanglement: Art and the More-Than-Human World.” Crespo emphasized the importance of demystifying AI, arguing that it is simply “a tool that allows us to extract patterns and rearrange them.” Her methodology is based on a cycle of data input, neural network processing, and explorative output. Crespo discussed how her most recent project highlights issues of data bias in endangered species conservation; she trained a neural network using a dataset of 10,000 species then used the resulting tree of life to attempt to map out the nearest neighbors of critically endangered species. Certain species — specifically those with faces, which are easier for humans to understand and interact with — had larger conservation efforts and more data to train the AI on. This imbalance was reflected in the project’s outcomes. Crespo’s recent work has thus been shaped specifically by the lack of certain data, drawing attention to this pervasive issue in current AI technologies.
The second panel discussion, moderated by UTSOA Dean Michelle Addington, Associate Professor Clay Odom, and Assistant Professor Daniel Koehler, revolved around the dual concepts of data and evolution. The speakers discussed why architecture doesn’t have a shared database, as many other fields do, with Steinfeld suggesting that the built environment is an architectural database in and of itself. Crespo reiterated how the limitations of available datasets for AI programs affect their outcomes, recognizing that all AI work contains human bias due to the nature of visual selection methods. These selection processes can be viewed as a kind of evolution; Nowak described how certain ideas “die along the way” as concepts continue to develop. All three participants agreed that the current rise in AI is part of a larger, ongoing evolution of technology and artistic thought. “Art has always been in motion,” explains Crespo.
The symposium concluded with closing remarks from Dean Addington, who encouraged designers not to box this new technology in. Instead, she suggested that architects push the boundaries of what is possible with AI, allowing it to help them ask questions that were never up for consideration before. “It should be one of the most revolutionary things that could happen to us in architecture,” she says. “It does require that we go ahead and take it as far as we can take it, and let somebody else worry about limiting it.”
A full recording of the “Architecture After AI” symposium is available for viewing on the UTSOA YouTube channel.
Abigail Thomas is an undergraduate student at the University of Texas at Austin School of Architecture.