Artificial intelligence is hot in the architectural news space. I seem to be getting at least an e-mail a day from PR firms advertising some artist or architect “subverting” or “reimagining” the world through the lens of programs like Midjourney or DALL-E. The AI tag on design news aggregator designboom (always a reliable trend bellwether) encompasses everything from lazy listicles about what AI thinks supercars would look like if designed by famous architects to claims that XYZ designer’s AI images of buildings imagine the future of architecture. The virus has since spread to trade publications like ArchDaily and Architizer. While much of this content may be pablum, the implications of the technology involved are not unremarkable.
As a testament to the AI hype’s reaching the highest echelons of the field, in February, architect Thom Mayne attended a media event at the Center for Architecture in Manhattan, where he claimed that architecture was too obsessed with “looks” and that it should instead embrace the potential complexities introduced by artificial intelligence. When asked to elaborate, he replied, in typical archi-speak, “There needs to be the advancement of something much more complex, as architects take responsibility for shaping the world.” Whatever. Yet, amid all his opacity, Mayne, who runs the bicoastal firm Morphosis, did end up revealing some of the actual motivation behind why firms are increasingly keen on AI: It will reduce the number of individual architects in a given studio to what he calls a more “intimate” level. Those blobby Instagram-fodder non-buildings suddenly feel a bit more threatening.
Let’s back up. It is a myth that starchitects like Mayne are responsible for the creation and execution of a building—much less “the world”—and that architecture is the purview of a few geniuses who turn napkin sketches into reality. The truth is that firms like Morphosis run thanks to the labor of dozens (and in larger firms, hundreds or thousands) of employees whose responsibilities range from construction documentation and computer-aided drafting to project management. What most people don’t know is that many aspects of architecture are not handled by architects at all and are in fact handed off to outside contractors, specialists, and engineers. When looking at the possibilities for AI’s becoming a labor disruptor in architecture, it is first necessary to understand just how deeply stratified the division of labor is in architecture as a field. Despite its reputation as a creative enterprise or as an “art,” the vast majority of practicing architects are not reified designers but workers putting in unholy hours at firms that look less like the romantic drafting-desk images from the movies and more like the depressing office environs of any other white-collar job.
To understand which parts of architecture AI is capable of displacing, one must keep in mind two things: the technical capabilities of the software and the less-flattering truths about architectural production. There are some AI startups like Swapp promoting the potential uses of AI in construction documentation (the assembling and management of all the diagrams and specifications of a building to be handed off to builders). In an Architizer editorial, Swapp claims that architects will be liberated from the drudgery of this work (and firms from the threat of worker unionization) through AI. This kind of rhetoric has its roots in earlier promises that computer-aided design (CAD) software would give architects more opportunities to pursue forms and ideas, when, in reality, CAD has increased workloads by allowing firms to take on more projects at once—contributing to the burnout AI promises to solve.
However, I believe that while design and construction documentation is the most boring part of making a building, it is also most likely to be the last to be automated, for the simple reason that architects and firms are extremely risk-averse. Firms are terrified of lawsuits to the point where it stifles aesthetic innovation—it is much less risky to use materials that are well-cataloged, documented, and tested than something new. Architects would generally rather leave such features such as retaining walls to contractors rather than elaborate on specifics that would make the architect themselves legally responsible. This is, simply, not a practice open to “disruption.”
When ordinary people think of what happens in an architecture office, they imagine the visual communication most often presented to them: the rendering. A rendering is an image shown to clients and the public demonstrating what a building will look like when built. If plans and sections are esoteric diagrams that are not considered aesthetically beautiful to anyone but architects, the rendering is how architecture portrays itself—and the architectural ideal—to the world. For most of architectural history, architecture firms made their own renderings. An architect’s renderings are sometimes even elevated to the status of art, often displayed in museums. But even this most intimate architectural expression has now been outsourced to outside firms.
The same can also be said for public relations and marketing, which also used to be in-house. Now these tasks, which used to be seen as the written, public-facing expression of a firm’s goals, have been shipped out to PR firms. The reality of the modern architecture firm is that it makes more renderings, more proposals, more press releases than it does buildings. It’s part of how firms compete against one another for relevance and eyeballs when it comes time to submit bids for actual competitions or projects. This has been the case since the dawn of mass media; it’s merely accelerated in the social media age, where everything is hype-driven and supposedly world-changing.
If you look at what AI is presently capable of doing, it’s not hard to see where I’m going with this. I type “hyperrealistic rendering of a modernist apartment block” into DALL-E and it spits out four images that look like completely plausible mid-rise apartments. I type “press release for a new luxury apartment building in Chicago” into Chat-GPT and it gives me something absolutely usable. None of what has been generated, however, is particularly creative, inspiring, or novel. Why would it be? AI, after all, gets its source material from human creativity as distributed across the entire Internet.
When most new-build apartment buildings look a certain way, AI will pop out an apartment building that best matches what machine learning has crawled through the Internet and assigned as an apartment building. It is misrepresented as some inherently revolutionary technology, when in fact it is rather conservative aesthetically. The image generator Midjourney creates images that look a certain way—digitally painted, pastel-colored, clay-like in texture—because a great deal of its source material comes from digital art sites like DeviantArt, which itself trends toward those aesthetic features as users compete for popularity. Not only is this a gray area in terms of intellectual property; it also is, well, boring.
An architecture dependent on AI will produce content of the lowest common denominator. Architecture, for better or for worse, has always offered commentary on what the future of the built environment should look like. It provides a service to the world. AI does not exist in the spatial realm perceived by ears, noses, hands, mouths, and eyes. It cannot spatially understand a site or its context; it cannot understand the history and politics of building because these are value judgments requiring human critical acumen. All it can do is look at what’s on the Internet and spit out images and D-tier text. That’s not the future of architecture. If anything, it’s the death of everything that makes architecture interesting.