AI Is the Answer, But What Was the Question?

The title of this review is a slight modification to a mid-1960s lecture by the brilliant architect Cedric Price titled “If technology is the answer, what was the question?” The obvious swapping of technology for AI is suitable today, given how artificial intelligence is seemingly everywhere: Google search results are topped by AI overviews. Email software offers AI assistance. As I write this, nearly every banner on the every web page open on my browser tabs includes some mention of AI. Even a headline in today’s New York Times contends that “Everyone Is Using A.I. for Everything,” while also asking, “Is That Bad?” After years of being hidden inside various technologies, AI is now front and center, making people equally excited and trepidatious: how far will AI advance, and what will it mean to me, to my job? ChatGPT and other chatbots make writers quite nervous, while Midjourney and other diffusion models do the same with architects. Should architects worry about AI’s impact on their jobs, their career choice? This is one of many such questions addressed in Leach’s timely update to Architecture in the Age of Artificial Intelligence.

Leach mentions Price’s rhetorical question on page 176 of his 320-page book, in a chapter that should be of interest to every architect: “The Future of the Architectural Office.” By this point in the book—in the seventh of nine chapters—readers will have realized that Leach is a strong advocate for AI, for technology more broadly, and for the advancement of the profession via the adoption of AI. Thirty years ago, architects were on the hinge between hand drawing and computer-aided drafting (CAD). Ten years later, after the turn of the millennium, building information modeling (BIM) made its move into architectural offices, while Grasshopper and other parametric softwares infiltrated studios in architecture schools. While the trajectory is clear, the embrace of these technologies by now most architects keeps control in the hands of architects. AI, on the other hand, shifts that control to a machine that has been trained on more data than any single architect could ever digest in their lifetime. While the images produced by Midjourney and other generative AI programs are just that, images, often steered by the input of practicing architects, the rapid acceleration of AI technologies should nevertheless give architects trepidation.

Leach fans any flames of fear over AI earlier in the book. In the second chapter, “A Brief History of AI,” one of the advancements he writes about is AlphaGo, a deep learning computer program whose name implies its purpose—playing the game of Go—and its successor, AlphaGo Zero, which taught itself Go without using data from human games, and without even knowing the rules of Go! Much attention has been given to ChatGPT and other chatbots “scraping” data from the internet to learn, but AlphaGo Zero did a similar, yet admittedly narrower, thing completely on its own: it trained itself in three days by playing 4.9 million games of Go against itself. AlphaGo beat Lee Sedol, a professional Go player, in 2016, with the AI player’s moves so unexpected that Lee described AlphaGo as creative—a term often reserved for humans. Three years later he retired from professional Go, stating he could never top an AI player. Again, the writing on the wall is clear: career obsolescence at the hands of an entity that is has the potential to be as creative as it is intelligent.

AI doing things unexpectedly and creatively in the realm of architecture happened a few years ago, not long after DALL-E, a text-to-image machine learning model, was announced, in 2021, and the first edition of Leach’s book was released. Midjourney’s open beta was released in mid-2022, and shortly thereafter photorealistic images of fantastical buildings were populating Instagram. This trend has continued as updates to DALL-E, Midjourney, Stable Diffusion, LookX, and other so-called generative design applications have ironed out some of the quirks and made them more responsive to the needs of architects. As more architects embrace these technologies, some of them are becoming what Randy Deutsch calls “superusers,” such as those within Foster + Partners’ Applied R+D team, Zaha Hadid Architects’ CODE computation and design research group, and Coop Himmelb(l)au’s Deep Himmelb(l)au. These three firms and their respective “superusers” are discussed in Leach’s book—in a chapter titled “AI and Architecture”—but outside of Foster, which is using AI to boost productivity, AI is still being used as a shallow, image-focused technology that is far from architecturally intelligent.

The biggest impediment to AI being intelligent enough to take the jobs of architects is the leap from 2D to 3D. “This remains extremely challenging,” Leach writes, “not simply because the computation power required would be extremely high, but also because there are as yet no tools available for dealing with the complexities of working with neural networks in 3D.” So, for now at least, AI can generate 2D images but not 3D forms, though there are startups focused on architecture and development who could realize such a shift. One is Forma, which was created as Spacemaker in 2016 and then acquired by Autodesk in 2020, and another is XKool, which was founded in 2016 by two architects, one of them, Wanyu He, a former OMA employee (XKool = “ex-Koolhaas,” get it?). Spacemaker/Forma was created to crunch codes and other data for developers, such that AI would become an “invisible assistant” for architects, while XKool was created by architects for architects, so they could “hand over the routine and repetitive work to the computer,” in Leach’s words, “so as to leave the architect more time to devote to the creative process.” In both cases, clients wanting architects to use AI—”to maximize their return on investment,” Leach writes, “and optimize the performance of their buildings”—will push architects into firmly incorporating the technology into their offices.

Any book about AI, and especially one about AI and architecture, is bound to make predictions. Leach grounds his own predictions within those of Toby Walsh, who wrote Machines that Think: The Future of Artificial Intelligence in 2009, and Ray Kurzweil, who published The Age of Intelligent Machines in 1990 and The Age of Spiritual Machines: When Computers Exceed Human Intelligence in 2009. The latter’s predictions “about technology in general and AI in particular,” Leach writes, “have proved to be remarkably accurate.” While Kurzweil was spot-on about “wireless networks [allowing] the easy sharing of courseware [...] and other creations,” his misses are laughable, as in long distance travel by 2009 being made using ”cybernetic chauffeurs.” Leach’s own predictions balance the practical concerns of architects and the aspirations of an AI booster. Will AI “become an indispensable assistant”? Most likely. Will AI “be able to design a building completely autonomously”? Maybe … but why? Should buildings for people be designed by machines? Or to put it another way, if AI is the answer, are we asking the right questions?