AI in architecture
January 8, 2025 4:45 PM Subscribe
How is AI being (or might eventually be) used in architecture?
When I google "AI in architecture", the articles that come up by Autodesk and the RIBA (1st and 3rd in my results) cover the bases well enough that I'd just be repeating what they say. Autodesk makes software that architects use extensively; RIBA is a professional organization, kind of like the UK version of the AIA.
posted by LionIndex at 6:50 PM on January 8 [2 favorites]
posted by LionIndex at 6:50 PM on January 8 [2 favorites]
I would expect licensure requirements & liability exposure to keep "AI" largely in the domain of space-planning prototypes, trying out different facade combinations, and the like. So, useful, but not exactly a paradigm shift. More "place the windows so we get more sun in these locations, but not too much, on the following dates...." type of stuff.
Given the flaws of current tech, I would expect any E&O carrier to go completely apeshit if anyone with their coverage used "AI" to any degree in any significant area of a project ("significant" being in this case "an area of a project which would be amenable to a lawsuit if there is a problem").
Generally speaking the AHJs (in North America, anyway) hold the legal stance that if you signed off on it, it's your problem and you will be required to suffer accordingly (not OpenAI or Autodesk).
I could see a paradigm shift in the future, but only if whomever created the software were willing to accept the various penalties and punishments associated with failure to perform, insufficient documentation, design errors, and so on.
Remember, a HUGE chunk of the AEC universe is effectively determined by legal requirements, contractual obligations and insurance bonding capacity -- technology is actually kind of a minor player.
posted by aramaic at 9:09 PM on January 8 [4 favorites]
Given the flaws of current tech, I would expect any E&O carrier to go completely apeshit if anyone with their coverage used "AI" to any degree in any significant area of a project ("significant" being in this case "an area of a project which would be amenable to a lawsuit if there is a problem").
Generally speaking the AHJs (in North America, anyway) hold the legal stance that if you signed off on it, it's your problem and you will be required to suffer accordingly (not OpenAI or Autodesk).
I could see a paradigm shift in the future, but only if whomever created the software were willing to accept the various penalties and punishments associated with failure to perform, insufficient documentation, design errors, and so on.
Remember, a HUGE chunk of the AEC universe is effectively determined by legal requirements, contractual obligations and insurance bonding capacity -- technology is actually kind of a minor player.
posted by aramaic at 9:09 PM on January 8 [4 favorites]
I spoke with a structural engineer from NASA last year about this - he was very excited. For load-bearing structures it can put together really interesting geometries that allow for different or less materials to be used, or sci-fi looking beams and supports with holes in them that save weight but have no effect on rigidity, etc. These things are only possible through extensive training of models on this stuff, simulations that they test stuff out in, and (what is most useful about AI generally) a willingness to perform human-esque tasks like designing a new beam, assembling a structure with it, and performing several stress tests — millions of times with no preconceptions about what is "correct" or "best." He said it produces really alien looking stuff but with remarkable structural qualities. So I suspect we will see more buildings and features with unusual shapes or styles, which have been arrived at by an AI (or AI-adjacent) system.
posted by BlackLeotardFront at 10:59 PM on January 8 [4 favorites]
posted by BlackLeotardFront at 10:59 PM on January 8 [4 favorites]
Regarding what aramaic said: I don't think there'll really be a limit with regulations, insurance or licensing. AEC (Architecture, Engineering, Construction) firms already have their work largely produced by unlicensed people who basically know what they're doing, but the people with licenses are expected to review the construction documents before they go out to be built and any liability still falls on the licensed person approving the drawings. I don't think having the documents produced by AI really changes that. But I think the idea that AI takes care of a lot of the banal iterative tasks involved in trying out different approaches (finishes, room layouts, solar and energy stuff, etc) is basically correct. The throughline with technology in architecture has been making alterations to a design much easier, quicker, and more complete while requiring a lot more input on the back end before the technology even starts doing anything. Going from hand drawing, where if a wall needed to move three feet right before construction documents went to the contractor it meant A LOT of work - possibly a week - and maybe even starting drawings over; now with revit I just move the wall and my drawings update automatically because it's a full 3D model instead of a bunch of lines, but I have to have a library of things that make up the model from the start. I still have to figure out how to relocate all the ducts, lighting, and plumbing stuff that might be associated with that moving wall, and AI might be able to handle that.
Regarding what BlackLeotardFront says: until we get to where we can 3D print steel affordably, we're stuck with using a very standardized kit of parts that people can put together in the field. I think AI will be able to analyze things using that kit of parts - we already do load test 3D models in a computer doing that, it would just be AI optimizing given the framework we have to work within.
But that's what AI is doing now - wild, impractical, unbuildable things; which is something architects have been drawing for hundreds of years.
posted by LionIndex at 11:21 PM on January 8 [5 favorites]
Regarding what BlackLeotardFront says: until we get to where we can 3D print steel affordably, we're stuck with using a very standardized kit of parts that people can put together in the field. I think AI will be able to analyze things using that kit of parts - we already do load test 3D models in a computer doing that, it would just be AI optimizing given the framework we have to work within.
But that's what AI is doing now - wild, impractical, unbuildable things; which is something architects have been drawing for hundreds of years.
posted by LionIndex at 11:21 PM on January 8 [5 favorites]
If you'll forgive the self-link, I wrote about this recently for Wallpaper* magazine after interviewing the team at Zaha Hadid Architects. It's not a particularly in-depth article, but the gist is that ZHA are all-in with AI in two different ways; generative AI as a replacement for traditional rendering and sketching, with an AI trained on their massive image database. This allows them to iterate designs within pre-established parameters, pretty much on the fly.
For more in-depth use of the technology, they've teamed up with NVIDIA, all part of that company's much-vaunted Omniverse platform. The idea here is to create exact 'digital twins' of buildings, from structure right down to the HVAC, just as manufacturers do with factories. The resulting dataset will then inform AI-shaped building design, capable of catering for any number of parameters (energy consumption, material use, programme, etc).
To be honest, it felt pretty dystopian and I said as much. Their response was that what they usually build is unbelievably bespoke and complex - 'construction is not rocket science, but pretty much every building is a prototype'. In contrast, tens of thousands of substantially pre-fab and basically identical towers are being built around the world right now with little to no care for context or form. A well-trained and governed AI, they believe, can counter this and shape a more efficient and interesting architecture. We shall see.
posted by jonathanbell at 2:55 AM on January 9 [5 favorites]
For more in-depth use of the technology, they've teamed up with NVIDIA, all part of that company's much-vaunted Omniverse platform. The idea here is to create exact 'digital twins' of buildings, from structure right down to the HVAC, just as manufacturers do with factories. The resulting dataset will then inform AI-shaped building design, capable of catering for any number of parameters (energy consumption, material use, programme, etc).
To be honest, it felt pretty dystopian and I said as much. Their response was that what they usually build is unbelievably bespoke and complex - 'construction is not rocket science, but pretty much every building is a prototype'. In contrast, tens of thousands of substantially pre-fab and basically identical towers are being built around the world right now with little to no care for context or form. A well-trained and governed AI, they believe, can counter this and shape a more efficient and interesting architecture. We shall see.
posted by jonathanbell at 2:55 AM on January 9 [5 favorites]
Remember, a HUGE chunk of the AEC universe is effectively determined by legal requirements, contractual obligations and insurance bonding capacity -- technology is actually kind of a minor player.
IMO: we are trying to use AI for software design. There are honestly so many legal rules, business rules, design considerations, regulations, and such that no (normal) human can remember all of it. It's currently stored in documents, and excel sheets, and who knows who has access or remembers or even thinks of what they might need to get from a document or from that magic person who can recall it by memory.
What does that mean for new hires? Or for making sure all that is actually met?
I can tell you we have trouble getting new people up to speed and making sure all the regulations are met is a crapshoot. So there is extreme value in AI for taking all that into account, and making sure that more people than just the very best, or most senior consider and recall it.
So IMO (again) getting all that into an AI that normal employee can recall, make sure it's taken into account will be a real game changer use for AI. Software is also bespoke for no particular reason.
And you may say, 'well all that sounds nice' but approximately 60% of most infrastructure projects' budgets are spent on consulting upfront before anything is done. Years of design work occurs. That means 40% is the throngs of actual hardhat wearing workers and materials that you see! The rest is invisible.
If you could get that to 50/50 or slightly less it would be a real game changer, and all those savings could be for a 2nd building or for higher wages, or profit or whatever.
posted by The_Vegetables at 9:35 AM on January 9
IMO: we are trying to use AI for software design. There are honestly so many legal rules, business rules, design considerations, regulations, and such that no (normal) human can remember all of it. It's currently stored in documents, and excel sheets, and who knows who has access or remembers or even thinks of what they might need to get from a document or from that magic person who can recall it by memory.
What does that mean for new hires? Or for making sure all that is actually met?
I can tell you we have trouble getting new people up to speed and making sure all the regulations are met is a crapshoot. So there is extreme value in AI for taking all that into account, and making sure that more people than just the very best, or most senior consider and recall it.
So IMO (again) getting all that into an AI that normal employee can recall, make sure it's taken into account will be a real game changer use for AI. Software is also bespoke for no particular reason.
And you may say, 'well all that sounds nice' but approximately 60% of most infrastructure projects' budgets are spent on consulting upfront before anything is done. Years of design work occurs. That means 40% is the throngs of actual hardhat wearing workers and materials that you see! The rest is invisible.
If you could get that to 50/50 or slightly less it would be a real game changer, and all those savings could be for a 2nd building or for higher wages, or profit or whatever.
posted by The_Vegetables at 9:35 AM on January 9
I’m not an architect, but I do follow an architect on Instagram named Andrew Kudless who writes and posts about both computational/parametric architecture and using generative AI in architectural design. His website is https://www.matsys.design/profile
posted by hapticactionnetwork at 4:09 PM on January 9
posted by hapticactionnetwork at 4:09 PM on January 9
« Older What did I do to my keyboard and or laptop to make... | What’s the longest someone has tested positive on... Newer »
You are not logged in, either login or create an account to post comments
posted by moosetracks at 6:34 PM on January 8 [13 favorites]