A colleague noted that he had been asked to speak at a conference as a panel member, on the subject of AI in architecture. He expressed a few misgivings, saying that he had some reservations as he had little experience of AI in practice, having retired a few years ago. Matters these days do change quickly. He said that he would only accept the position if he had something to say.
One wondered about this. We all have something to say about AI as it creeps into everyday life in a subtle and almost unnoticed manner. Simple Google searches now involve AI’s Gemini with responses that can be seen to be somewhat generic and disinterested, obviously gathering sundry information and presenting it as though it might be learnedly authoritative: such is AI. Sometimes one gets useful information; other times one gets garbled nonsense; but how might AI have an impact on architecture?
One knows how computers can be useful for the usual mechanical tasks that they are capable of, like searching, copying, itemising, and for similar challenges like schedules and filing that one uses in architecture; but it is in presentation and design that one first confronts the presence of AI. Practice has grappled with a range of CAD developments that started years ago as simple drafting, and now has become 3D documentation allowing an enormous complexity to be described and detailed.
This has led to documentation that becomes the contractor’s tool, with the creation of a seamless link between the architect and the builder. As matters have developed, AI has crept into this world that already surprises even the creator of the file, offering opportunities for amazement beyond what anyone might have comprehended as being possible. One notices that AI slowly takes over as the master, remaining in control by having others act according to the rules it works to; and as the slave, by responding to simple verbal and written instructions to manage complex matters whenever these directions are given. One is left in an ambiguous situation of initiating processes that one has little control over.
So it is that one can not only ask AI to prepare a report on such and such, e.g., but one can also direct it to design whatever is described, and fine-tune this output with further directions, with each step offering slick, complete solutions for consideration arising from the digital void. These are drafts only in the sense that the instigator, the architect, might consider them to be interim steps; stages of development; but they all remain tempting final possibilities. One might ask AI to design a house in the style of Wright, or Corb, or whoever, (see: https://voussoirs.blogspot.com/2023/10/cosplay-in-architecture.html), and then further refine the outputs with more instructions, e.g. by asking for the scheme to be less bold, or smaller and less grand, or whatever: e.g., make it blue, and it will. The refinement can go on until one gets an image that one feels ‘happy’ with. It is something like a lucky dip, with the architect taking as many ‘dips’ as he wants until he gets the prize he prefers, like a spoilt child. AI has an impact on how people act and how they see themselves – something like self-important heroes; god-like figures given apparent total control by AI, a perception which, on reflection, fades away as a misguided illusion.
The possibilities are endless. AI is becoming increasingly powerful, capable of producing complex visuals under direction, to request, in a matter of minutes, with a precision and variety of expression that empowers the fool and the genius without any discrepancy. AI is simply mathematics; the intricate matching of numbers and sequences, code that gives the ‘answer’ to the calculation it has been set. Just as ‘one plus one’ will give the answer ‘two’ on the calculator, the direction: ‘design a beach house,’ will give a beach house of some styling. After seeing this answer, one might be able to understand what is wanted or not wanted, and continue with as many steps as one desires to achieve the preferred design. They could be called the ‘Goldilocks steps’: not too big; not too small; not too . . . , etc., and so the outcome is varied to respond just as though it was: ‘now try four plus three’ when one is searching for an appropriate ‘seven’: no, how about ‘twenty-nine minus twenty-two’?
The concern with AI is one that already exists with 3D CAD: its ability to amaze the operator. Each step in AI provides a perfect graphic submission that can be altered to be a ‘hand sketch,’ a video, a water colour rendering, or whatever might be available as a programmed option achievable at the push of a button, leaving folk literally gobsmacked; astonished at the ‘mindless,’ instant outcome – ‘their work.’ The result has literally come from nowhere, just with a bit of mumbling as a maybe instruction, with no roots other than the thought as words, and the mathematics the response has been structured with. Its mystery is like magic, with schemes appearing complete before one’s very eyes: “WOW!”
The ‘learning’ process that has been adopted for AI is the same as that of the human brain: one tries, stumbles, and finally, after much corrective repetition, becomes fluent and skilful in the tasks – and then an expert. The more complex matters are developed and managed likewise, as an aggregation of simple steps that are intertwined and layered to offer an integrated awareness of a situation. AI manages things similarly in text and graphics. There is no fault here. The mechanism has been adopted and implemented, and it works well. The complication comes with other issues, some rather basic matters, and some other very subtle concerns.
The accuracy of the outcome is the most predictable problem that can be checked and corrected. All results are presented with a supreme confidence and sophistication that encourages one to see them as ‘the gospel,’ the 'right' answer, totally accurate and all-knowing, even when the whole might be an inappropriate, fanciful fabrication. One has to remain vigilant.
Then there are the more subtle matters. The human brain and its mode of operation might be able to be replicated and ‘improved’ upon by AI, but it never works alone. This human brain, (not the digital ‘brain’ we sometimes refer to as ‘thinking’), is not a simple machine that works in isolation; there are feelings, senses, assessments, reviews, judgements; flesh and blood – the body flexing rhythmically – that are involved in the brain’s operation. There is other 'knowledge' that is involved, always more than can be explained, giving moral guidance and personal responsibility a place in outcomes; shaping their role and relevance. AI creates from a void, a book of rules defined by the coding. It is a complex version of colouring by numbers on auto that is then assessed by the human brain for its degree of satisfaction that sees difference and surprise as enduring positives.
Both these situations demand that if AI is to have a responsible role in architecture, it needs to be managed as a tool, not treated as the ‘messiah.’ It has to be constantly questioned and challenged by a living, breathing integrity in order to shape meaningful outcomes. It has to be treated as the lesser gadget that can mislead and conjure up hopeless, irresponsible, perhaps alarming visions, rather than being seen as giving unforeseen, 'genius,' (c.f. genie as in mystery and magic), results. Here one thinks of that relationship between the craftsman and his work, where his beliefs and attitudes have an impact on the outcome, with these matters lingering in the thing itself, to be sensed as that known unknown. Given this, one can envisage how important the involvement of the architect and the personal rigour that comes with the individual is in the designing of a building.
We are the ones who are making AI what it is. It can be whatever we want it to be. It can be the prompting guide hinting at possibilities that need to be tested as scientific theories might be, even discarded, or the bold director of the designer, taking over completely, turning the architect into an irrelevant slave, a nobody who can be the anybody who is giving the instructions.
One thinks of the report on AI and aboriginal art. The outputs are astonishing and believable, but they come with no cultural roots; no social or historical provenance: they just are ‘lookalikes’ – the outcomes of calculations and rules; nothing more: dare one liken them to fakes? They present nothing emotionally; they hold no power other than what we might choose, in our naivety, to give them: they have no inherent value other than saying that they have been compiled from sources Aboriginal.
Here lies the issue with architecture; we could end up with a valueless architecture, with nothing to support the soul; zero Zen wholeness: merely the game of make believe; of let’s pretend we are real – mock-ups, as in theatre sets shaped for flimsy entertainment: perhaps we have it already? – see; https://voussoirs.blogspot.com/2017/01/herzog-de-meuron-architecture-with-no.html. This is a serious matter. Being needs substance to be; essence. AI is so persuasive in everything it does; it is up to us to let ourselves be fooled by it or not.
We seem to have granted AI a particular place in our understanding. While we describe the use of a computer in architectural work as CAD – computer-aided design; perhaps ‘drafting’ might fit too – we do not refer to the use of the computer to respond to our questions as CAR – computer-aided responses or replies – but as AI: artificial ‘intelligence’; nor do we say that the architectural presentations and designs are not CAP – computer-aided perspectives or possibilities – but as AI in architecture: artificial intelligence, giving AI an unwarranted stature and sense, in the same way as we do when we say that a computer is ‘thinking.’ We assign intelligence to the computer, the very same piece of equipment that we use for ordinary searches, schedules, and writing, with the only difference being the programming. From the basic programs that allowed us to manage 0 and 1, then other numbers and letters, we developed programmes that could help with maths and spelling; then grammar; writing; then essays; etc.; then drawings; then design: all as the complexity of the programming developed, .01; .02; .03; etc., and the finer details and possibilities were compiled, assembled, and manipulated.
It is a situation that we cannot dismiss or criticise, for it is how we learn too. How could this be a problem? The difference is not the mechanism, it is the context. One ‘mechanism’ comes as a part of the operation of a living body, with a thinking and feeling that embodies, literally, consideration for another; for context, empathy, meaning; ethics; the other is a calculation manipulating the numbers that have been entered into the digital system that responds to the input to provide the 'answer.'
No doubt the argument can be made that a computer can be programmed to accommodate matters like symbolism and the like if the software is appropriately assembled with more and more subtle layers and versions. It may even embody ‘meaning’ itself when this has been shaped as a set of rules; but it seems that there will there always be more. Tradition says that if things ephemeral and enriching could be told, they would have been. So, can meaning be programmed? It seems unlikely, even if we might try, leaving us with the quandary of human involvement and its unknowns that machines seek to match.
AI gives some less subtle results too.
Aboriginal art best highlights this situation that remains difficult to articulate, being unsatisfactory in spite of the stunning outputs by AI when asked to present an Aboriginal painting.# Cultural relevance and intimate meaning involve something deeply relevant and personal that leaves us questioning AI ‘Aboriginal’ art. We can admire the effort, and be cajoled, but remain outraged at the sheer insult that we are expected to see this art as a style that can be replicated by anyone, anywhere; by any machine, in any place. We have to remember that computers compute; that human involvement carries more that we need to respect even if we cannot understand what this might be. Here lies the problem that can always be quashed by the debater's logic. Art, like life itself, is not the outcome of a rational dictionary reference, an analysis, or a calculation.
The craftsman of old knew that his very being was involved in the making - 'having concentrated, he set to work' - and that this essence came to be embodied in the outcome to engage and involve others in this mystery, this commitment. Once we forget this, or think the notion fanciful, as we do today, then it makes perfect sense to envisage the computer creating ‘meaningful’ aboriginal or any art - anything at all, even architecture.
Already ZHA uses AI to find, to go fishing for, an ‘interesting’ form for most projects – to see what comes up - and then to go for it.* It sounds just like a game of ‘pick a box’ - and we consider this intelligent? The process would have been mocked in earlier times when ideas were rooted in ideals, ideas, and theories; when there were no computers to play with to leave us astonished and amazed, distracted from the real state of affairs – that we are dealing with a programmed machine.
We are in danger of becoming blind outsiders, occupying a blank space while manipulating the digital void to develop some different visuals – an architecture of surprise appearances. Yet, even being aware of this situation, the argument for the benefits of AI continue to be put with much force and determination, such is its guile. Perhaps we should not be surprised, for, as René Guénon has noted, we are in an age of quantity that concerns itself only with what can be measured, or programmed, with our search for meaning lingering in the amazement of progress, revealed in the bespoke differences that AI can generate with things 'futuristic.' We need to think more carefully about how we manage AI rather than delight in its wonderment.
#
*
https://www.dezeen.com/2023/04/26/zaha-hadid-architects-patrik-schumacher-ai-dalle-midjourney/
Some articles to ponder:
https://www.dezeen.com/2025/08/14/seraphinne-vallora-interview-andreea-petrescu-ai-imaging/
https://share.google/cMRTKtLxLtvOvjWVG
https://share.google/EU4DLqZNHrTQ88d0X
9 SEPT 25
NOTE
One could liken AI in architecture to cruise control in a vehicle. This system cleverly keeps the speed automatically, distancing the vehicle from slower traffic, but it knows nothing of the road conditions, and can become dangerous on rough surfaces, on certain bends, on some climbs, and in various densities of traffic, a situation that is only aggravated by any combination of these factors. This computer-aided driving lacks a subtlety in its response that only manages the mathematics it has been structured with.
.jpg)
.jpg)
.jpg)

.jpg)
.jpg)
.jpg)
.jpg)
.jpg)


.jpg)
.jpg)

.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)

.jpg)
.jpg)
.jpg)
.jpg)
.jpg)


.jpg)
.jpg)
.jpg)
.jpg)
.jpg)


.jpg)

.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
No comments:
Post a Comment
Note: only a member of this blog may post a comment.