The Worst Way to Write an Email, But For Buildings
February 20, 2020 11:05 AM   Subscribe

Why Generative Design in architecture is unlikely to succeed.

Daniel Davis provides a detailed takedown, including a lovely GP2 email generator, explaining why complex design processes are difficult to automate.
Will Robots Take My Job agrees.
posted by q*ben (23 comments total) 18 users marked this as a favorite
 
as someone who completely adores "algorithmic smarts", I completely agree
posted by wym at 11:52 AM on February 20, 2020 [1 favorite]


I just noticed that he provides a find and replace tool if you don’t like the term “Generative Design,” which makes me so happy. I’m going with “Hogwash Computerism”.
posted by q*ben at 11:55 AM on February 20, 2020 [1 favorite]


I have horribly uninformed, and half-formed opinions about this.

Like, I feel like mostly, computers are good at search, and mostly that's when we use them for. Search within huge, but finite data sets. The number of configurations on a Go board is incomprehensibly huge, but still finite, and the end goal is pretty easy to define. A talented two year old could write something to look at a go board and decide who won.

Computers are not so good at interpretation. In order to write an email, you have to interpret what it means, have a theory of mind in order to think about the effect it'll have on the other person, and decide how effective it'll be.

GANs throw all this out, and treat it as a search problem. Finding the optimal face doesn't care that it's a face, it's just searching for a number. With the right training set, it'll find something whose components are basically contained in the training set, but it won't go farther than that because it doesn't really know what a face is. There's no meaning, and no interpretation, since that requires assigning meaning.

Like the Chinese room problem, the guy in the box is just following arbitrary rules, and we're prone to paredolia, so we assign the meaning ourselves.

But design really, really requires empathy, and that theory of mind. Generative whatever is cool, but we're doing to ourselves what these beetles are doing to that beer bottle.
posted by fnerg at 12:01 PM on February 20, 2020 [3 favorites]


Also, considering that generative design forces us into the adversarial part of a GAN, this seems especially stupid.
posted by fnerg at 12:02 PM on February 20, 2020 [2 favorites]


My Precious Deleted Comment
×
Hey cortex,

You do not know what I meant. I am just going to go to this guy's website and read it and I will tell you that he can be your friend, and that he has a lot of power over you, and he can destroy you.

Generative MeTa!
posted by GenjiandProust at 12:06 PM on February 20, 2020 [3 favorites]


Architecture seems like the dumbest possible place to try generative design. I could see maybe some element of a building like a door or roof truss where the fitness criteria was easier to nail down, but the whole thing?
posted by BrotherCaine at 12:08 PM on February 20, 2020 [3 favorites]


One of the hardest lessons I've had to learn - and am still learning - is that in almost every case in my job it's faster to do most stuff by hand and automate only the very simplest and most repetitive bits. Circuit board and VLSI design tools have had sophisticated automation for decades. Even with incredibly well-defined goals - use these connector locations, insure no inductive coupling between lines exceeds some specific amount, then minimize vias, area, trace-length, etc - it usually takes longer to fix the automated stuff than just to lay most things out by hand and copy-and-paste them in a sensible way. The software invariably seems to find the one condition that you ignored because it's just obvious that nobody would lay out a circuit that way and then churn out something either deeply flawed or so wacky no human will ever be able to troubleshoot it. Spending twenty hours automating something that would have taken three hours to do by hand is a trap I've fallen into too many times.

I was pretty surprised to learn from the article that they're using this for whole design concepts. I was expecting an article on auto-placing conduits and framing inside walls. That seems hard, but not impossible.
posted by eotvos at 12:17 PM on February 20, 2020 [3 favorites]


“Oops! All Bathrooms!”
posted by Huffy Puffy at 12:17 PM on February 20, 2020 [12 favorites]


The same problems have, more or less, bedeviled generative videogame level design forever. If people aren't using generative tools to create videogame buildings, why the heck would it work for real ones?
posted by BungaDunga at 12:24 PM on February 20, 2020 [6 favorites]


> The same problems have, more or less, bedeviled generative videogame level design forever. If people aren't using generative tools to create videogame buildings, why the heck would it work for real ones?

There has been exactly one game advertised as containing level design: Bungie's Oni. To my knowledge, the approach has not been replicated since. I guess architecture is far too important to be left to architects.
posted by pwnguin at 12:31 PM on February 20, 2020


I had college professors who were working on this sort of thing in the mid '90s. I couldn't have articulated it nearly as well as this article, but it did strike me as an intrigue of the novelty "look what the computer came up with!" far more than anything that was leading somewhere useful. I do think there are possible tools that exist within this realm but on a much smaller scale. For example: a tool that allowed you to define the floor plan limits available for an ADA bathroom and then helped quickly sort through options in terms of door location/swing and whether it was helpful to build out the toilet wall and place the sink in a recess versus expanding the overall room width. But, we already use templates based on prior proven solutions to quickly get to the 90% solved state on these problems and it's hard to imagine a computer getting us much farther any faster.
posted by meinvt at 12:41 PM on February 20, 2020 [2 favorites]


I am probably one of a very small number of people who is trained as an architect and also is a certified SCRUM master so I was sort of intrigued about what would be in the article. Because I was thinking "iterative" for "generative" and those are both interesting concepts. I utilizied my scrum skills in the tech space for building websites and interfaces and then I left the world of tech to go back to the world of building buildings with a focus on residential. I thought it would be interesting to try to use my iterative concept design theories and collaboration skills in architecture but I quickly abandoned it. Ain't nobody got time for that when you are going to be building physical objects that will stand on the earth for potentially hundreds of years! I work in the residential space and when I think about the generative design for houses, like, surely there's only so many ways a living room can exist near a kitchen...but I agree with the author of the paper. The key is in the editing and why bother generating 100 crap designs when you have to assess and edit each one? There's a complex interplay between practicalities of building–materials, codes, zoning, skilled labor–and external factors like terrain and sun pattern and then your clients hopes and dreams combined with your own skills and experience and aesthetic opinions. When I work, I do fast iteration and come up with some truly terrible combinations that I can quickly assess and trash. But it's through that iteration and trashing that I come up with something that works. Having a computer put a bunch of trash in my face doesn't help.
posted by amanda at 1:04 PM on February 20, 2020 [11 favorites]


I just had that gp2 app write a mean letter to Bob Seger. I don’t know why I did this. I am entirely neutral on Bob Seger. But it made my day:

Dear Bob,

You can't make it, you can't do it, you're not doing it anymore. No matter how good you are at your craft, Your songs suck.

Now go ahead and write your lyrics. Don't tell me I don't love you. I don't like you, I don't know you, I don't even want to know you. Don't tell me you don't want to write a song about me. I didn't even make it to "Worst of Both Worlds" when you made your first song.

With gratitude,

q*ben
posted by q*ben at 1:15 PM on February 20, 2020 [8 favorites]


The letter writing component is the most important part of this post. It wrote a letter to hank from clarise on the subject what is wrong is wrong all the way through you that was just "I don't care if you have a bad day" 15 times.
posted by clockwork at 1:36 PM on February 20, 2020 [1 favorite]


The same problems have, more or less, bedeviled generative videogame level design forever. If people aren't using generative tools to create videogame buildings, why the heck would it work for real ones?

Skyrim seems pretty successful and most of it seemed pretty generated, and the data they fed the algorithm was "Morrowind" with this goal of "make this, but worse -- and prettier." They didn't use it on buildings, they just made one or two of those to copy+paste, but dungeons and crap were just dungeon chunks fitted together with quests for generated items.
posted by GoblinHoney at 2:31 PM on February 20, 2020


From the F.A.: The fact of the matter is: one hundred shitty designs aren’t anywhere equivalent to one considered design. If your software was any good, it’d produce fewer designs, not more.
posted by signal at 6:11 PM on February 20, 2020 [1 favorite]


I am probably one of a very small number of people who is trained as an architect and also is a certified SCRUM master

Amanda — there are more of us than you might think... even here on the blue.

Amyway, generative design, as it stands now, seems to be little more than a mechanism to transfer wealth from architecture firms to software companies.
posted by turbowombat at 7:29 PM on February 20, 2020


as far as I know Skyrim dungeons were handmade
posted by BungaDunga at 7:42 PM on February 20, 2020


I think the "generate a bunch of options and pick the most interesting" method can work, with the caveats that you are also creating / iterating on the generator itself (that is, you aren't simply a consumer of a generator someone else wrote, but can actively build your intent into the generator itself), and that you use the generator output as a starting point rather than as a final result.
posted by Pyry at 7:46 PM on February 20, 2020


Have to agree that the email generator in this article is hugely fun, in a surreal and musical way. The email subject line I use most often at my day job is "Materials received" (usually as in, you sent us the wrong stuff and this is what was wrong), so I put that in, and one of the options it came up with was this (repeated lines indicated for brevity):

"I've been thinking about Materials received. (x34)

Warm wishes,"

This quite emotional one is from the script to a one-act play written and performed by high school sophomores:

"Hey Robert,

What's up, what you're doing here is not your fault. I'm not trying to be rude. I'm just trying to let you know that I am okay with it.

I'm glad.

Why would you even let me talk about Materials here? I don't understand, what do you mean. I'm not talking about Materials! No matter what, I know that I'm a little bit different.

I'm just in this part, and so if I'm not careful, it will get worse.

This is something I can't get away from.

But what do you mean by "nothing"?

I can only be a little bit better.

Many thanks,"

I for one am swayed by the argument that just as this generator is entertaining but can't actually efficiently create an email I could send to someone who has mailed my office the wrong stuff, an architect might find something creatively provoking in trying to assess the output of generative design, but not a tangible usable output.
posted by C. K. Dexter Haven at 8:32 PM on February 20, 2020 [1 favorite]


Dear Metafilter,

I'm a software architect, and I have been a software architect for over 20 years. I've worked for Microsoft, Apple, and Google. I was one of the original programmers for the Windows operating system. I'm a software architect for many companies. I'm an engineer. I'm a software engineer for many people. I'm a software engineer for many companies. I'm an engineer for many people. I'm an engineer for many people.

Thank you,

romanb
posted by romanb at 5:50 AM on February 21, 2020 [1 favorite]


I'm a software architect, and I have been a software architect for over 20 years. I've worked for Microsoft, Apple, and Google. I was one of the original programmers for the Windows operating system. I'm a software architect for many companies. I'm an engineer. I'm a software engineer for many people. I'm a software engineer for many companies. I'm an engineer for many people. I'm an engineer for many people.

With a little bit of work, these can be the new lyrics for Guided by Voices' I am a Scientist
posted by The_Vegetables at 8:02 AM on February 21, 2020 [3 favorites]


short and sweet but it works:
Are you there?

Hello God,

Respectfully,

Margaret
posted by Kabanos at 9:54 AM on February 24, 2020 [1 favorite]


« Older "Well, no, I’m sad, and I want to make you sad...   |   "I can’t believe I’m being gaslighted by a room... Newer »


This thread has been archived and is closed to new comments