Artificial Intelligence, as a discipline, has already been permeating countless fields, bringing means and methods to previously unresolved challenges, across industries. The advent of AI in Architecture is still in its early days but offers promising results. More than a mere opportunity, such potential represents for us a major step ahead, about to reshape the architectural discipline. — Towards Data Science
Stanislas Chaillou, a Master's candidate in Architecture and Fulbright fellow at the Harvard Graduate School of Design, believes that artificial intelligence can offer in-depth analysis and alternative strategies to the design of floor plans.
Chaillou argues that higher degrees of optimization, contextualization and classification are just a few of the benefits of applying artificial intelligence to the design of floor plans. "Generative Adversarial Neural Networks — or GANs- are here our weapon of choice," Chaillou writes. "Within the field of AI, Neural Networks stands as a key field of investigation. The creative ability of such models has been recently evidenced, through the advent of Generative Adversarial Neural Networks. As any machine-learning model, GANs learn statistically significant phenomena among data presented to them."
We have already seen design software - such as Revit, BIM and CATIA - simplify the task of detailing (typically dealt with in the Construction Design process, but what Chaillou and other architects influenced by the AI movement offer is an ease and verifiability of the Schematic Design process. "Far from thinking about AI as the new dogma in Architecture," Chaillou concludes, "we conceive this field as a new challenge, full of potential, and promises. We see here the possibility for rich results, that will complement our practice and address some blind spots of our discipline."
@miles, design or architecture are and must be a “stylistic” respond to specific conditions. If you take away the “stylistic” part, then you are talking about mere engineering. When I talk about style I am not referring to trends or architectural movements, but of my own or your own “signature” that takes quite a few years to develop.
And if you read the article, the particular subset of AI called GANs is way different and advanced than power tools.
And to be honest (unless you work for Herzong and de Meuron that consider themselves “design consultants”), right now I (we) spent 80 percent of our time drawing stupid/bland stuff like making sure there is 800mm between the stove and the sink. Or how many desks we can fit. Or tagging 1000 pieces of ceilings. Or numbering rooms. Or calculating stairs to a basement will only be used once a year. Or calculating gross areas. The time dedicated to actually conceptualising and designing is minimum if we are honest. All of this I just mentioned can be automised with linear-descent type of AI. The architects at WeWork are doing it. Look for Andrew Heumann on YouTube.
But GANs are quite something else. They are more like breeders of ideas.
My feeling is that ignorance is making many of us afraid of the new.
All 17 Comments
As a practicing architect of + 20 years. I have naturally given much thought to an AI based approaches to design. At every turn I keep arriving at a similar conclusion, it may be appropriate for making a "building" but it cannot make "architecture". Finally there is a dead end to what optimization can produce. NASA has tested self contained food/ waste loop with very little loss of nutrients in each cycle. All it asks of the astronauts is that they only eat a diet of algae.
Is our greatest aspiration the architectural equivalent of an all algae diet?
This line of thinking emerges from the belief that every aspect of life can be quantified and therefore optimized. This line of reasoning is as misleading as it is vapid. Could someone actually say with total sincerity that The Odessy is 37% better that Naked Lunch?
This self legitimating approach displays an deep lack of confidence in our profession. We are so much more that what we can measure.
Let's step away from the numbers, read novels, and make architecture worthy of our humanity.
We don't need artificial idiocy, there is already a vast surplus of the natural kind.
Google Flux attempted to do the same when it first launched - create an app that produces an optimized building (complete with MEP, envelope, structure) based on developer inputs. They gave up on that software though, in favor of a Grasshopper/Dynamo plugin.
From a developer's point of view, it really is a number's game. What vastly complicates matters are zoning, code, and coordination - which stands in the way of most optimization attempts that deal with the whole building. I'd compare a building more to, say, an aviation project (Developing the A380 or F35 JSF, for instance) in terms of sheer complexity. While optimization is a critical process of those kinds of project engineering, it is also just one tool amongst many.
There is much potential for this in the realm of interior design. And already we see WeWork doing exactly this in their offices. Inadvertently, in their relentless quest to quantify and optimize, the aesthetic converged to that Instagram-friendly, neutrally chic, inoffensively cool style, like so many interiors of today. Yet that appears to be what the market "wants".
All good points. But should we design buildings only for the Instagram moment when the lived experience of them vastly different? I really wrestle with this as I was trained to focus on the " money shot" at several good firms. It does make for a successful practice but it lacks something essential. A Carlo Scarpa building can be difficult to capture in a money shot but being in the space is otherworldly!
I suppose the bigger question remains.... how do we lead the market to better, more inspiring work?
That’s the irony of architectural representation isn’t it? That the audience vastly exceeds the number of actual users and that we seldom get to experience a space in its totality. Regarding leading the market to better aesthetics - that is indeed the value add of designers. To the client, the big factor is risk - they just don’t know what to expect in terms of concept and even more importantly, execution. Like so many big investments of resources and time, risk management is key. In industries that allows risk takers to reap economies of scale, innovation can be lucrative in the long run. But with most building design being one offs, such rewards usually accrue only to retail, hospitality and now office designs too - where one design, done right, can be rolled out worldwide. Unfortunately this leads to yet another standardization of design style - so aptly naked Airspace in an article commenting on the universal banality of today’s photogenic interiors.
" so aptly naked Airspace"
I don't follow, please explain :)
My bad! Here’s the article where I first read of the term “Airspace”
https://www.theverge.com/2016/8/3/12325104/airbnb-aesthetic-global-minimalism-startup-gentrification
You could say it’s the inevitable design choice if the tech world has its say! How different this is from the styles of yore and whether or not it is fundamentally bad are tough questions. For me, there has to be truth in art and design - the designer should not lie about the intent nor effect of the work. If a designer genuinely digs this style then so be it - but faking authenticity seems disingenuous. An AI designer though ...
The images above are simply thoughtless graphic morphing of one nonsense into another.
A computer can beat a human at chess because it can analyze millions of possible moves and countermoves towards a single goal (checkmate). A computer will never design The Sydney Opera House or Sistine Chapel.
ALEXA orders up porn because it misinterprets the command of a six-year old. People drive into trees because the GPS says "turn right here". People become dependent on devices that do not require them to write (or even read) and lose the ability to perform those basic functions. Computers do not make people smarter, in fact they amplify their stupidity.
To err is human, to really fuck things up requires a computer.
I totally concur. But a worse consequence is that it will continue to make is less skilled and intelligent. Awhile back I was at an architectural technology seminar here in NYC surrounded by some the the best in our profession. One unnamed leader of these firms was touting how their stair automation tool will ensure that no young architect would have to spend hours making and updating stair drawings... " Isn't this wonderful?"
To me it is a nightmare, how can anyone learn deeply our craft when we automate all the opportunities for learning? It's staff destruction in serivce of making fees cheaper. If unchecked, there will be no future architects, only stylists.
the craft turns to computation skills. If you can make a quick script in rhino/grasshopper that allows us to optimize tread height for a tall-ish building, with uneven floor heights (because its mixed use), then you are ready for the modern world. The way we used to do this was trial and error. It was painful. Nothing is learned from it and we missed better solutions because it was too much work. It is too easy to settle for less-bad answers. Or just standardize everything, which is I guess how it was in the bad old days before computational design became standard office skills.
A simple, fundamental mathematical operation (division) calculates stair risers. Different floor height require individual calculations.
Not exactly rocket science. Especially considering we put men on the moon with slide rules.
Technology is as neutral as its user. Also “Division” is not architecture… neither is style
The original article/paper says clearly that the algorythm can be led or directed to learn different styles. That is, we can potentially personalize them. Yes, the tool will fix the mundane things, but the final output (the artistic part) will be control by ourselves. That is a massive potential laying there.
@libnypacheco,
By your definiton, are architects only responsible for and motivated by "the artistic part"? If so it seems a narrow and increasingly fragile definition of the profession. If we believe that AI can do" the mundane stuff" and be a tool to create the personal stylings of the client on the skin of the building, we just made a software package that puts us all out of work.
If however our definition of architecture has a much broader scope, we can claim our previous position as master builders. Louis Kahn, Kevin Roche, Walter Netsch and many other masters' works explored design as a comprehensive consideration of all systems. They integrated the disparate challenges of building to make a whole that is greater than the parts. AI can make a collection of very efficient parts with cute skin but it not a comprehensive whole, it is not architecture. It could never make the Salk Institute. It lacks the capacity of that which makes us human, it lacks imagination.
https://www.youtube.com/watch?v=2oFMgJeK2cA
Not AI, but slowly trying to make sure we pay as few humans as possible while cloaking it as a benefit.
I have a computer with a mouse. I am an architect.
Human labor will be completely replaced by AI eventually...100-1000 years... Seems sad but inevitable... Art will probably never/last be replaced...but that could be wrong...
@franciscooke, I am with you regarding the reduction of the architect’s work to design pretty, exciting, inspiring skins. You are right that we’ve lost autonomy because we let others took parts of our job (aka consultants). In our firm we are working right now with BIM on the cloud with the others consultants loading their own 3D files. Interestingly enough we are always the ones leading and being ahead of all the parties. We set the start. This is happening in three different projects the clients asked us to use this cloud thing. What I am saying is, regardless of how many consultants and how much of the work we used to be in charge of, right now the architects are the ones leading projects. Design still trumps. We might not be the masters of the old days. But we are still in charge.
Now, coming back to the AI. When I talk about style, I am not talking about a particular look you search for to please a client. I am talking about the style each architect develops during her/his career. The lines, the curves, the dispositions and sequences of space. That “style” is what we can transfer to the AI. We will train them to learn our style. And then it will start giving us proposals that we can discard, choose, mix, but they will always be ours at the end of the day. The “style” transfer tool explained in the article just reveals that possibility I am talking about.
Kevin Kelly has taught us that technology is very much a tool we create to transfer some cognitive tasks to a device so we can use our brain to think above that. A pencil and a paper was a way to transfer knowledge so we didn’t have to remember every single bit of information. A calculator was the same. Enciclopedias were also a transfer of knowledge so we could go “lighter” and focus on things in a higher cognitive level.
Finally, instead of trying to recover some knowledge of the old days, or try to recover autonomy by taking back the work we’ve given to consultants, we should take AI and transfer even more to it so we can operate in a higher level where we can produce even more impressive and affecting spaces. Or produce art, as defined by Deleuze: a block of percepts and affects, bypassing the senses directly into the nervous system, that is, make people see/experience the world in a new way than they had before entering one of our buildings.
I am glad that you work in s place that has the spirit to drive the process. All good offices take this approach. This is what I mean when I say master Craftsman in a modern context, perhaps "ringleader" would have been more apt.
Your proposal for how to use AI in your process I find interesting. It's kind of a William S Burroughs Cut Up technique. In writing about cut ups, he said their true power lies in that they reveal what you know but don't know that you know.
The key to this is using this process to trigger your creativity and not accept the outcome as the inevitable, perfect, sacred computer-god ordained solution. My fear is that AI will be treated exactly this way and it's golem like revenge will not bode well for us. Editorial control is key.
Style is irrelevent. Design is a process that responds to conditions and programmatic requirements. Creativity in design is largely dependent on an intimate and detailed knowledge of processes and conditions, hiding these behind some automated process does not free one's creative process, it cripples it.
Reminds me of the carpenter who can only use power tools. The nature of the material is no longer relevant, and a simple cut with a hand saw is beyond his ability.
@miles, design or architecture are and must be a “stylistic” respond to specific conditions. If you take away the “stylistic” part, then you are talking about mere engineering. When I talk about style I am not referring to trends or architectural movements, but of my own or your own “signature” that takes quite a few years to develop.
And if you read the article, the particular subset of AI called GANs is way different and advanced than power tools.
And to be honest (unless you work for Herzong and de Meuron that consider themselves “design consultants”), right now I (we) spent 80 percent of our time drawing stupid/bland stuff like making sure there is 800mm between the stove and the sink. Or how many desks we can fit. Or tagging 1000 pieces of ceilings. Or numbering rooms. Or calculating stairs to a basement will only be used once a year. Or calculating gross areas. The time dedicated to actually conceptualising and designing is minimum if we are honest. All of this I just mentioned can be automised with linear-descent type of AI. The architects at WeWork are doing it. Look for Andrew Heumann on YouTube.
But GANs are quite something else. They are more like breeders of ideas.
My feeling is that ignorance is making many of us afraid of the new.
If you have trouble "breeding" ideas you should be in a different line of work. As for "style", you need to consult a dictionary. As for "new", it isn't necessarily better and is often worse.
Miles, some of the reflections regarding the “breeding” of objects/events/ideas, check Manuel Delanda’s articles on genetic algorithms in architecture: https://www.cddc.vt.edu/host/delanda/pages/algorithm.htm
As for the definition of style I am referring to, please check Deleuze’s: https://muse.jhu.edu/article/179424
You are welcome to go as far down the rabbit hole as you like, I prefer to work in reality.
Besos y abrazos, Miles
This:
"As any machine-learning model, GANs learn statistically significant phenomena among data presented to them."
This is one of the most important summaries about what and how GANs actually do. Already this is why architects should pay attention to GAN generated vision. To find the overseen tendencies, to detect the dynamics, which are crucial for human, but probably aren't still visualized. We shouldn't idolize AI, but it can be a great assistant for humans, if meaningful used. Anti-AI tendencies and attitudes are childish, I'm sorry.
GANs are just the latest greatest bullshit in AI. There is no intelligence here, just algorithms. You should try to understand the difference. Practical use in astronomical image analysis has people concerned about deepfakes in art.
Wait a couple of minutes and there will be another new piece of magical code that will transform the world.
Main premise formulating arguments for data input and learning "(1) the creation of building plans is a non-trivial technical challenge, although encompassing standard optimization technics, and (2) the design of space is a sequential process, requiring successive design steps across different scales (urban scale, building scale, unit scale)" already makes unfounded architectural design decisions. These decisions, guised as "intuition," highly restrict the development of architecture and ignore vast possibilities of architecture. This research is a good precedent to further emphasize many architecture schools' lack of interior architecture coverage and societies' deep infatuation with the exteriors, to the fault of categorizing the interior as a technical challenge. Weakly trained algorithms are good for organizing one's closet or optimizing sewage infrastructure, not yet thorough enough to design better inhabitable spaces.
"...design software - such as Revit, BIM and CATIA - (is already) simplify(ing) the task of detailing... "
as much as cad did... it's still copy pasta from the last job
Simplification =/= improvement.
I would argue every step has been away from simplification. Cad doesn't really simplify detailing over hand drawing. It makes it more efficient perhaps. I would argue all day that the same can be said for BIM if you are doing it right. I work on projects where people treat Revit like cad, and it is actually making it more difficult, because it doesn't do cad quite as well as cad, shocking. When projects are actually set up to do detailing in a "BIM" way, it can improve efficiency.
Block this user
Are you sure you want to block this user and hide all related comments throughout the site?
Archinect
This is your first comment on Archinect. Your comment will be visible once approved.