Architecture Students on AI – Common Edge

This past fall, I led a graduate design studio at University of Illinois Urbana-Champaign School of Architecture in which students used AI to develop innovative, repeatable, and affordable housing solutions. 

A few things about this studio cohort: None of the students had opted in. They were not part of the competitive lottery system but were, instead, assigned the course by the school. In other words, none entered having a familiarity with, or comfort in using, AI. Most were undergraduate exchange students whose fear was that there would be a need to quickly learn “new software.” 

It turned out there was no need for worry. The 10,000-hour rule doesn’t apply to AI; students were productively and seamlessly using the tools in just 10 hours. AI empowers most students, helping the inexperienced, and their design projects, be far better versions of themselves in less time. It also enabled international undergraduate exchange students in a graduate studio—for whom English, architecture, and AI were at the start of the semester second languages—to perform from the beginning at a high level of competence.

The premise of the AI studio was simple: Chicago needs 126,000 affordable housing units to address its homeless population. While no single project is going to solve this housing crisis, use of AI in the design studio helped students move beyond the design of one building on an urban site to come up with expansive solutions that were innovative, demonstratively affordable, buildable, and repeatable in multiple locations in the city. The goal in using AI was to consider solutions with a potentially greater impact on a larger urban scale than one building could achieve.

In teams of two or three, as an icebreaker, to reduce anxiety and to help students steer clear of preconceived ideas, each commenced with the same deliberately provocative AI prompt: In what ways is a building like an octopus?

Several of the exchange students used AI to translate their prompts to English. Some of the AI responses were expected, related to geometry and appearance. Some were unexpected, having to do with biomimicry and structures that, like octopuses, change their color. It soon became apparent that one comparison eluded AI but not the students. One mentioned tentacles, in which the best buildings reach out, like an octopus, beyond their sites to engage with the surroundings. The seemingly non-sequitur icebreaker prompt proved especially effective, where the idea of reaching out beyond became the design impetus for each team moving forward.

The students’  contributions to the list of shared attributes between buildings and octopuses had another important effect: it helped them see how AI doesn’t rid the designer of agency but, in fact, requires it. The collaboration worked two ways. It soon became apparent that AI became a design collaborator, another team member, helping student teams prototype faster, more imaginatively, and at a more advanced level than they had previously experienced. AI, in other words, served not as new software that they previously feared required mastering, but as a seamless, agreeable, helpful, responsive, suggestive and always-available teammate. 

For their projects, students used AI to learn how to mitigate the negative impacts, including from auto exhaust and noise, of placing housing near highways. AI also made helpful suggestions for how to make housing that is more affordable. Since the most expensive cost of housing is land, several student teams used this information to seek out building sites that were less costly or even, in the case of unused land along a major freeway, free. The goal wasn’t to emulate public housing prototypes but to create innovative design and planning solutions that also happened to be affordable.

Ultimately, over the course of a semester, AI was used for a multitude of formerly time-consuming tasks: to flesh out, refine, and provide potential building envelopes for massing models and rough site sketches; streamline formerly linear and cumbersome design processes; optimize floor plan unit layout variations; simplify building envelopes; determine energy-use and energy-saving strategies; and provide urban environmental performance analysis. Given where students were in their academic and professional journeys, the results were nothing short of remarkable. Here are a few key takeaways from the semester. 

Design teams benefited from AI’s tendency to hallucinate, where AI models inadvertently generate incorrect or misleading results. Instead of thinking of hallucinations as errors or mistakes, as is often reported in the media, students reframed them as unexpected or provocative design suggestions, recognizing that to innovate using AI, hallucination should be seen as a feature, not a bug. One example involved a student team’s predominantly concrete housing development bridging over a highway that was mistaken by AI to be built using mass timber. Instead of stressing, ignoring the mistake, or trying to correct it, the team took this incorrect observation on the part of AI as a provocation, recommendation, or design suggestion, looked into the advantages of using CLT for their project, and, with input from a structures professor, made the design change to great improvement and design impact.

Another positive takeaway for both students and instructors is that AI helps designers start projects in the middle of the design process by providing, not completed building designs, but informed-yet-sloppy first drafts, reducing student’s tendency to “shop” for project ideas online on sites such as Architizer or via Google images. Augmented, amplified, and empowered by AI, students no longer need to search for designs online. Another unexpected benefit of using AI is that it is no longer only good for creating atmosphere but provides helpful input and feedback from the project’s context. While including the specific project street address in the prompt didn’t necessarily improve results, referencing the more general neighborhood, milieu, or site features—for example, building near an urban highway or campus—reaped huge rewards for our students in terms of rich and meaning-laden design suggestions. This capability should only improve over time.

My students used several popular and lesser-known AI tools throughout the semester. While no one AI tool at this point does it all (and some apparently do not even do what they purport to do), students came to realize that there is currently a best AI tool for each task at hand. One tool suggested building envelopes based on photographs of their quickly-built cardboard massing models; another streamlined formerly time-consuming and unwieldy design processes. It was instructive to see students leverage AI to optimize floor plans, analyze urban environments, and develop energy-efficient building strategies, demonstrating that technological literacy will soon be fundamental to not only architectural education but practice itself.

As with any studio, there were a couple of hiccups. The use of multiple AI tools can be costly, with significant out-of-pocket expenditures for both student and professor. More freely available AI image generators can result in elaborate eye-catching 2D graphic designs and 3D biomorphic sculptures, but not buildings per se, and not in specific urban settings. When the tools were first introduced, AI didn’t know an AI-generated image was a building; now it apparently does. Along with recognizing a project’s larger context, this feature ought to be beneficial to anyone planning a large-scale urban planning project that’s responsive to its specific locale and environment.

Another unexpected liability in the use of AI is that it can be frustrating to use as a push tool, trying to get it to perform specific functions and deliver predictable results. Instead, AI is better employed as a pull tool by playing to its strengths; suggestibility, recommendation, metaphoric and lateral thinking.

After a semester of AI use in the studio, it was clear that the benefits of using it for design assignments outweighs any potential concerns as the tool provides all of us with the means to face new challenges, ways to automate time-consuming and repetitive tasks, and simpler ways to address complex operations. AI arrives just in time today for designers to address intractable societal problems at all scales: mass migration, affordable housing, social and racial justice, climate crisis, ethical sourcing of materials, community action, adaptive reuse and historic preservation. All of these can be addressed with the use of AI. We’re going to need all tools at our disposal, including AI, to help address these issues, especially ecological catastrophe. 

To be sure, AI is not yet a science. In one experiment each student was given the same prompt, resulting in 12 different outputs. In another instance, a student would use the same prompt twice, producing a different result each time. AI is not a science because results are not repeatable. If AI is not a science, what is AI research? If you cannot repeat results, what does it mean to be an AI expert in our profession and industry? Related to this, AI results are impacted by the observer effect. As with quantum theory, one’s presence changes the results.
All in all, despite these concerns, the results from a semester leveraging AI in the design studio were nothing short of transformative. Students used it not as a replacement for creativity, but as a collaborative partner that expanded their design possibilities. They weren’t just learning technology; they were solving urban challenges.

Looking ahead, teaching the next generation of design practitioners, our mission is clear. We’re not just teaching students to use AI; we’re teaching them to think critically about technology’s role in solving complex human challenges. The future of architecture isn’t about replacing human creativity with AI, but augmenting and amplifying it. For every designer, planner, educator, and student, use technology as a tool that helps you become a better version of yourself in less time, with more agency.

Images courtesy of Julia Camardans Cirera, Julia Leigh Ganser, Carla Monserrrat Girones Sheehan; Haroon Ahmed, Preston Caleb Ireland.