Just over a century since The Manifesto of Futurist Architecture declared the city must be rethought and rebuilt like an “immense and tumultuous shipyard” – “everywhere dynamic”, and the house like a “gigantic machine”, it may be that author Antonio Sant’Elia had things the wrong way around.
Because although his machine-fetishising sketches inspired our common vision of a science-fiction future – as in Fritz Lang’s 1927 film Metropolis, with its technological Tower of Babel an imposing centrepiece – it might be the gigantic machines that are making our houses.
Architecture and AI visionaries – forming especially around MIT in the 1950s, through to the later work of MIT Media Lab co-founder Nicholas Negroponte – and design pioneers have long thought about automating the creation of our environments. Now the technology is catching up to their ideas, and a radical shift into AI-assisted design is taking hold, with implications that could radically transform the form, feel and function of the places we inhabit.
Completely automated design is not quite there yet. This crop of generative, AI-assisted tools is rather new. But there are signs that we could be on the cusp of a revolution in how our buildings, towns and cities are created. Will these begin to take on a homogeneous shape, recognisable as AI-planned spaces? And is this the beginning of the ‘copy-and-pasted’ city – or do we already inhabit those, with the identical new-build properties that seem to crop up everywhere?
AI helping hand
Advocates argue that AI-based city design could remove burdensome manual labour, allowing architects, designers and planners to focus on creativity. On the other hand, could AI accelerate more of the same – a ruthlessly efficient approach to stuffing more people into buildings and maximising rents. Whatever happens, AI-assisted design appears set to radically change the future of architecture.
Despite the long tail of thinking around automated design, the drafting process was largely manual until very recently, even in software like the ubiquitous AutoCAD or the building information modelling tools that added more context to designs and have become dominant. “There was always the dream of automating design and urban planning but, little by little, it happened over the last decade or so,” comments Imdat As, architect and co-editor of The Routledge Companion to Artificial Intelligence in Architecture.
The machine learning revolution has helped create the conditions for adequate computing power, with the imitation-thinking enabled by neural networks finally making generative design commercially viable. Nudging this AI-assisted world into reality are new tools backed by Silicon Valley, such as Delve, owned by Google subsidiary Sidewalk Labs, and SpaceMaker, which was recently acquired by computer-aided design giant Autodesk for $240m (£196m).
There was always the dream of automating design and urban planning
Unlike the painstakingly crafted line-by-line processes normally associated with architecture proposals, these tools allow the user to view and play with a huge range of variables – prioritising or adjusting nuances we may take for granted, like noise levels, temperature or window views – and then generating design options. With the traditional approach, planning teams are limited by their time and their tools, so proposals rarely exceed a selection of three to five designs. But by using AI-assisted tools, planners can explore hundreds, if not thousands of options, with their subtle differences illustrated on a 3D map so various stakeholders can view progress or collaborate as plans evolve.
A huge shift in architectural planning
SpaceMaker co-founder and CEO Håvard Haukeland says that this is one of his platform’s key benefits: because urban planning is very much about competing interests, projects can end up bogged down by meetings where people are spending more time putting forward their cases, rather than exploring multiple solutions. The latter, he argues, is “better for the city, better for the people living in the apartments or using the office spaces, and it’s usually better for the economics of the project and the developer”.
Haukeland adds that this approach could represent a huge shift in architecture and planning – and one that can virtually eliminate what those in the industry colourfully term ‘Oh Shit Moments’: when a design has already been fixed, but the team had forgotten to carry out essential tasks like noise analysis, thereby potentially setting project deadlines back – sometimes quite literally to the drawing board.
That was a fate avoided by the growing Kivistö district in Vantaa, Finland, where a new railway line will see its population swell to 45,000 in the coming decades (a relatively large population in Finnish terms).
For planners, it was crucial to balance new, dense urban neighbourhoods within it with the district’s proximity to nature and its silver birch-lined streets, while also remaining on course for a 2050 carbon neutral target. Even at a late stage, designers were able to use SpaceMaker to refine plans for interior courtyards, reducing wind effects and placing a sunny terrace for future residents. “The software almost downright persuades one to try different options,” commented town planner Ville Leppänen.
Meanwhile, in Sofia, Bulgaria, city planners applied Delve to a city unit to map out future development strategies for the area. They told RIBAJ that the test project provided “valuable and important insight into the many possibilities that parametric planning offers”, with one of the key benefits being able to “rapidly change input data and generate output results that may not have been even considered before”.
This kind of experimentation was just not possible at this pace or scale previously. Michele Pelino is principal analyst, edge computing and the internet of things, at Forrester Research. She notes that although future-looking cities like Singapore had experimented with digital twin technology – where virtual copies are made of existing places and are then subject to computer simulations – applied generative design is new terrain and how it will shape our buildings and cities is yet to be determined.
Model renaissance
With the right prompts and some patience, algorithms appear capable of helping to create stunning portraits and fantastical worlds, as evidenced by OpenAI’s Dalle2 system: an easily navigable fountain of artistry that anybody can use. This AI-generated art is a window into what will become possible on a larger scale with our buildings, especially when combined with 3D printing, coming together to encompass one automated process. So says Eleanor Watson, IEEE AI Ethics engineer and AI Faculty at Singularity University.
This could, she says, build works of incredible complexity but at no extra marginal cost – and would be an opportunity to push back against the “stark utilitarian mass-produced simplicity, inoffensive and timeless yet dull and soulless”. “We might soon see a renaissance, whereby plainness becomes passé, in a world where beauty has become next to free,” she says.
First, though, there are many more complexities to creating a building (and even more on the scale of towns and cities) than generating a pleasing portrait image. With all their variables, location-specific considerations, the context-dependent nature of floorplans, and the algorithmically impossible-to-pin-down overall feeling of a place, it may be some time yet before machines are helping to bring about that renaissance.
For now, it is the nuts-and-bolts stuff where the latest AI-generative tools excel; the design is not quite end-to-end – meaning, pushing a button won’t instantly generate you a building or district to your liking.
You want to encourage positive mutations and that’s what AI and machine learning make possible
But even these optimisations have the potential to change the look and feel of spaces, adds Pelino.
Being able to analyse, calculate and map predicted temperatures, for instance, could help developers avoid creating urban heat islands, and create cooler conditions for residents as buildings and cities evolve. And as sustainability becomes a more pressing concern, it may be proven that our approaches to buildings and cities are woefully inadequate – and that AI-imagined geometric models surprise designers with the optimised shapes they take.
In time, as technology marches forward, new, surprising, aesthetic forms may crystallise.
Stephen Barrett, partner at architectural designers RSHP, believes that AI will be able to take the more mechanistic day-to-day activities of planners and designers, and “autocomplete” some of the laborious processes. There are “great advantages” to this, he says: it frees up time and space to work on the interesting stuff, to innovate and create.
At present, for instance, an AI-generated Picasso scene will create a rough caricature of the artist’s style based on the inputs fed to it. But however impressive it might be, it is no more than an approximation – a kind of evolved copy – of existing images and aesthetics rather than something altogether new.
So, determining the future with algorithms needs to be considered “very carefully”, he says: “It’s a little bit like Modernist architecture in the mid-to-late 20th century. It was meant to be a form of architecture free of baggage and values but, in the end, you could argue it was fairly post-colonial dominant – and you have the same buildings in São Paulo as you have in Riyadh as you have in Singapore or Moscow or wherever.
“You want to encourage positive mutations and that’s what the rapid processing and multiple iterations of AI and machine learning make possible,” he adds. “But also, to ensure that the output is intelligent, and not simply a reflection of the limitations of the inputs.”
So designers will have to tread carefully and remain conscious of algorithmic bias, where software reproduces errors due to the prejudices of the software designers.
Flair over form for the foreseeable
Haukeland argues that more design options and therefore more variety can hardly ever be a bad thing. But if you look at history, he adds, revolutions in architecture have occurred across thousands of years and in the end, there’s always something new, better, or smarter that builds on the past. “It’s easy to stand here at the beginning of a new era and say this will change everything,” he says. “But in 10, 20 or 30 years’ time, we might say that generative design and AI was super interesting but it had its weaknesses. So I don’t think humans have come to the end.”
Imdat As wonders what designers would work on if AI were to produce 90% of the buildings.
“The top 10 designers – the Zaha Hadids, and so on – will always be there, with new ideas, new aesthetics. Those will be the designers who come up with a new design idea,” he says. “And what if they trained in-house AI systems? Instead of, say, 10 buildings a year, they might build a million, all over the world. The power of AI for an architectural company could be amazing. The business structure could be: if you use my AI system, you pay royalties. It could change architectural practice models. I think there will be those types of changes.”
While machines are learning to navigate environments, they can never know the experience of doing so
Presently, it’s unlikely that residents would notice at all if computers helped to build their housing, angling a complex five or six degrees to maximise liveability. And it may be too early to tell if algorithmically designed buildings and cities will leave telltale AI marks in their floorplans or on their façades, but As is hopeful that “its own sort of urban morphology” will emerge.
Barrett wonders if the “step-change revolutions you get with accidental, erratic, unique or eccentric human inputs” could ever occur with artificial intelligence thrown into the mix. “If you took a city like Paris and ran that as an existing data set, you’d have more Paris,” he comments, “which is not a bad thing. But would you ever have had a Pompidou Centre?”
One thing does, though, seem certain. Given the efficiency gains, AI-assisted design will play an increasingly important role in that future planning, developing, building. But just as with software and the complex data sets that inform or mediate our lives, keeping a human in the loop is likely to be a fixture in the foreseeable, “marshalling, judging, and continuing to select”, says Barrett.
“After all, while machines are learning to navigate environments, they can never know the experience of doing so,” adds Watson. “We must never sacrifice the feel of an urban landscape at the altar of efficiency, nor cause a malfunction in any person’s enjoyment of a resource.”
