Automatic Design

In February 2016 I wrote about something called associative design and linked to this next video. I wrote that it seemed a genuine attempt to improve things in that its design decisions are shaped by the same variables by which the project and its performance are to be judged and you don’t get much better than that. The only question that remains is whether the chosen variables are the right ones. Nothing much has changed since. Whether those variables are necessary and sufficient is always going to be the question.

This next was only a small gif showing how an adaptive planning algorithm works but it made a bit of a splash on LinkedIn after having been reported on ArchDaily circa 2019. It was meant to show how an adaptive planning algorithm can be used to replan an apartment in accordance with the area and proportions of a space reserved for it. This algorithm is called FINCH and was written for Grasshopper but I’m not sure why we needed to know that. Just as a certain kind of student thinks any SketchUp image or Rhino model must surely be better than a hand drawing, a certain kind of architect seems to be think something is only real once Grasshopper can be made to do it. These people already converted are who adaptive planning algorithms are being promoted to.

I noted how abysmal all the actual layouts were. Each may have zero time or effort to generate but it also looks like it. My first thought was that, before presenting this faux innovation, it might have been a good idea to add a couple of windows and an entrance door, and that a dependence upon artificial lighting and ventilation has been designed into these layouts that are intended to show us what this albgorithm can do.

If it’d just stay still for a minute I’d mark it up but offhand I’d say the flaws of Grasshopper three-dimensional design have been successfully mapped to two-dimensional design. It all depends on what you choose to make a variable, what weighting you give it, and how you choose to link them. Never mind for now that these three weightings will alter at certain thresholds. For example, if the area shrinks too much you might want to re-think the necessity of having a single bed and a sofa. But these and other things such as re-thinking the amounts of kitchen and storage space can all be refined. Since then, the algorithm has surely been tweaked and the program refined but it was presented as if it were proof of concept despite failing to deliver the crucial output of layouts that were efficient, workable and pleasant. In some world that happens to be ours, it must serve somebody’s purpose if it was shoved in our faces as it was and when it was.

Something similar exists in order to “automate” house design, whatever that means. We need to find out what that means. And also what “to design a house means”. There may be situations where it doesn’t matter that much, just as there may be people happy with being given a false choice from preselected options.

And there’s nothing wrong with that. If you visit the Japanese version of the MUJI website you’ll see how it’s done, and without invoking Grasshopper. MUJI produce four types of house with twenty variations of size and proportion. All work well.

Someone has done the work and learned that the original design assumptions and goals aren’t valid for dimensions and proportions outside a certain range. The modular fabrication and construction system adapts to all variations but those variations are, by definition, limited.

MUJI aren’t asking us to be wowed by the labour-saving and cost-savings potential of their technology. It’s not even visible. All they’ve done is systematically apply some rationality to the design inputs of a manufacturing system. I may yet have my life bookended by Christopher Alexander and Patrik Schumacher but I’m not going to spend it debating the merits of humans vs. computers when it comes to solving problems until it’s clear whose problem is being solved.

My third example of automatic non-design is a relatively recent one I saw at the ZHA exhibition in Shanghai in summer. I quote. “The proposal develops housing as micro-communities arranged around distributed shared courtyards. This encourages community living and socialization through more intimate community and connected spaces. Thus, [?!] facilitating a wide range of shared spaces of differing sizes and characteristics in comparison to more ubiquitous, stacked or central courtyard configurations.”

The year was 2018 so this was produced on Patrik Schumacher’s watch. Many persons inhabiting a peripheral apartment will pass through a restaurant, launderette or communal area on the way home but they are the lucky ones who won’t be looking into someone else’s window five metres away.

Those windows are full height glass walls along which the beds are placed à la Nakagin. This level of community living and socialization through more intimate community and connected spaces is just plain bad. Privacy and noise transmission aside, I sense the presentation is pre-emptively defensive about the amount of light in these light wells because the caption to the diagram below right is “Modules exposure verified through solar radiation analysis”. It may be true from this angle if, in our brave new data-driven algorithmic world, yellow appears to mean bright and orange a bit brighter. It’s odd that we’re told this proposal is data driven without being given any data to prove it. I’ll assume the units facing other directions are blue a.k.a. dark, cold and miserable.

Otherwise, the rooms seem modeled on those of Kurokawa’s Nakagin Capsule Tower. Inside the front door is a curious ante-room. At least a typical floor has sufficient fire escape stairwells but having to pass through communal and or open areas to reach them is not ideal, and most likely a breach of code that, were it complied with, would negate the stated premise of the project.

I’m not a total Luddite. In the 1980s I did some work for a Japanese company responsible for some of the first attempts at machine translation. The task of translating technical documents was split into three. Native Japanese speakers edited the source Japanese into simpler constructions and syntax which was then input to a computer and converted and people like me would then reshape it into more natural English. Translation work could therefore be performed by less skilled people who didn’t even need to know both languages. Live interpreters and translators of literature have highly-developed skills and my full admiration but the market for technical translators can’t not have shrunk. I’m glad I’m not a technical translator anymore. I’m not even an architect or senior designer anymore but the main thing that worries – no – offends me about these examples of automatic design is that they’re not very good.

The act of applying one’s knowledge and skill to a problem resistant to reduction and thus to meaningful scripting is challenging and enjoyable. It’s not onerous. It’s what we do. If persons champion algorithmic or associative or otherwise automatic design as the solution to all problems, it’s not because it is the solution to all problems but because they’d like it to be. But why? What’s the attraction of automatic/associative/generative design? What forces are driving it? What need is it fulfilling? And even if it’s not actually fulfilling that need at present, what kind of world is it making us more readily accept in the here and now?

The idea of living on Mars makes us more readily accept that this planet was trashed in the name of unbridled capitalism, that the future exists only for the wealthy and productive, that private enterprise will save humanity … [c.f. Mad for Mars]

The idea of buildings shaped like mountains makes us more ready to accept a world in which mountains are leveled to make way for buildings that look like mountains.

This is already getting out of hand. I don’t know how far MVRDV’s Long Tan Park proposal has come since I first saw these but it’s a bit too close to home for comfort.

Over-publicized technologies such as these may one day produce acceptable designs but, in the meantime, they function to make us more willing to accept a world in which design is seamlessly integrated with a means of production. Given architecture’s track record in pandering to the wet dreams of the construction industry, this is a reasonable suspicion to harbor. Unionized construction workforces with their unreasonable pay demands and their sissy concern for health and safety regulations are being designed out of the industry. It is no accident that the ZHA artifact constructed as a result of their digital thought experiment was 3D printed in gypsum. It couldn’t not have been.

These proposals may be poor but they make excellent propaganda if all it takes is the merest simulation of novelty to keep us happy. One side effect of putting these undercooked proposals in circulation is to debase design and what was once thought a core skill of architects. This hasn’t mattered for a long time now as the main metric for architectural influence continues to be success in manipulating the opinions of an audience of peers. That hasn’t changed. In the end nothing changes. There’s a certain contradiction in automatic design attempting to satisfy base and not-so-base human needs and desires, of (a representation of) rationality claiming to satisfy our various human needs and subjectivities. I can object and say I don’t want anybody’s algorithm on my case but I’d be forgetting it was never about me in the first place.