The Architecture of AI
Archinect speaks to designers pushing forward the integration of self-learning systems into architecture to understand how the rise of artificial intelligence is shaping the profession.
“Let us consider an augmented architect at work. He sits at a working station that has a visual display screen some three feet on a side, this is his working surface, controlled by a computer with which he can communicate by means of small keyboards and various other devices.” Douglas Engelbart (1).
This vision of the future architect was imagined by engineer and inventor Douglas Engelbart during his research into emerging computer systems at Stanford in 1962. At the dawn of personal computing, he imagined the creative mind overlapping symbiotically with an intelligent machine to co-create designs. This dual mode of production, he envisaged, would hold the potential to generate new realities which could not be realised by either entity operating alone. Today, self-learning systems, otherwise known as artificial intelligence or ‘AI’, are changing the way architecture is practiced, as they do our daily lives, whether or not we realise it. If you are reading this on a laptop or tablet, then you are engaging directly with a number of integrated AI systems, now so embedded in our the way we use technology, they often go unnoticed.
As an industry, AI is growing at an exponential rate, now understood to be on track to be worth $70bn globally by 2020(2). This is in part due to constant innovation in the speed of microprocessors, which in turn increases the volume of data that can be gathered and stored. But don’t panic — the artificial architect with enhanced Revit proficiency is not coming to steal your job. The human vs. robot debate, while compelling, is not so much the focus in this feature but instead how AI is augmenting design and how architects are responding to and working with these technological developments.
Assuming you read this as a non-expert, it is likely that much of the AI you have encountered to this point has been ‘weak AI’, otherwise known as ANI (Artificial Narrow Intelligence). ANI follows pre-programmed rules in that it appears intelligent but is in effect a simulation of a process of mechanical logic. With recent innovations such as that of Nvidia’s microchip in April 2016, a shift in development is now occurring towards what might be understood as ‘deep learning’ within AI systems, where a system can, in effect, train and adapt itself.
The interest for designers is that AI is now being applied to more creative tasks, such as writing books, making art, web design, or spontaneously generating design solutions. This is in part due to an increased proficiency in the recognition of speech and images. Commentators such as philosopher Nick Bostron suggest the AI industry is now on the cusp of an explosion which will not only shape but drive the design industry in the next century. It is widely recognized that AI has the potential to influence the architectural design process at a series of different construction stages, from site research to the realisation and operation of a building.
01. Site and social research
“By already knowing everything about us, our hobbies, likes, dislikes, activities, friends, our yearly income, etc., AI software can calculate population growth, prioritize projects, categorize streets according to usage and so on, and thus predict a virtual future and automatically draft urban plans that best represent and suit everyone.” — Rron Beqiri on Future Architecture Platform(3).
Gathering information about a project and its constraints is often a first stage of an architectural design process, traditionally involving travelling to a site, perhaps measuring, sketching and taking photographs. In the online and connected world, there is already a swarm-like abundance of data for the architect to tap into, already linked and referenced against other sources allowing the designer to, in effect, simulate the surrounding site without ever having to engage with it physically.
This ‘information fabric’ has been referred to as the ‘internet of things’. BIM tools currently on the market already tap into these data constellations, allowing an architect to evaluate site conditions with minute precision. Software such as EcoDesigner Star or open-source plugins for Google SketchUp already allow architects to immediately calculate necessary building and environmental analyses without ever having to leave the office. This phenomenon is already enabling many practices to take on large projects abroad that might have been logistically unachievable just a decade ago.
The information gathered by our devices and stored in the ‘cloud’ amounts to much more than the material conditions of the world around us. Globally, we are amassing ever expanding records of human behaviour and interactions in real-time. Personal, ‘soft’ data might, in the most optimistic sense possible, work towards the ‘socially focussed design’ that has been widely publicised in recent years by its ability to integrate the needs of users. Is it possible that the internet of things create a socially adaptable and responsive architecture? One could speculate that, for example, when the population of children in a city crosses a maximum threshold in relation to the number of schools, a notification might be sent to the district council that it is time to commission a new school. AI could therefore, in effect, write the brief for and commission architects by generating new projects where they are most needed.
02. Design decision-making
Now that we have located live-updating intelligence for our site, it is time to harness AI to develop a design proposal. Rather than a computer program, this technology is better understood as an interconnected, self-learning system that has the ability to upgrade itself. It is possible to harness a huge amount of computing power and experience by working with these tools, even as an individual — as Autodesk president Pete Baxter told the Guardian(4): “now a one-man designer, a graduate designer, can get access to the same amount of computing power as these big multinational companies”. When working with AI driven interfaces, the architect must input project parameters, in effect an edited ‘design brief’, and the computer system will then suggest a range of solutions which fulfil this criteria. This innovation has the potential to revolutionise how architecture is not only imagined but how it is fundamentally expressed for designers who adopt these methods.
I spoke with Michael Bergin, a researcher at Project Dreamcatcher at Autodesk’s Research Lab, to get a better understanding of how AI systems are influencing the development of design software for architects. He explained to me that while the aim of Autodesk’s Research Lab was to produce software for the automotive and industrial design industries, Dreamcatcher’s use cases are increasingly filtering into the architecture and construction sectors. For example, Dreamcatcher was used recently to develop The Living’s generative design for Autodesk’s new office in Toronto and MX3D’s steel bridge in Amsterdam. The basic concept is that CAD models of the surrounding site and other data, such as client databases and environmental information, are fed into the processor. Moments later, the system outputs a series of optimised 3D design solutions ready to render. These processes effectively rely on cloud computing to create a multitude of options based on self-learning algorithmic parameters. Lattice-like and fluid forms are often the aesthetic result, perhaps unsurprisingly, as the software imitates structural rules found in nature.
The Dreamcatcher software has been designed to optimise parametric design and link into and extend the capabilities of existing software designed by Autodesk, such as Revit and Dynamo. Interestingly, Dreamcatcher can make use of a wide and increasing spectrum of design input data — such as formulas, engineering requirements, CAD geometry, and sensor information — and the research team is now experimenting with Dreamcatcher’s ability to recognise sketches and text as input data.
Bergin imagines the future of AI-driven design tools as “systems that accept any type of input that a designer can produce [to enable] a collaboration with the computer to iteratively target a high-performing design that meets all the varied needs of the design team”. This would mean future architects would be less in the business of drawing and more into specifying requirements of the problem to operate in sync with their machine counterparts. Bergin also suggests architects who adopt AI tools would have the ability to “synthesize a broad set of high level requirements from the design stakeholders, including clients and engineers, and produce design documentation as output”, arguably in line with Engelbart’s vision of AI augmenting the skills of designers.
“[Integrating AI systems into architectural design will enable architects to utilize] the computer as a true partner in solving hard design problems rather than a super-powerful 3D drafting board” — Michael Bergin
AI is also being used directly in software such as Space Syntax’s ‘depthmapX’, designed at The Bartlett in London, to analyse the spatial network of a city with an aim to understand and utilize social interactions in the design process. Another tool, Unity 3D, is built from software developed for game engines to enable designers to analyse their plans, such as the shortest distances to fire exits. This information would then allow the architect to re-arrange or generate spaces in plan, or even to organize entire future buildings. Examples of architects who are adopting these methods include Zaha Hadid with the Beijing Tower project (designed ante-mortem) and MAD Architects in China, among others.
03. Client and user engagement
Smart design tools such as Materiable by Tangible Media are already beginning to experiment with how AI can begin to engage with and learn from human input into computer systems. Such a change in representational method has the potential to shift what is possible within the field of architectural expression, perhaps as CAD drafting did at the beginning of this century.
As a significant proportion of AI technology has been developed for the gaming industry, its ability to produce forms of augmented reality, ‘AR’, also appear to be significant. AR holds potential to change both the perception and engagement with the design process for architects and non-architects alike. Through the use of additional hardware, AR can enable people to experience a design prior to construction. The lights, sounds, even the smells of a building can be simulated, which could reorder the emphasis architects currently give to specific elements of their design. It is possible that many architecture projects will also remain in this unbuilt zone, in a parallel digital reality, which the majority of future world citizens will simultaneously inhabit.
04. Realising designs and rise of the robot craftsmen
AI systems are already being integrated into the construction industry — Swiss studio Computational Architecture are one of the first to work with ‘robotic craftsmen’ to explore AI in construction technology and fabrication. Michael Hansmeyer and Benjamin Dillenburger, founders of Computational Architecture, are investigating the new aesthetic language these developments are starting to generate. “Architecture stands at an inflection point,” he suggests on their website(5), “the confluence of advances in both computation and fabrication technologies lets us create an architecture of hitherto unimaginable forms, with an unseen level of detail, producing entirely new spatial sensations.”
“The confluence of advances in both computation and fabrication technologies lets us create an architecture of hitherto unimaginable forms, with an unseen level of detail, producing entirely new spatial sensations.” — Michael Hansmeyer / Benjamin Dillenburger, Computational Architecture
3D printing developed alongside AI technology can offer twenty-first century architects a significantly different aesthetic language, perhaps catalysing a resurgence of detail and ornamentation, now rare due to the global decline in traditional craftsmanship. Hansmeyer and Dillenburger’s Grotto Prototype for the Super Material exhibition, London, was a complex architectural grotto 3D-printed from sandstone. The form of the sand grains were arranged by a series of algorithms custom-designed by the practice and allowed forms significantly different to those produced by traditional stonemasonry.
AI-driven robotics are also becoming more common on construction job sites, at the time of writing now mostly concerned with human resources and logistics. According to the Association of Equipment Manufacturers(6), their applications will soon expand to bricklaying, concrete dispensing, welding and demolition. Another example of their future use could include working with BIM to identify missing elements in the construction quality assurance process and update the AI in real-time.
In part due to cost, large scale projects, such as government-lead infrastructure initiatives, might well be the first to apply this technology, followed by mid-scale projects in the private sector, such as cultural buildings. The challenges of the construction site will bring AI robotics out of the indoor, sanitised environment of the lab into a less scripted reality. Robert Saunders, a researcher into AI and fabrication at the University of Sydney, told New Atlas(7) that “robots are great at repetitive tasks and working with materials that react reliably…what we’re interested in doing is trying to develop robots that are capable of learning how to work with materials that work in non-linear ways… like working with hot wax or expanding foam or, more practically, with low-grade building materials like low-grade timber.” Saunders foresees robot stonemasons and other ‘craftsbots’ working in yet unforeseen ways, such as developing the architect’s ‘skeleton plans’, in effect, spontaneously generating a building on-site from a sketch.
05. Integration of AI systems
The integration of AI into the built environment and furniture systems appears to fall under two categories: either integrating artificial technologies with existing infrastructure or designing around AI systems. There is a lot of excitement in this field, especially within product marketing agencies, influenced in part by Mark Zuckerberg’s personal project to develop networked AI systems within his home, which he announced in his New year’s Facebook post in 2016(8). Zuckerberg’s wish is to develop simple AI systems to run his home and help with his day-to-day work. His technology would have the ability to recognise the voices of members of the household and respond to their requests.
Some designers and architects are taking on the challenge of designing home-integrated systems, such as the Ori System of responsive furniture, or gadgets such as Eliq for energy monitoring. Behnaz Farahi is a young architect activating her research into AI and adaptive surfaces to develop interactive designs, such as in her Aurora and Breathing Wall projects. Farahi creates immersive and engaging indoor environments which adapt to and learn from their occupants.
Researchers and designers working in the field of AI are attempting to understand the potential of computational intelligence to improve or even upgrade parts of the design process with an aim to create a more functional and user-optimised built environment. It has always been the architect’s task to make decisions based on complex, interwoven and sometimes contradictory sets of information. As AI gradually improves in making useful judgements in real-world situations, it is not hard to imagine the human and machine overlapping and engaging with each other in the design process. While the increasing power of AI systems may raise questions in terms of ownership, agency and, of course, privacy in data gathering and use, the upsurge in self-learning technologies is already altering the power and scope of architects’ engagement with design and construction. As architect and design theorist Christopher Alexander said back in 1964(9),“We must face the fact that we are on the brink of times when man may be able to magnify his intellectual and inventive capacity, just as in the nineteenth century he used machines to magnify his physical capacity.”
“To think architecturally is to imagine and construct new worlds, integrate systems and organise information, which lends itself to the front line of technical development.”
During our interview, Bergin gave some insights into how he sees AI technology impacting designers in the next twenty years. “The architectural language of projects in the future may be more expressive of the design team’s intent”, he suggested, “generative design tools will allow teams to evaluate every possible alternative strategy to preserve design intent, instead of compromising on a sub-optimal solution because of limitations in time and/or resources.” Bergin believes AI and machine learning possesses the ability to support a “dynamic and expanding community of practice for design knowledge”. He can also foresee implications of this in the democratisation of design work, suggesting that “the expertise embodied by a professional of 30 years may be more readily utilized by a more junior architect”. Overall, he believes “architectural practice over the next 20 years will likely become far more inclusive with respect to client and occupant needs and orders of magnitude more efficient when considering environmental impact, energy use, material selection and client satisfaction”.
Autodesk present Pete Baxter also suggests architects have little to fear from artificial intelligence: “Yes, you can automate. But what does a design look like that’s fully automated and fully rationalised by a computer program? Probably not the most exciting piece of architecture you’ve ever seen.” At the time of writing, it is still proving difficult to automate design decision-making that would at first glance seem simple for a human. A number of research labs, including the MIT Media Lab, are working to solve this.
Architectural language and diagramming have been part of programming complex systems and software from the start, and they have had significant influence on one another. To think architecturally is to imagine and construct new worlds, integrate systems and organise information, which lends itself to the front line of technical development. As far back as the 1960s, architects were experimenting with computer interfaces to aid their design work, and their thinking has inspired much of the technology and interfaces we now engage with each today.
(1) Douglas C. Engelbart. 1962. Augmenting Human Intellect: A conceptual framework. [ONLINE] Available at: http://www.dougengelbart.org/pubs/augment-3906.html. [Accessed 7 March 2018].
(2) Reuters. 2016. Tech CEOs Declare This the Era of Artificial Intelligence. [ONLINE] Available at: http://fortune.com/2016/06/03/tech-ceos-artificial-intelligence/. [Accessed 7 March 2018].
(3) Rron Beqiri. 2016. Architecture and Urban Planning in the age of Artificial Intelligence. [ONLINE] Available at: http://futurearchitectureplatform.org/news/28/ai-architecture-intelligence/. [Accessed 7 March 2018].
(4) Michael Tom Meltzer. 2014. Robot doctors, online lawyers and automated architects: the future of the professions?. [ONLINE] Available at: https://www.theguardian.com/technology/2014/jun/15/robot-doctors-online-lawyers-automated-architects-future-professions-jobs-technology. [Accessed 7 March 2018].
(5) Michael Hansmeyer. 2017. Computational architecture. [ONLINE] Available at: http://www.michael-hansmeyer.com/profile/about.html?screenSize=1&color=1. [Accessed 7 March 2018].
(6) AEM. 2016. How Artificial Intelligence Could Revolutionize Construction. [ONLINE] Available at: https://www.aem.org/news/october-2016/how-artificial-intelligence-could-revolutionize-construction/. [Accessed 7 March 2018].
(7) Richard Moss. 2015. Creative AI: Algorithms and robot craftsmen open new possibilities in architecture. [ONLINE] Available at: http://newatlas.com/creative-ai-algorithmic-architecture-robot-craftsmen/36212/. [Accessed 7 March 2018].
(8) Mark Zuckerberg. 2016. [ONLINE] Available at: https://www.facebook.com/zuck/posts/10102577175875681. [Accessed 7 March 2018].
(9) Alexander, C. 2002. Notes on the synthesis of form. Cambridge, Mass. [u.a.]: Harvard Univ. Press, p.11.