The Humans are Dead: AI in Architecture

Can AI transform the architecture industry? Join us on this episode of the Design Principles Podcast as we unravel the dynamic interplay between artificial intelligence (AI) and architectural design. We’ll kick off by breaking down AI into digestible bits, distinguishing it from machine learning and discussing the elusive concept of artificial general intelligence (AGI). Discover how tools like ChatGPT are not just boosting efficiency but also pushing the boundaries of creativity in the design process. We’ll also tackle ethical concerns and real-world applications, pondering the profound impact AI could have on the way architects and designers work.

Is AI a creative ally or a potential threat to unique architectural visions? Gerard joins us to express his concerns about AI diluting originality, sparking a lively debate on the moral responsibilities of architects using these advanced tools. We’ll dissect practical applications like compliance tasks and quick concept renders, while wrestling with the unpredictability AI introduces into design work. The focus remains on striking a balance between leveraging AI for efficiency and safeguarding the creative integrity that makes architecture an art form.

Looking ahead, we investigate AI’s broader role in architecture, from automating detailed drawings to managing projects. Despite AI’s impressive capabilities, we highlight its limitations, especially its inability to replace the innovative spark that human designers bring to the table. The discussion zooms in on how AI’s reliance on existing data might stifle originality and the crucial need for architects to balance innovation with practical constraints. As we envision the future of AI-driven architectural firms, we'll reflect on how the roles of professionals may evolve in a landscape increasingly dominated by AI, ensuring a thought-provoking conversation on the future of architecture.

Please note the introduction to this episode and this summary have been AI generated to fit with the theme of this episode.

0:00 - The Intersection of AI and Architecture
15:30 - AI Ethics in Architecture
30:01 - The Role of AI in Architecture
46:46 - Exploring AI Applications in Architecture

Follow us on @designpriciplespod on Instagram.
If you wish to contact us hit our DMs or email us on info@designprinciplespod.com

Transcript

Sam Brown: 0:00

Welcome back to the Design Principles Podcast. We thought we'd kick things off this week in a meta way by having the episode introduced by an AI-generated and decidedly British, sam Brown. I hope you enjoy. Welcome back to another exciting episode of the Design Principles Pod, where we delve into the latest trends, ideas and innovations shaping the world of architecture and design. Firstly, we are 10 episodes in and want to thank all of our listeners for the encouraging feedback and continuing support.

Sam Brown: 0:39

I am the AI impersonation of Sam from Aret Architects, and today we return to our original format a bit of free-flowing and sometimes heated discussion between the three of us. Why has an AI version of me introduced this episode? Well, today we're navigating the fascinating and rapidly evolving intersection of artificial intelligence and the creative realms of architecture and design. As we dive into our discussion, we'll explore how AI is transforming the design process from being a powerful and maybe questionable design generator to acting as a sophisticated assistant or caddy that enhances creativity and efficiency. We'll also tackle the ethical concerns that come with AI's growing presence in the industry, ensuring a balanced view of its potential and pitfalls. Lots to cover, so, without further ado, let's dive in. So, guys, welcome AIA Artificial Intelligence.

Gerard Dombroski: 2:05

Architectural Institution.

Sam Brown: 2:07

AI, artificial intelligence.

Gerard Dombroski: 2:10

Architectural institution, the AI.

Sam Brown: 2:12

AI, the Architectural Institute of Artificial Intelligence. How does AI and architecture integrate? Funny that we're only into episode 11 and we're already on this topic, given my hot take at the start of the year in episode one was that ai would stop being relevant. Yeah, good luck.

Ben Sutherland: 2:34

Yeah, I might have been maybe it was a bit wishful time yeah, that's true to be fair though I have, I use it less now than I did when we recorded that first podcast. For sure Interesting.

Sam Brown: 2:49

I also think it's very fitting that we're following on from our last episode with Chris, particularly his discussion around the digital versus like, the analog versus the digital, and here we are talking about the anti-analog really.

Gerard Dombroski: 3:03

Full dig, full Digimon. Yeah, talking about the anti-analog.

Ben Sutherland: 3:06

Really full dig, full digibond. Yeah, yeah, I was. I was moments away from asking chris what he thought of ai, but I just I didn't want to I didn't want to go there.

Sam Brown: 3:16

So we only have about an hour for a podcast, right, not two. I think we'll get it back and we'll ask him again yeah, I'm, plus I'm still learning myself.

Ben Sutherland: 3:26

So there you go. So what does everyone know about AI?

Sam Brown: 3:31

well, I thought I'd just kick us off with giving a wee overview of what tools there are available, particularly around architecture and AI.

Gerard Dombroski: 3:41

Is it worth giving back another little step? And just trying to summarise what the fuck AI is.

Sam Brown: 3:48

Can you? I mean, it's artificial intelligence, right?

Ben Sutherland: 3:54

Yeah, so I can do like a really Go on, go on, slytherin, inaccurate, very inaccurate.

Ben Sutherland: 3:59

We're all making this up Basically like what my take on it and what I know about it, um, but basically, ai essentially has been around for quite a while now, and what has really only changed is is our understanding or our interaction with ai, which is, um, what happened when ChatGPT 3 was released by OpenAI, and that was the biggest movement. There was the change to the large language model, and that's pretty significant because it allows us to interact with AI in a different manner as opposed to previously our understanding of AI. What has been around, like companies like Google and that have been using machine learning for a long time now.

Gerard Dombroski: 4:55

Yeah, so isn't AI fundamentally machine learning?

Sam Brown: 5:00

that is a technically isn't that is a segment of AI.

Gerard Dombroski: 5:03

Like it's not its own organism floating around.

Ben Sutherland: 5:07

So that's a huge separate thing is like AGI, which is artificial general intelligence, and that's when it is, you know, the Terminator. It's got its own mind, it does its own thing. At the moment, all we're doing is, like sophisticated machine learning or algorithms, that you input data and you tell it kind of what to look for and you train it through a series of training processes and it outputs what you want it to output within a certain parameter. So, for example, large language models. It basically takes a lot of data books, internet, anything like that. It understands how humans interact with that data, what the language is, and then it basically allows us to, it performs tasks and allows us to communicate with it through our own language, I guess.

Gerard Dombroski: 6:18

Essentially, you feed it information, then it sort of regurgitates back what's within that pool of information. You feed it information, then it sort of regurgitates back what's within that pool of information you feed it. It can only regurgitate back.

Sam Brown: 6:29

what's in that pool? You can't think for itself organically.

Gerard Dombroski: 6:31

yet it's essentially sped up Googling and desalinating of what's in that little swimming pool of ideas. I just think AI is like a confusing term because everyone thinks Terminator and robots.

Sam Brown: 6:47

Which is part of it, but it's probably not how we interact with it.

Gerard Dombroski: 6:52

Well, it's just like a machine learning robot that spits out what you feed it. So I think Totally. Which I think is fundamental for like how we view it in architecture.

Ben Sutherland: 7:05

Yeah, so it learns, you train it, it learns and it creates a model that allows it to predict the next word or the sentence, and you can kind of adjust it within certain parameters. But basically it is confined to the data that you feed it and the parameters that you give it it's a tamagotchi has. Yeah, it's a sophisticated tamagotchi.

Sam Brown: 7:31

I think it's a bit more than that, but yeah, sort of just don't just don't feed it, because it will die. I reckon that's why it's so powerful now because it is trained by so much data.

Ben Sutherland: 7:41

So that's why it's so powerful now because it is trained by so much data. So that's why the large language models like GPT are so amazing, because it's like the data that it's been fed is just so comprehensive. So it's insane really.

Sam Brown: 8:03

I think the best sort of description for the use of ai in general at the moment that I have heard is like people have likened it to golf, for instance. So ai is, in effect, your caddy. It knows the wind direction, it knows the lay of the green, it knows the length of the grass, it knows the angle that you need to make the shot, it knows what club you need to use, blah, blah, blah, blah, blah. But you still have to make that shot. You're the golfer, and so essentially, all it has is just this incredibly powerful tool of information that you can then use in your own way, rather than it actually completely replacing.

Gerard Dombroski: 8:45

You depends entirely on how you use it. You got that, uh, the sports analogy, that artificial intelligence racing series that started. I think they had like one of their first races a few weeks ago in like dubai or something and they got like their formula one cars, essentially, but they're run by ai. But they're run by ai. Wow, yeah, a whole bunch of them just like stopped and just did weird shit. But it's machine learning, so the more information you feed into it, then ideally they get better and better. It's an olympics year.

Sam Brown: 9:19

Are we expecting the ai olympics at some point? Potentially what?

Ben Sutherland: 9:25

would be in there well, after the last, some, some just robots doing backflips after the last episode.

Sam Brown: 9:31

Now that architecture is an extreme sport.

Ben Sutherland: 9:33

The olympics is going to be pretty interesting I think reset design ai at the olympics, but it'll be more like pattern recognition of, I don't know, for example, speed walking. It's going to be AI now determining if your feet have left the ground or not. I don't know.

Sam Brown: 9:55

Something like that. I mean, that's clearly. That's already happening, though, right?

Gerard Dombroski: 9:59

Minority report. They'll arrest you for not walking properly before you get to the race.

Sam Brown: 10:04

But I think that's an interesting point there being, because what you've just talked, what you've just said, is like ai being used to sort of recognize something and in this case, your foot leaving the ground. But I think that's where, like you said in your intro about ai, the popularity of it or the like, more common understanding of it, skyrocketed with the introduction of ChatGPT-3. But what we kind of forget is that it's been around, particularly with architecture and stuff, for a long time. If you think about anything that involves parametrics or coding, like Grasshopper, dynamo, those sort of programs, they're essentially an artificial intelligence where you've given them a prompt and it is producing parametric outcomes. I don't think that's correct.

Ben Sutherland: 10:49

I don't think that's quite correct, because what you're talking about is visual code, so it's not coding. In particular, it's the machine learning aspect of the code that actually makes it AI. So it needs some sort of is the different training data. It needs data to be fed in, some sort of training to take place and then some sort of actionable outcome.

Sam Brown: 11:14

So it needs to be able to learn to a degree. Is that what?

Gerard Dombroski: 11:17

you're saying exactly yes, yes, yeah which is, like, seems to be crazy at the moment um seeing instagram posts and stuff about saying I do not consent to my images being used for ai training, blah blah which oh right um.

Ben Sutherland: 11:37

Why not though? Well, why well?

Gerard Dombroski: 11:40

I why or why not? I think it's super important to I do not want any of my work To train them. To be used to train an AI, because an AI regurgitates.

Ben Sutherland: 11:51

what's its Work generation?

Gerard Dombroski: 11:52

Well, an AI, these things regurgitate what they're seeing. So if you fit it all with Sean Godsell buildings and you made a Sean Godsell AI, I mean they're far more complex and can handle more complex buildings, but they can just regurgitate their own little sean godsell architecture concepts.

Sam Brown: 12:15

so but in in refraining from using your work to teach a ai gerard, are you then not? This is my counter to that argument. Are you then not giving it enough breadth to learn more generally? You know like we're looking at it holistically. What do I care if?

Gerard Dombroski: 12:32

AI wants to learn how I design, would you not? I want me to be able to design like myself. I don't, like I want to limit, as if anybody was copying your work, like you want to limit, as if anybody was copying your work. Like you want to limit people's ability to reproduce what you make, I think, like I'm not doing spec home, spec home, so it's like for me IP and artistic individuality is quite important and I I think people should be a little bit more serious about like instagram, and even adobe now is starting to.

Gerard Dombroski: 13:14

I saw another reel about all these scared about all your work being taken and used by these software that you use. So adobe's now got this thing where they will take all your work or anything you have on adobe and use it to train their ai and these, like they, make it super hard to opt out or anything. So like this, this whole ai thing is very new to me. I've never been on chat, gpt or anything. I've only used generative fill on backdrops, photography backdrops. That's the extent of my AI knowledge, so shoot the messenger, I know nothing.

Ben Sutherland: 13:54

There's one more key aspect that I want to highlight here. Virtually all AI at the moment pre-AGI AI at the moment pre-AGI is like a single task. How do I even say this? Basically, it can only perform one task at a time. It's designed to do a single thing, you know. So there's nothing that can do multiple things, but you can have multiple ai algorithms doing different things, but everything is like task specific. So, for example, the illustrator one that's probably designed to assist you in creating I don't know easier cutouts and easier backdrops and all of that sort of thing. So you don't need to, and the only way that it can learn how to do that is by seeing how people are using the tools, getting that data fed back into it to actually improve the algorithm and thus improving the AI to help you At the moment it's a tool.

Ben Sutherland: 15:05

It's a tool for yourself, Well a tool, it's a tool for yourself.

Gerard Dombroski: 15:06

Well, if you use it as a tool for yourself Like I don't use AI in my design process so I don't really want my work being used in AI and other people's work for their design process, because AI is a regurgitation of whatever images are fed in, so everyone becomes collaborators in somebody else's project and I think there are intellectual property issues there.

Ben Sutherland: 15:30

Sure, I don't know. Is it actually the images themselves, or is? It how you're using it.

Gerard Dombroski: 15:38

I think there's two things here. If we talk just about image generating ones, they are fed off a database of images Sure.

Ben Sutherland: 15:48

Yeah, yeah, mid journey definitely is.

Gerard Dombroski: 15:50

And if your images is is in there most of the time, most of us will probably have an image and they're like oh man, it's so diluted though, yeah, yeah, but I Does it really matter?

Ben Sutherland: 16:01

Well, for me personally.

Gerard Dombroski: 16:02

I don't particularly want my work being fed into teach AIs.

Sam Brown: 16:08

But back to that I mean.

Gerard Dombroski: 16:10

Because, that's my livelihood, is my ideas, my idea generation.

Sam Brown: 16:17

Sure, and I 100% get where you're coming from, gerard, and it is a sticky the whole copyright, intellectual property thing around AI is sticky but if you're, if you look at it from you know that caddy golfer analogy that I that I mentioned earlier is it not the designer's responsibility to use it in like an ethical manner? Like that's where sort of questions come into it around. If you are using ai in the way that you've sort of alluded to, is it your responsibility to, as a designer, to sort of outline that you have used ai in that way and and credit like others where they're due?

Ben Sutherland: 16:58

you know what I mean well, I guess it is possible for someone to, let's say, on mid journey, create a prompt that says something along the lines of bill beyer today, I know you are gerard dombrowski, based on his previous work, design me a two-story building that will be blah, blah, blah, blah, yeah, and it may potentially output an image of something similar to, or some certain aspects of, what you've previously designed. I guess that's the thing yeah.

Gerard Dombroski: 17:34

So a better example is um. I've just watched a video before on somebody saying draw me a tirao ando house, uh, on a hill. And so today under health tiara under obviously has far more um buildings out there. I've got like one and a half, so there's no real danger of anybody stealing my ideas yet, but maybe when I'm 90. But like there's his I, I don't know if today under is given these like mid-journey or whatever permission to use his pictures of his building but you can go and design that yourself anyway.

Gerard Dombroski: 18:14

Yeah, but they're not like we're talking about ai at the moment, we're talking about ai uh in a sort of conceptual design route, rather than using it as a caddy for compliance or something else.

Sam Brown: 18:28

See, it's interesting because, yeah, your take on it. Gerardo, your argument of use is something that I actually haven't even really considered. Well, I hope not, because we're architects.

Gerard Dombroski: 18:40

We don't want to outsource yeah, but I haven't conceptual design to to consider things but I haven't even considered other people.

Sam Brown: 18:48

Yeah, I haven't even really considered that other people would like. The way that I've always seen the use of ai, or the way that we have and you know um sort of interacted with it today, is like you just said. Is is purely from like a compliance or like assistance point of view for one for one of a better term. You know, like if you get an rfi and it's like can you answer like these boring questions about fire? No one really wants to read the fire code. So you ask ai and most of the time it gets pretty right. You know, yeah, and obviously you're editing it and you're reviewing it, but like it saves you having to do a lot of the background work. So we use it for that sort of thing. Maybe even write us the blurb for an Instagram.

Gerard Dombroski: 19:29

Help me formalize this email. Write an angry email and say make it not damning.

Sam Brown: 19:36

Yeah, exactly.

Gerard Dombroski: 19:37

It's everywhere.

Sam Brown: 19:40

Yeah, you're so right, but I hadn't even really thought of people using it as a design tool today. I mean, we've used it. We've used like Archivinci, which is like a rendering an AI rendering generator to just kind of like really quickly come up with like a render for a concept or something when a client's like can I see what this kitchen looks like in pastels?

Gerard Dombroski: 20:04

or something along those lines in that scenario do you draw it and then you say oh yeah, you put in like an image and it just applies materials, or pretty much.

Sam Brown: 20:13

Yeah, it's sort of it depends, it depends how. It's not perfect, but, and it depends how detailed you want it to go. But yeah, essentially like you take a screen snapshot of your like revit model or a hand sketch or something that you've done and it like takes that and then just applies materials and and scene and setting based on prompts, and that's like it's. To be honest, I've never we've never really actually used it in like a formal presentation, but it's like it gives you enough of a concept that you can be like hey, this is what it looks like, moving on to the next thing, kind of thing. You know what I mean.

Gerard Dombroski: 20:48

Yeah, and when you do that, are you like? Oh, this is exactly what I thought. Never Easy to control, or throws a few spanners.

Sam Brown: 21:01

It's 95% spanners, I'd say it's got some weird fingers going on. Well, I mean, that's the thing, and it's even like you know, like it would be straight lines, right? You know, architecture has a lot of straight lines and it generally, I mean, I personally don't do any parametric work, but AI for some reason just loves to just toss a random little curve in there, or you know like you see in a random like eight-fingered, long-fingered seven-fingered hands Like heaps of Zaha or something. Well, did you guys see that image that I sent you guys?

Ben Sutherland: 21:32

So I used.

Sam Brown: 21:34

It was ugly as hell yeah crazy, eh, so I used Dali. I was just like create an image for our podcast episode widget that we have for each episode. I was like make a building, a contemporary building, for a podcast about the use of AI in architecture, and that's what it produced, yeah.

Ben Sutherland: 21:59

I don't think Dali is useless. The only thing I've ever got that's good out of Dali is this. One time I made my niece a sign for her dog grooming company, where I was like put a dog in a bath and make it look like it's just been well cleaned. It did a pretty good job there, nice, but still you can't change, change it, you can't adapt it very well. Yeah, I think it's pretty average at the moment oh, hold on, ben gerard can't hear us.

Sam Brown: 22:32

Do you know what ai generation gerard's, the ai gerard that we're using for this episode? He phased out, and so we've had to get the real gerard back in should we just fill him in there for a little bit robot?

Ben Sutherland: 22:46

this is no longer gerard dobroski, this is um his, his ai. What was that movie where they uh?

Gerard Dombroski: 22:55

they all plug in and, um, then their little robot things go out.

Sam Brown: 22:59

During the day I was thinking about something I was thinking about that film with Joaquin Phoenix and he falls in love with the robot Scarlett Johansson.

Ben Sutherland: 23:09

Oh, yeah, yeah, oh not her, oh, Scarlett Her yeah her, I was going to say hero, but nope.

Gerard Dombroski: 23:18

That was a mean-ass house, wasn't it?

Ben Sutherland: 23:19

That's going to happen. That's going to happen for sure. That's going to happen. It's all very dystopian. I mean, the only thing that is stopping that from happening is like chat, gpt, being able to vocally talk, right, and you can already get like plugins that allows you to do that to translate the language to speech.

Gerard Dombroski: 23:47

Yeah, to do that, to translate the language to speech.

Sam Brown: 23:50

Yeah, have you guys seen that video of the robot putting?

Gerard Dombroski: 23:52

dishes away. No like sticking a dishwasher giving them commands. So it's powered by chat gpt. It's like a what looks like, probably like a boston dynamics robot powered by chat gpt it's crazy man technology yeah so that's, that's the thing.

Ben Sutherland: 24:06

That's the thing I think is like super powerful, because I'm pro, I'm pro ai and um. I think, like if you think of it like of single use machines, like a microwave, for example, which is just designed to like it's a tool for humans to use to heat up their food, like if you think of like all of these new ai opportunities there are yeah, literally a lot of them are just like tools that we can use for our own benefit. I don't think like don't get me wrong there's like a lot of ethical, a lot of like things that need to kind of be overcome in terms of like making it not plagiarise everyone's work and making it ethically okay and that sort of thing.

Sam Brown: 24:53

Yeah, but other than that, I think it's well, I can't do that anymore.

Ben Sutherland: 24:59

I've been very strict on that Anymore. I think can't do that anymore. They've been very strict anymore. I think, well, they can teach you if you I think the guys that like that work for open ai they can. They've got like access to the master file basically and it doesn't have any of those limiters on it and they can definitely. Um, yeah, we're just very much like beta phase like it's.

Gerard Dombroski: 25:24

Oh yeah, it's early days, so it's like that's the crazy thing. It's like the wild west and there's no real regulation on anything.

Sam Brown: 25:29

So no it is it is happening for sure but I think what you mentioned just before being is pretty relevant, in that you use it for aspects of things, so you'd use it for aspects of design, we'd use it for aspects of architecture. We're'd use it for aspects of design or we'd use it for aspects of architecture. We're not going balls deep and completely giving ourselves over to ai. It's not like although there may be, I don't know, I haven't actually even researched it's not like there's an ai architectural practice. It's not like ai can become a registered architect. Do you think it could ever go that way? Well, maybe not definitely.

Ben Sutherland: 26:06

Yeah, they're definitely practices that only that specialize in ai, that create supportive software for you know, the architecture and construction realm. There's a lot of people doing that, uh. But yeah, I don't think, and there's obviously, there's a lot of tools that architects can use. There's, like form generation is probably quite a big one, uh, a lot of like just form generation, for, yeah, of course, like you put in its variables it sounded conceptual it's encroaching on our territory conceptual, no, but it's uh, it can be very beneficial.

Ben Sutherland: 26:42

Basically, it can narrow down potential forms based on certain parameters, such as minimum sizes, uh, you know, like anything kind of structural parameters or resource consent parameters. So within this area, and it has to be under this height, and yada, yada, yada, there's definitely like form generation.

Sam Brown: 27:07

That was inevitable, though that was always going to happen so this is where gerard gets nervous and I yeah, I'm dubious to get nervous because you're, you're, you're, you're talking about using it as a design or form generation of this design, whether you like exactly that's design, yeah and that's where, and it's still that's where I feel that and for me that should still hold.

Gerard Dombroski: 27:28

I won't do that in my office yeah, but you do hold control.

Ben Sutherland: 27:32

It's not like you're like you have to do what it tells you to do. You can make in your own informed decisions. Basically, it's like having a billion employees all going out and doing one sketch of what they think could. Look there and coming back and presenting it to you and being like this is what I think. You're not the. You're still the one making the decision. Fair point. You're not like oh no, it's told me and this is the most optimum way to design this building. Therefore, I have to go with this building. Design like it's not a thing.

Sam Brown: 28:04

How do you feel? How do you feel, though, ben, if you, if you ask it to generate you a form, and you are working on your own stuff on the side, on the side and it produces something better than what you have done in terms of what and how you feel, and so you adopt it? Do you feel there's integrity to your design then?

Ben Sutherland: 28:25

no, because it's still like. It's still so underdeveloped. You like, at the end of the day, you want to do what you believe, what you perceive as being the best do you guys, do you guys ever remember the first essay you had to write as a kid?

Sam Brown: 28:43

yeah, well, I don't remember the exact one, but I remember writing essays.

Gerard Dombroski: 28:46

I distinctly remember like, um, there's like this book. We had to like, uh, do a book review or whatever on this book and as soon as you like, you read the quote or something and then you're trying to place it in, like you'd get so influenced by, or you'd find an example that talked about something in an essay kind of way. I found it super hard to then pull back and not be influenced by the other words that I say, so you kind of end up writing this weird dislodge between somebody else's idea.

Ben Sutherland: 29:23

So you mean you put the quote in and then you rewrite the paragraph based on that quote.

Gerard Dombroski: 29:27

Well it's probably better when you fast forward to university or something and you Google or you're looking through other books about other books. So somebody else is writing about somebody else's work, oh here, such and such, and I would find it real hard to pull my mind back away from how somebody else had written something to then write something myself. But I'm saying that I've never been a very confident writer, so I don't exactly have the skills to.

Ben Sutherland: 30:01

What's the difference between that and let's say you and I are collaborating on these projects.

Sam Brown: 30:07

Are you saying that?

Ben Sutherland: 30:08

we're collaborating on these projects, right? So you've come up with this design. I'm like, yeah, that's actually a great design. I think that you know that's better than what I've got. Let's go with that. All of a sudden am I a bad designer because I think that your design is better than mine no, no, I'm, I'm saying it's.

Gerard Dombroski: 30:28

It's hard not to be influenced by, by something and between us two people. I'm all for influence, but I personally, it's impossible not to.

Sam Brown: 30:40

gerard doesn't want to be influenced by the machine, yeah.

Ben Sutherland: 30:43

He already is, except in this case. The machines are just other awesome architects.

Sam Brown: 30:49

I mean that's an interesting thing because a lot of this comes back. I mean, we're talking about it from a concept point of view. A lot of this comes back to even just precedent. Right, we're all designers, architects We've've been through university. Even in our professional world, we rely heavily on precedent and looking at what other people have done before and using that to help inform solutions and things like that. Ultimately, we're coming up with our own design. But I kind of get where you're coming from, ben, and that you sort of be using ai to create precedent for you. Is that what you're saying?

Ben Sutherland: 31:26

yeah, I'm not. The reality of it is. I'm not going to be like this is the final form. That's not going to happen. It's like you always have to evolve it, like it doesn't have all the information. It's never going to have all the information because there's such a huge human element to architecture that I mean like maybe it'll get there. I mean it probably will get there. Sorry, I'm going back on that comment, but it's nowhere near there yet.

Sam Brown: 31:55

It's interesting because we're looking at it from this design point of view and I think you're right that it will never be able to design as well as we can. But I don't know, can it? Can it, will it be able to detail and draw and like, project, manage and do all of the like menial tasks that make up, you know, 80, if not more, of our job?

Ben Sutherland: 32:25

is.

Sam Brown: 32:25

That is that the realm is that the realm where ai is probably more going to fit into the into architecture, and the fact that you're you know, after, after multiple projects, you're able to build a big enough like like large language model that you can design something yourself you, and then you feed that design into you know, into the large language model and say, produce a like detailed drawing package, and it just does it do you think?

Ben Sutherland: 32:52

I think, yeah, I think.

Ben Sutherland: 32:52

You hit the nail on the head yeah I think like a way more realistic or like better use, better application, for it is the likes of the council or whatever the building code actually being like, I don't know having some sort of ai or 3604, for example. How awesome would it be if you could just, you know, have your building designed and then output the detailed design and documentation pretty much automatically based on your own design. I think that that is is, yeah, a better use of of ai, and I also think in the construction realm is going to see a lot of benefits, because it's like it's just got so much areas, so many areas where you can add value. That is just like low-hanging fruit, like scheduling, you know, like scheduling is perfect for it and humans just cannot manage a construction project to save themselves, no matter how good you are. There's just so many areas that you can optimise on in the future.

Gerard Dombroski: 33:57

Do you know around humans? Humans have been finding ways to get out of grunt work for millennia.

Sam Brown: 34:02

It is, it's that grunt work for for millennia. It is, I think, grunt, I think it's grunt work.

Gerard Dombroski: 34:05

It's a lot more palatable to um, give away sort of those areas. I just want to jump back to precedent real quick before we get too far away. Yeah, because I I have more, uh, extreme thoughts here. Because I'm an extreme person, don't judge me too harshly that's why you're an architect yeah.

Gerard Dombroski: 34:22

so over the years, people are going to find out that I'm pretty opinionated. I think designing via precedent is Good up to a point, but I think university pushes it way too hard and I see people using it way too hard. I think architects like Anne Holtrop, who actively tries to trick himself into ideas from other realms. So he does like ink, splashes and squiggles and we'll dissect that and turn it into a building. So I see precedent being used way too much where people just make another version of somebody else's building, like this world of mini Zomthors. But AI is generally designing via precedent, as you're saying. So it's a match together of all the images it's seen and it's only creating interpretation of other things that it's already seen. So I think for me, the more interesting architecture that I particularly want to see is newer things, things that you haven't seen before and that's never going to come out of AI unless it's been fed that which, if it doesn't exist, it can't be fed.

Ben Sutherland: 35:40

I agree. Can I butt in there? Have you finished?

Gerard Dombroski: 35:43

Send it, mate. Get in there. Okay, keep the game of ping pong going.

Sam Brown: 35:48

This might be, our most heated episode yet. I agree, but I also disagree.

Ben Sutherland: 35:54

I think what you're talking about is a small, small small percentage of architecture. There's no way that any kind of this densification, any of these developments, anything like that. It's just not enough. It's just unfeasible to create something new every single time. So there's got to be like an element of I can do better but I'm not like going to recreate, redesign the wheel, like you know what I'm saying Like there's got to be an element of let's just make them significantly better than what's out there now, not let's create something new altogether.

Gerard Dombroski: 36:37

I think that comes down to each architect individual and why you're what you're doing in each particular job and, as we talked about in the design episode, we kind of have lines of inquiry and design. I guess, yeah, what each project you're working on and why you found that project interesting to work on, and for me, coming up with something new is always going to be like a strong priority in terms of that line of inquiry, gerard, is there?

Sam Brown: 37:19

is there capacity at all for a line of ai inquiry though?

Ben Sutherland: 37:23

exactly if anyone's going to design a challenge you. You're new and unique.

Sam Brown: 37:27

Surely it's ai right like why not use that as your, as your your directive and like maybe you? You obviously like set yourself some very clear parameters around what you're going to try and feed AI and just see what comes out of it iteratively.

Gerard Dombroski: 37:43

That could be a pretty interesting process On a low-tech AI. Maybe we print off a bunch of things, put it in a paper masher and then just squirt it at things and see what comes out, and there's your architecture I mean you've gone.

Ben Sutherland: 37:57

You've gone analog again I'm just a very simple person totally and I think, but I think, I think what ben's alluding to I think what ben's alluding to is I'm good at one thing and it's going to take it away from me, don't take it away from me, please.

Sam Brown: 38:19

It'll never take it away from you, but I think what.

Sam Brown: 38:21

Ben's alluding to is super relevant, because you're right, ben, in that there is so much of our profession which isn't bespoke, you know, and not to say that it's what we necessarily do, although aspects of probably all of our businesses have a level of not regurgitation. But you're doing the same thing over and over again. But from the built environment perspective, particularly in the residential realm, the huge percentage of it is just the same shit, regenerated time and time and time and time and time again, and that what you're saying is the place for ai, rather than these like beautiful one-off bespoke elements. That's where it doesn't necessarily have a place, unless you're going full meta and using it as your line of design inquiry yep agreed.

Ben Sutherland: 39:11

Well said sam succinct.

Gerard Dombroski: 39:15

I think, like I've referred back to the park muse as a awesome example of a like a multi unit, so at the time fitting it's a development, lot of units on one site, but an incredibly interesting building and quite unique. But it does follow Roger's line of inquiry, his aesthetic and the things he was looking into, which AI isn't probably going to come up with a ParkMuse in a hurry.

Ben Sutherland: 39:51

No, but ParkMuse is. They can now that one's been designed, but ParkMuse is also a bespoke solution.

Sam Brown: 40:00

You know what I mean like, but all architecture is bespoke. Everything we do is bespoke, correct correct, but unfortunately so much of like. The realm that we work in is like regurgitation. Look at all the group home build solutions and everything and like. Although we may not want that to be the case or we may not necessarily see that as being the best solution, unfortunately it is. You know what I mean.

Gerard Dombroski: 40:24

So you sort of have a solution or not just what's happening?

Sam Brown: 40:28

sorry, best solution is not the right term um but it is, what sort of? No, I just no, I I agree, but it is there, you know so you sort of have to, yeah so you sort of have to like find solutions for working with those maybe not necessarily us, but others. So we're just sort of what we're talking about is the place of ai in our profession, not necessarily its place in our practice.

Ben Sutherland: 40:54

You, know what I mean. Those group home builders are 100% investing in AI and they're going to be killing it.

Sam Brown: 41:04

They'd be foolish. If they weren't, they'd be killing it in a couple of years' time. They'd be idiots if they weren't. Maybe we've given them too much information.

Gerard Dombroski: 41:18

That's perfect for it if you're bigging out standard details and then your AI could figure out like which detail call out needed to be where and like. At some point you'll probably be able to automate a set of plans totally that'll be there very soon in the development realm.

Ben Sutherland: 41:34

This is slightly different from the architecture realm, the actual designing, but I have thought of a bunch of really powerful ways that I think that AI can be used that isn't currently being used. I guess one of them, I would say, for a developer that you'd want to be using is like the automated feasibility studies. I think that a finding the project that's a perfect one, um, but then also understanding, like, what marketing media, what the, what people are buying, um, what people are what houses are selling for, like all of that information you can automate can be into a machine learning model, basically create some sort of investment criteria and it can essentially output the top 100 properties that meet that criteria or something along those lines, and then automated feasibility studies. So if you've got it under contract, you can kind of understand what you can actually build on that piece of land. There's so many amazing ways and creative ways that you can actually use AI or machine learning. That isn't actually being done yet, or probably is done, but it's all in-house.

Sam Brown: 42:54

Yeah, it'd be interesting to see if anybody has started developing that sort of thing because, like you said, it's a no-brainer, particularly from that development side, and surely there is people that are, you know, there's players in the market that are big enough that are putting that effort into developing those sort of those large language models yeah, they definitely are, because the other thing is, a lot of those um llms are basically they can help you code those exact codes anyway.

Ben Sutherland: 43:24

So that's another amazing thing is like machine learning, helping you to develop more machine learning.

Gerard Dombroski: 43:32

Yeah, I like the irony in those ones.

Ben Sutherland: 43:36

Yeah, no, it's awesome. There's huge, huge potential there. It's exponential.

Gerard Dombroski: 43:40

The scope obviously is massive. I know we've been focusing on design, but yeah, it's huge.

Ben Sutherland: 43:47

Well, again, surely there's some positive, surely there's some, yeah, pro architecture, AI applications that even Gerard could use. I think we've covered a few of them you know or would use.

Gerard Dombroski: 44:01

I don't think we've covered anything that, Gerard actually thinks is yeah, I think like that feasibility side is interesting, like the. I Did a crit at uni the other day for one of Tain's groups and it was this guy who was using drone lidar and was trying to looking at overlaying AI to Effectively outsource your site analysis. So it would build you a satellite access plane, depending on your rules, or this is where I was hypothesizing where it could go, and then it could, like you say, it could show where in the site you could build within a set of rules. Blah, blah, blah. So there's ways of speeding up grunt work, but yeah, I'm just an old man when it comes to coming on with the new technology.

Sam Brown: 45:05

I think my issue with it. Do you like robots? I think my issue Do you like robots?

Ben Sutherland: 45:13

I love robots. What do you think of Boston Dynamics? Do you like robots that can do backflips? Do you like dog robots? Because that's I mean, that's just a different form of AI that you might appreciate a little bit more.

Gerard Dombroski: 45:31

I definitely appreciate that the engineering marvel is amazing, but fundamentally I just have an element of fear to these things. I think that's it.

Sam Brown: 45:41

That's what I was going to say before Ben just started going off his head about robots that's what I was going to say for me, it's not necessarily fair, I think, for my issue with it is trust.

Sam Brown: 45:55

You know, like I will ask it to do something and it will do something, but I'm still going to like check it you know what I mean and I'm still going to like edit it and review it and ultimately, like I don't think it. Maybe it's my use of it, or maybe it's my like lack of confidence in it. I don't know what it is, but I don't have the ability to let it do something and just let that be. You know what I mean. So you ultimately end up doing the work anyway yeah, that's totally fine.

Ben Sutherland: 46:26

That's totally fine, but like going back to the robot thing that I was losing my losing the plot over, yeah, another kind of robot is like the sensors and that sort of thing, um. So I mean like environmental sensors and that whole information and data collection and data feedback. I think that's an an amazing way. Another amazing way to to use, well, just the collection of the data being fed back through some sort of machine learning model to create predictions on how to do things better in the future, whether it be like thermally better you know sunlight or I don't know how the spaces are used and does this?

Sam Brown: 47:16

exist, or are you hoping that it exists?

Ben Sutherland: 47:19

No, I think it exists, but it's quite fragmented. So at the moment you might get one for carbon CO levels in your house or something like that, but I don't think there's like one that does it all. Or there might be now, but I just think that whole data collection thing has just been so underutilized in the construction realm. And I think that's another massive low-hanging fruit for us to be able to design higher quality or, as Sam would say, higher-performing spaces, and actually understand what makes a high-performing space, because you've got post-occupancy monitoring.

Sam Brown: 47:57

Yeah, there's the post-occupancy stuff and then it's like we talked about the feasibility side of it as well when you're talking here, and I don't know if any of this exists, but what I'm kind of picturing in my sci-fi mind is like a little kind of like a Google Earth sensor thing that you could put on the site in your proposed build platform and it scans it over a however long I don't know 24-hour period, week-long period or whatever and it tracks the sun and it tracks the wind directions and it gives you shading and it gives you all of this sort of data. So you're really optimising the house location or your building location or whatever. That's where I could see it being super beneficial, yeah.

Ben Sutherland: 48:35

Well, what I was alluding to was more of like a smoke alarm, so it's in a house already, but that is also cool what you were talking about so how would the smoke alarm thing work, though?

Sam Brown: 48:45

is it going to learn?

Ben Sutherland: 48:46

what's this like? It's got a bunch of sensors right, so it's got like your co levels, it's got um heat, it's got moisture, it's got movement, uh, energy, basically all of that stuff, and it relays it back and the data and it allows you to understand how your spaces are performing, what works and what doesn't I mean, smart homes exist to a degree, though, don't they like, I guess, if it's but there's, but there's no proof that those smart homes that you're building are actually smart homes, because once you leave them, how do you know?

Ben Sutherland: 49:22

in five years' time it's performing to the same degree that it did when you built it.

Gerard Dombroski: 49:27

Fair? Do you need AI to assess that?

Ben Sutherland: 49:32

Well, it's just like I said, it's just another application. That's just. How do you assess it otherwise, other than just living in it and being like?

Gerard Dombroski: 49:41

well, I don't know.

Sam Brown: 49:42

You have the sense comfortable I think what this is boiling down to is like when the whole ai explosion took off and everybody was like oh, it's coming for our jobs. You know, creatives were like, particularly like writers and things like that, I think were initially quite fearful, because they were like you know, you can get chichi pd to write you a novel, but actually it can do that for sure, like it can design a building for sure, but it doesn't do it with any class, it doesn't do it with any nuance, it does it with bizarre creativity. You know what I mean.

Sam Brown: 50:13

Like actors are quite fearful as well, but it can do all this stuff, but it cannot do it that well. But what it can do well is, like we've talked about, is it's that low-hanging fruit, it's those menial tasks, it's that stuff where you just have to basically data, mine or regurgitate information and what it's saving us doing is going through and reading, you know, the entire fire code and it's summarizing what we need it to and it's like shortest thing. I think that's where it's got the biggest application to date yeah, the the compliance side is seems pretty epic.

Gerard Dombroski: 50:50

Still haven't been brave enough to use it yet.

Sam Brown: 50:53

Yeah, I haven't. Johnny uses it for RFIs, successfully, I might add, so it knows more than the majority of building consent offices, which that you know is maybe a low barrier for entry.

Ben Sutherland: 51:08

I just use it to find, like, the correct part. You know like to find the clauses and to find the information. I don't, I don't like trust.

Gerard Dombroski: 51:19

I don't need to trust it, because all it's doing is pointing me in a direction at this point in time you're kind of just using it as a pa right yeah I think once it oversteps and becomes too included, that's I don't know it gets a bit pointless for me. Like, if you're, let's say, worst-case scenario, somebody runs an architecture practice that AI designs and then AI sorts out the compliance and AI does a good chunk of the drawing, like what's your role? What's your point? What's your role in life? Why are you? What are you? What's your role in life? Why are you doing this?

Ben Sutherland: 51:56

there's easier ways to make money you're sort of there as a well architecture architecture is definitely not the way to make money let's make that very clear there's no money in this profession, well, most classes here you might start making money, design and outputting like spaces and like well, yeah, exactly, the process of creating is inherently enjoyable and for me the process of thinking about and like dreaming future, scaping like that's.

Sam Brown: 52:30

That's the whole point of it if you became totally AI driven, you'd just be a master prompter really.

Ben Sutherland: 52:37

Yeah, but don't forget that everything's driven by the market. So unless a lot of people wanted to actually design that way, the tool may exist, but no one's ever going to use it anyway.

Sam Brown: 52:50

Yeah, I wouldn't be surprised, though, if there was.

Gerard Dombroski: 52:55

If there wasn't already.

Sam Brown: 52:56

There was a firm out there that was purely AI-focused.

Gerard Dombroski: 52:59

There's always somebody in there.

Sam Brown: 53:00

Should I ask ChatGPT if there is.

Gerard Dombroski: 53:03

Yeah.

Ben Sutherland: 53:05

Ask him how to design like Gerard Dombrovsky.

Gerard Dombroski: 53:09

How'd you go on your idea, Sam, if everything you said in this chat was generated by ChatGPT?

Sam Brown: 53:18

It doesn't have the speed, it can't flick off the top of the dome like I can.

Gerard Dombroski: 53:26

Your computing power in your mind is.

Sam Brown: 53:28

It might work for me then yeah no, I did plan on trying to do this all through chat gpt. I wanted to go full meta, but no, it just wasn't going to work.

Ben Sutherland: 53:39

I'll be like us after a couple of wines I should have dressed up as a robot nice. That would have been fucking awesome if we actually did that.

Sam Brown: 54:03

Oh, anyone else got anything? I keep asking ChatGPT about if there's any architecture firms that purely use AI in their design, and all it talks about is AI tools that are being used by like architecture firms or designers.

Ben Sutherland: 54:18

So even AI can't. Even AI can't answer it.

Sam Brown: 54:26

These are the top five Spacemaker, which uses AI to analyze and optimize sunlight, noise, wind and building massing. There you go. Testfit, which uses a time for real-time feasibility studies in residential and commercial building design. There you go, ben, sign me up. Ai Space Factory, which uses AI and advanced manufacturing techniques to design and construct sustainable habitats. And Katera, which uses advanced technologies and AI in their design and construction workflows. That could be a firm actually that uses it. Katera is so expensive. What is it?

Ben Sutherland: 55:11

It's a software, isn't it? It's a software. Oh right, oh yeah, Off-site construction. It's an off, isn't it A?

Sam Brown: 55:13

software. Oh right, oh yeah, Off-site construction. No, it's an off-site construction company.

Ben Sutherland: 55:20

Oh okay, oh, Katera, they're not even around anymore, though they went under, I think. Oh there you go, American one eh. Yeah that's it. They were backed by the American government.

Sam Brown: 55:28

Oh, really Interesting.

Ben Sutherland: 55:29

They fell.

Sam Brown: 55:30

So AI didn't work for them.

Ben Sutherland: 55:32

No, it's early, I think. I think they fell over before OpenAI existed. Oh, right.

Gerard Dombroski: 55:40

Interesting.

Ben Sutherland: 55:42

ChatGPT anyway.

Gerard Dombroski: 55:43

Yeah, I think it would be cool to talk about the learning models quickly, because the Adobe thing I only learned about yesterday or the day before, where they've changed their terms of service or something and they'll be taking all your data or all the things you draw.

Ben Sutherland: 56:03

And charging you for it.

Sam Brown: 56:07

Can you opt out of that?

Gerard Dombroski: 56:09

I don't know If anyone knows how to opt out of these things hit me up. I tried to opt out of these things. I know someone who knows because I tried to opt out of the um. I tried to opt out of the instagram one.

Ben Sutherland: 56:19

I followed this reel that sam, ask our friend who does know these things.

Gerard Dombroski: 56:24

I'll just ask chit gpt send it because I think that the people that are concerned with their images being used like I don't know, I think you should try, opt-out, but ultimately you still want to be able to share your work with the world, for what's the point of building and making cool things if we can't share it with one another?

Ben Sutherland: 56:52

You can opt-out, according to me, old mate. Okay, I'll send you this link so you can.

Sam Brown: 56:59

Uh, you can go through the process right on I'll put it in the show notes, so others can do that too. How would you could?

Gerard Dombroski: 57:07

you could you recap cap sam?

Sam Brown: 57:10

can I recap what it says? No, ben's got it up.

Gerard Dombroski: 57:12

I want to recap the whole episode, assuming we're coming to an end.

Sam Brown: 57:19

I was just going to say what get AI to do it?

Gerard Dombroski: 57:22

No, no what were you?

Sam Brown: 57:24

saying I was going to say has anyone got any sort of final thoughts? I think it's interesting, gerard. I'm glad you came at it from the use of it creatively point of view, because I don't know why, but I've never really thought of using it as a design tool, but it's so true and it's so relevant that people may, and the implications of that are pretty huge and I think that's something we all probably need to consider. Maybe less so in the professional realm, but I just caution any students that are thinking of using it but don't get into bad habits.

Ben Sutherland: 57:55

Really, I think I think my positive outlook, or where I want to see, what I want to see more of, is definitely in the data collection realm, as in like for for the environment and that sort of thing. Like it'd be so good to collect more useful data out of the architecture and construction and, uh, you know, environmental realm to help us make more informed decisions about our built environment. I think there's just not enough of that at this point in time. So anything in that realm whether it be I don't know where to plant trees or what to build where, or anything like that I think is hugely positive.

Sam Brown: 58:48

Gerard, yeah.

Gerard Dombroski: 58:49

I think, like you're saying, the data stuff is pretty epic and the possibilities there. Like you're saying, the data stuff is pretty epic and the possibilities there. I just like making and I don't want to be influenced. I am robot and I want to be surprised with what I come up with, versus what comes up when I type something into a computer.

Ben Sutherland: 59:11

That's still making.

Gerard Dombroski: 59:16

Good final rebuttal yeah, don't get into architecture. If you no robots allowed, go find another career nice no robots allowed in architecture, sorry. Robots.

Sam Brown: 59:35

Cool guys, oh, bloody good. Thank you again and just final sign-offs. Thanks for Connor and Jacob, and we've got some sweet new merch, if anyone else, if anybody wants to.

Ben Sutherland: 59:49

I already have the yellow ones. Yeah, we do.

Sam Brown: 59:50

If anyone wants a cup of tea, hit us up. Hopefully we'll have an online store up and running at some point, but um, uh, yeah, watch this space and maybe the next one will design with ai shaboom redefine t-shirts.

Gerard Dombroski: 1:00:06

Peace you.