AIA and AI
I’m afraid it’s some more musing on AI. Recently, I was given a copy of Neil Leach’s book Architecture in the Age of Artificial Intelligence: An Introduction to AI for Architects [thanks VpD!]. It’s not the only book on the topic out there but, with a field that seems to be changing so rapidly, I wondered how up-to-date any book on the topic could be. Nevertheless, I marveled at how quick the world of publishing is to jump when an on-trend book proposal comes along.
Some or all of the above books could just be bandwagon jumping so the only way to tell was to read a bit. I had some time on my hands so I read AIAAI:AIAIA anyway. The day before, I’d watched a YouTube video [thanks KT!] titled “The Future is a Dead Mall” about how contender metaverses are endlessly talked up but never actually deliver. It was illuminating, although I thought the title unfair to actual dead malls that at least have a potentially re-useable megastructure. But with a dead metaverse there’s nothing. The real problem seems to be getting a metaverse to live and this is the endless talking up about how it’s going to be better than the lame prototypes we get to see. This all seemed familiar. By now, we ought to be indignant that we’re not already living in 3D printed buildings on Mars but fortunately – by which I mean suspiciously – something else came along for journalists to get us excited about.
This first book of AIAAI:AIAIA – three are planned – talks up AI while an upcoming one talks it down and lets not worry about the third just yet. This first book focussed on how AI would assist architects in the “design” (i.e marketing) part of their work by speeding up the generation and development of alternatives to show clients. This seems to be what the verb design now means so I’m not going to upset myself too much over that.
On pages 30 and 31 is the following description of creating novelty using a “generator” and a “discriminator” that seem very analogous to an artist and an art market. The goal is to create novelty but nothing so novel as to be beyond understanding. This technology represents a money tree for our current novelty generators provided our appetite for their particular brands of novelty remains insatiable. It doesn’t make the treadmill of artistic creativity redundant. It just automates it.
To make us less suspicious of any threat posed by AI, the author talks, like many others, of augmenting human abilities rather than making them redundant. This is well trodden ground but time will tell soon enough. Many would be happy for that to be the case because it retains the same hierarchies and working practices they’ve worked so hard to make us think are normal. For one, it entrenches the system of design architects and architects of record that works in favor of the usual suspects. But, within the design architect offices, that same AI maintains the division of architects into those that don’t design, and those that design or choose between alternate designs, alternatives, options or states or whatever passes these days for the intersection of option selection and branding nous. So yes, there’s much talk like this in this next example. Unlike the advent of prefabrication a century ago, at least nobody’s telling us how much more free time we’re all going to have now that machines can do it all for us.
Perhaps I shouldn’t be too hard on a book that says it’s an overview but I was relieved to see signs of a broader outlook, even thought it’s the person at the top of the tree who gets to choose which potential outcome to run with. This seems worse than a top-down form-making process whatever that is. “Take this sketch and play with it and come show me at the end of the day.” ?
I thought I’d skip to Chapter 6 because everything before basically summarizes the past year but in the end I read on, passing some interesting-looking rabbitholes along the way. One drew a difference between creativity and a design strategy. Soon after however, was a description of how AlphaGo beat Korean Go master Lee Sedol in Game Two of a five game tournament. Lee Sedol said that AlphaGo’s decisive move 37 was beautiful and previously unknown and he had no doubt AlphaGo was “creative”. However, given the huge number of board configurations and plays possible, maybe it was just a winning strategy that human players hadn’t yet encountered? [To be fair, the author does ask the same question later.] We might be being too quick to call something creative when it’s really just something we haven’t seen before.This could be why our obsession with novelty ever since Delight began standing in for Virtue. A separate question is “Was AlphaGo pleased it had won? Did it enjoy the match?” The question Go players, professional and non professional alike, need to ask themselves is “Is Go still worth playing if it is impossible to win a game against AI?” I think they’ll say yes. Winning a game with novel moves is good, but there’s also a separate enjoyment to be had from just playing the game and exercising one’s brain. Architects might well ask themselves if they even enjoy architecture if there’s no emotional involvement? Finally, I’m not yet aware of any discussion of what
architecture our built environment might stand to lose from total adoption of AI so I won’t be spending time on these questions until someone tells me what architectural creativity is and want we as recipients want from it. Until then, I have the impression that, with creativity and AI, all we’re doing is redesigning the nails to suit our new hammer.
Chapter 7 of the book speculates on the AI city of the future. I think it’s too early but this bringing together of urbanism and AI made me think of Episode 15 of the Netflix adaption of Cixin Liu’s trilogy, The Three Body Problem, (or Chapter 17 in Book 1) where a city (styled in China’s Warring States Period 481–221 B.C.) is configured as a computer by having 30 million soldiers hold a black flag in one hand and a white flag in the other. If the two input soldiers of three standing in a triangle both raised a black flag then the output soldier would raise a black flag to forming an AND gate. An OR gate, NAND gate, NOR gate, XOR-gate and XNOR-gate and NOT gate were similarly formed. A motherboard was fthus ormed and, together with a CPU as the Aztec-like pyramid of the palace, configured a computer that could be loaded with an operating system and program. Scribes were the memory. Configuring a city as an analogue digital computer was an audacious idea in the book and a set piece in the adaption.
I only mention this because the scene reminded me of post The Microprocessor Is Not Trying To Be Beautiful from the early days of June 2010 when I was trying to make a case for performance beauty and seeing beauty in things designed for optimum performance and without any regard for how they look. “You don’t find an ugly microprocessor”, I wrote. It was profoundly uninfluential.
Like many of these posts, this one was prompted by two things entering my life at the same time. One was a copy of the book I mentioned and the other was an online news article with a title we’ve come to expect: ‘It’s already way beyond what humans can do’: will AI wipe out architects? Both turned out to be much the same thing. I only want to comment on two of the article’s images. The first one is an AI generated something in the style of Le Corbusier we’re told. Even if it looks like it’s made of cork, it does seem like something that’s possible yet haven’t seen before and perhaps this is why people say things like I don’t know what creativity is but I know it when I see it. Elsewhere, people are trying to bring Zaha Hadid back from the dead but, while we’re on the subject of branding, Foster+Partners are claiming a lead in AI but aren’t about to share their training data. I’m sure ZHA won’t be either but this will only lead to a universe of individual bespoke AI systems to which the same old brand value has been transferred. In the end, nothing changes. The big unspoken fear must be that future architectural AI systems are connected and able to share materials and construction cost data, enabling clients to make more informed choices about exactly what it is they’re paying for and whether it’s cost efficient or, if not, what level of cost inefficiency they will settle for.
It may be a spoiler as my author tells me (via this article) that the AI offering of a company called Xkool is now the new cutting edge [why do we never use “avant garde” for technolgies?] because it has the AI means to do everything from design and layouts to construction drawings. In the image below, I see a bit of motherboard logic in this development but that’s not a bad thing. At least has a real context that it seems to fit in with unlike the context-free image we see above. And so positions are being taken. We might be headed for another round of Expressionism vs. Rationalism, Modernism vs. Everything That Went Before, Post-Modernism vs. Modernism and so on. It runs and runs. The Chinese founders of Xkool used to work for OMA and say they became disillusioned with OMA’s conventional way of working. People see the kool in the name as homage to Rem Koolhaas and the prefix X is said to be a contraction of ex-Kool but that X could just be the universal symbol for No.
Finally, suppose we have the perfect AI “assistant” for architecture – all of it, not just the bits we see. I’ll call my hypothetical AI assistant ArchI. ArchI can of course pass the AIA registration exams as well as the RIBA Part III exams as hosted by London’s AA. Can ArchI be a registered architect? Probably, because ArchI has no criminal record, is not bankrupt or has any kind of financial arrangement with creditors, and so on, even if it’s never had the opportunity to. But suppose we employ unregistered ArchI anyway and get it (he/she/them) to do our work. It’s a huge project – a hospital let’s say – and ArchI generates the design, does the layouts and the internal elevations (even making sure each space has the required equipment like with those Autodesk add-ons in the early 1990s). ArchI also knocks out complete sets of structural drawings, construction details, as well as the electrical and mechanical and produces a full specification and cost estimate. With the help of some vizualizations, fly-throughs and a VR walkthrough ArchI has also generated, the client signs off on the spatial arrangements and the external appearance, as they do now. But who’s going to take legal responsibility for the rest? A registered architect isn’t allowed to accept a commission for which they don’t have the resources to complete but does ArchI constitute sufficient resources? Who is going to check all this work the architect didn’t have the skills or time to do. Who will be liable if something goes wrong down the line? I see this leading to a whole ecosystem of subcontracted verifiers who accept partial liability as they do in the system we have now. Liability and professional indemnity insurance will prevent AI from fulfilling the time and labour claims made for it. AI could easily go no further than generating options because architects aren’t going to accept any more liability than they can pay for despite being (presented as) the party with the most to gain from ArchI’s work [“Free ArchI!”]. Should anything go wrong at the level of the design architect’s remit, the architect could of course sue the AI provider and the onus will then be on the architect to prove the AI was fed information within its comprehension. Fault would probably be decided by whether or not similar simulations produce a similar result. A sole practitioner isn’t going to use AI to design a hospital or an airport if they can’t meet the professional indemnity insurance obligations. I don’t see the Bauhaus led separation of design and production disappearing anytime soon. If anything, AI further entrenches it.
• • •