How to Stop Worrying and Love AI
The architecture profession has much to gain from the evolution of artificial intelligence. Ultimately, the future of AI will drive improved collaboration, creativity, and critical thinking in both construction and design.
Artificial intelligence will increase “specific opportunities for architects to leverage data and analysis toward more robust and predictable results. Rather than a threat, maybe this is an opportunity to offer more valuable services to those clients and keep the robots at bay,” writes Phil Bernstein, associate dean at the Yale School of Architecture and author of Machine Learning: Architecture in the Age of Artificial Intelligence. So, what does artificial intelligence (AI) signify for our profession, and how will it impact the creation of buildings and cities?
AI has been part of our lives through algorithms and computer programs for decades. We are already familiar with Facebook, Alexa and Siri. We have AI algorithms to find our partners, suggest Netflix shows to watch or make book and clothing suggestions. But the release of DALL-E and ChatGPT in the fall of 2022 cracked open a new discussion about creativity, the authenticity of thought and the future viability of a profession seemingly constantly in a state of existential crisis. While we envision a typical architectural office to include more computer programmers, those offices will enable junior staff to spend more time managing projects and building relationships with the construction industry than collecting design precedents, optimizing material selection, or verifying building code compliance with Passive House standards. AI will make for more compelling client presentations and enable designs to quickly develop feasibility studies or energy modelling while liberating the critical skills of more senior practitioners.
Increasingly, architecture offices are incorporating AI-based processes. Global firms like Zaha Hadid Architects (ZHA) are taking on AI-generated images with vigour where firm principal Patrik Schumacher accepts text-to-image results into his firm’s body of work, noting in a recent roundtable that “Any of what comes out of this, I claim authorship for it in terms of validating, selecting, elaborating,” adding, “For me, it’s always been very similar to verbal-prompting teams, referencing prior projects and ideas and gesticulating with my hands.”
In an ever more litigious world, architects must ensure that AI will not “hallucinate,” a term popularized in 2022 when specific large language models (LLMs) such as ChatGPT began to create plausible-sounding falsehoods. After all, AI only makes sense of a world based on the data it draws from. Neil Leach points out in his Architecture in the Age of Artificial Intelligence that similar hallucinations can occur when we “Ask any architect what a ‘functionalist’ building looks like, [and] chances are they would describe a white building on pilotis with a flat roof, even though flat roofs are not very functional in most countries because they tend to leak.”
Leach makes clear that architects are conditioned to see the world in a certain way, and this falls into the theory of “predictive perception” a way of seeing that does not significantly differ from how artificial neural networks are trained. Leach refers to this phenomenon as “architecturalizations,” a process “where architects tend to ‘architecturalize’ whatever they see and read the world in architectural terms.” This process describes Jørn Utzon’s inspiration for the Sydney Opera House after observing the billowing sails of yachts in the harbour, or architects’ near-clichéd misinterpretations of Gilles Deleuze’s philosophical concept of the “fold.” As Leach ruefully declares, “Whether we understand this conditioned outlook…as a form of ‘controlled hallucination’…or as being constituted through ‘fantasy,’ it is clear that architects see the world not as it is but as they are trained to see it.” So then, should we be afraid of AI if it already processes information a bit too much like an architect?
The laws surrounding copyright, intellectual property and legal liabilities surrounding AI-generated solutions in architecture still need clarification, even in heavily manipulated imagery.
Automation of Design Progressions
Expect to see all kinds of software innovations triggered by the expansion of AI. Autodesk is now beginning to integrate AI into their software, like Project Dreamcatcher, a generative design platform where designers can input performance criteria, cost restrictions, and other constraints. Adobe PhotoShop’s content-aware fill is no longer sufficient for market demands, and it is beta-testing its AI-generated fill powered by Adobe Firefly to insert more creative backgrounds into designers’ image files. According to Adobe’s website, it is “a family of generative Al models designed to be safe for commercial use and trained on Adobe Stock imagery, openly licensed work, and public domain content where the copyright has expired. Content generated by Adobe Firefly in the Photoshop (beta) app is not permitted for commercial use.” The language here is worth commenting on because the laws surrounding copyright, intellectual property and legal liabilities surrounding AI-generated solutions in architecture still need clarification, even in heavily manipulated imagery.
Ambitious software startups like Xkool are creating new opportunities for less ambitious designers when it comes to writing their AI-related code, which is becoming simpler to achieve. Architecture offices are beginning to develop their proprietary plugins using programs like Grasshopper or Rhino to help them measure building performance criteria. Or they can also use Rhinovault as an AI-like algorithmic design technique. Ameba is yet another program using a Rhino-Grasshopper platform which can apply different loading and boundary conditions to an initial design, then evolve into various shapes to achieve structurally efficient organic forms.
But software tools are just that: tools. It’s never the technology but how humans use it. We’ve witnessed this before when CAD replaced drafting boards, and BIM began to shift how we re-conceptualize the construction process.
The world of construction is undoubtedly changing too. Toronto-based Promise Robotics is using AI to advance robotic fabrication technologies and improve the cost and efficiency of building construction. It’s adopting dumb robotics from the automotive sector with additional sensors and cameras to engage in dynamic construction capabilities while attracting financial support from the federal government and venture capitalists interested in developing product and service innovations across the construction sector.
In broader technology-based sectors like ClimateTech, GreenTech, and ConTech, incubators across the country like Foresight Canada and MaRS – and indeed worldwide – are helping entrepreneurs adopt AI to make better buildings faster and more efficient. Architects should be well-advised to capture these human and financial capital investments.
Another area of AI influence is in the evolution of digital twins, which continue to gain popularity on both the level of an individual building or a city. A digital twin effectively allows a performance simulator to occur and is constantly being updated in real-time using sensors in building columns, roads, or objects connected to the Internet of Things (IoT) to anticipate everything from pothole repairs to flooding. To this end, construction and construction-related companies like Pillar have been pushing for designers to anticipate the need to install sensors in their buildings to help collect data. Their investors include insurance companies wanting buildings to be more intelligent and dialled into better generative design decisions while anticipating more robust AI-related monitoring.
Found in Translation
We should consider AI as a continuum of how we are already conceptualizing our buildings. Terence Tourangeau, director of Digital Practice at SvN Architects + Planners, graduated from Carleton University roughly 15 years ago, where collecting precedents was not part of his design pedagogy. In contrast, other architecture schools used precedent study as a core teaching tool. In most offices today, collecting precedents for a client presentation is standard practice where the junior staffer assembles images of projects, layouts or materials from built work around the world. AI could instantly achieve this process, drawing from millions of images online. However, “An image of a precedent is never as good as having experienced a space that you’ve kept in your memory bank,” says Tourangeau, adding, “Cobbling together images creates a collage that can miss a lot of variables. But AI lets you become the editor.” AI design tools can quicken the design process, notes Tourangeau, creating elements for your renderings like trees or usable silhouettes of people.
When it comes to those fantastic and instantly created renderings from Midjourney or DALL-E, let’s consider the potential falsehoods and manipulations of the rendered image. Many firms can produce photorealistic renders in-house, while others gladly farm out the job to specialized renderers living in other time zones. We are already creating imagery using text prompts through frustrating back-and-forth emails with contract renderers attempting to translate design intentions. How is this process dramatically different than prompting an AI-fuelled image generator? To this point, technology and design are about translation. Tourangeau cites his old mentor and teacher, Marco Frascari, who reminded him, “The best translators are not bilingual but bicultural where the ideas go beyond language.” We can similarly include architectural photography in our toolbox, where a project’s merit is either gained or lost in translation by visually narrating (i.e., marketing) its purpose to the outside world.
Mark Cichy, principal and director of Design Technology at HOK in Toronto, is working on a lot of internal testing at his firm to fully grasp just how abrupt and significant a change AI will bring to the practice of architecture. He’s optimistic about the potential for AI to help senior practitioners and junior architects alike. Cichy is also very much aware of how intellectual property developed within a single office might get lost or absorbed into the LLM-based world of AI. He also doesn’t see how loading an HOK-produced sketch into a trained AI model is significantly different in terms of previous design processes, only much faster than manual drawing, preparing concept boards or precedent decks. “It will be faster, like 50 times faster and will create new variables and design relationships. This is one of those prolific moments in history,” says Cichy. “We have 25 to 30 years before AI is fully autonomous, but the creatives stand to gain the most. And the older practitioners have the experience to make them the smartest and most capable to tweak the process.”
Not to be dismissive of individual design talent, but most firms can produce relatively good designs. While AI might help lift all boats regarding design acumen, the true winners will be the firms that manage design well. And this includes finding a better way to work in groups. After all, Cichy notes, “People hang around for a long time. Critical thought and understanding are cornerstones of what we do.”
Gathering precedents efficiently and quickly will help a junior designer to build more essential skills, like building trusted relationships with contractors and consultants.
Thinkers Need Apply
To this end, how will architectural schools – or universities in general – respond to the impacts of AI over time. Architectural theory taught in university is essential but needs to teach us how to build. Meanwhile, colleges are teaching increasingly automated technology and need more critical assessment of its applications. Cichy suspects our traditional educational models will require rethinking and reverse-engineering for the current university-college symbiotic relationship to adapt to the changes ahead. What might change in the future for architecture practice? You will have one architect on a project team with a computer programmer, a game developer, engineers and an architectural technologist. In essence, he sees how this transformation has already begun.
Beyond thinking about seductive renderings, the future of AI will undoubtedly assist us in better decision-making when designing complex buildings. Understanding all the complex variables that go into architecture will make us better understand the drivers of affordable housing, for example, which can look at financial drivers in addition to site constraints. This belief is what Azam Khan, founder and CEO of Trax, calls “systems thinking.” In a recent paper, Khan writes, “Just as architects who begin to adopt parametric design software tools begin to think parametrically, those who adopt systems modelling tools will ideally begin to think holistically in terms of systems and subsystems and reinforcing and balancing processes.”
Khan is a computer scientist leading the development of Trax, a cloud-based building code and standards platform with enough flexibility to input building codes into an easy-to-use compliance platform cross-referenced with accessibility, sustainability and other guidelines like Passive House, BOMA, Toronto Green Standards or the BC Step Code. Trax has already inputted mandatory building codes across Canada and is now in the process of uploading the voluntary codes often related to sustainability. Most of its current customer base comprises building code officials often overwhelmed by the volume, complexity and even contradictory nature of building codes. They also include consulting as part of its business, working with clients like the University of Southern California on building a digital twin to run and automate their building compliance reporting.
Trax realizes the power of linking OpenAI to help with queries and ensure accuracy, realizing that translating natural language to output accurate answers is an evolving challenge. With room for improvement regarding AI-generated responses, at least the company software’s links to building codes and their amendments are correct. Trax has written automation scripts and machine-learning tools to help access specific building code sections, automate unit conversions for users, or produce responses in English, French and parametric-ready versions. Increasingly combining statistical and natural language computational processing, Trax and other companies are achieving easy-to-use results that were unrealizable just a few years ago. While building code compliance is a complex challenge to resolve, with so many permutations such as programmatic adjacencies or materiality configurations, it represents the kinds of industry-wide problems that urgently need to be addressed: the requirement to think holistically. As one of Khan’s building official customers noted, he doesn’t need a BIM model, just a compliance one. Eventually, services like Trax will facilitate and expedite the decision-making processes supporting every building, not just from a code perspective but a sustainability or accessibility one. Trax is already planning to incorporate ANSI standards and life-cycle costing, further building its appeal to a widening customer base.
In its relative infancy, AI is a crude tool in the design profession. Its results are often seemingly random, revealing its sensitivities and unexpected outcomes. But we’re just beginning a significant upheaval in designing and building tomorrow’s cities. Architects should be excited as the evolution of AI will give us more opportunities to collaborate and invent in ever-more dynamic and innovative ways, spending more time strengthening our relationships with clients, consultants and contractors. This revolution is no hallucination.
Ian Chodikoff is an architect currently based in Toronto.