0:02 Okay, this is the moment the computer is trying to trying to decide if it wants to work or not. So welcome. I was a part of CDFAM last year in New York. I’m I delivered the opening talk, so I think I didn’t do something terribly wrong, so I’m back here. One thing I realized here is that I saw all the present presentations but one it felt like CEFAM, “E” for engineering, a little bit. So I’m me, I’m not an engineer, actually. I dropped from electronic engineering, became an architect, became a footwear designer. I guess I’m still kind of like exploring. So I’m going to put the CD back a little bit and bring things back home.

And yesterday, you know, there are a couple of questions about computational design just the facts. Like, you know, I’m indexing zero is the first thing I want to talk about. Computational design is: you are working on a design, meaning intentions and knowledge, and you’re trying to construct it through computational thinking, representing, sensing and making. That’s computational design, okay? So this is design. This is defined by MIT Design and Computation Group, which I was a part of. I have both my masters and PhD from MIT in computation design that’s how I started. Interestingly, computational design, unlike what we think about it, it’s more about humans than computers, because if you think your difference from computer, actually, you can use it to the max extent. Like, our brain has a way of working, which I’m going to talk about, like, briefly. But then computers, they have a their own way of working. So if you differentiate that, things get easier. My talk is all about that. Last year design, it’s called Design Computation Human, if you want to check it out.

1:50 So I had to put this slide this morning at 4:00 a.m. Every talked about, like, heat exchangers this is the biggest he heat exchanger in town, so if you don’t know, read about the history. Okay, so this building, actually, it uses the undergrad kind of like tectonic plates, you know, gaps, to chill and warm the building up in in the winter and summer. So that’s just a kind of like, you know, a small note. So it’s interesting it also relates to what I’m saying. Is Al actually a practice? Like, I’m practicing it on my own. So when you when you’re looking like this, you see that when you kind of like you see R oh, it’s like, okay, this the heat exchanger. So it’s really like, how do we kind of like take our gaze away from the problems that we’re dealing with like computation, numbers, data and then how can we come back to them and deliver the the best we can? Being critical always counts.

2:48 Rena MRI, it’s 1929, right? So it’s a it’s the painting of a pipe, which is saying it’s not a pipe underneath it. So the artist here is trying to tell us something. It’s like, I made a picture which is not a photograph. It’s a picture of a pipe, which is not a pipe it’s a picture, you know, you figure out what it is, right? So that’s the kind of like the art.

3:06 And it’s it’s kind of relevant, because we talk about both, like, generative systems including Ai and image making, but also making models is pretty kind of like in a in a similar fashion as well. So this is my nut is shoe it’s actually a photograph of a 3D printed shoe that was designed by using AI. The sole reason for this to exist is this: AI takes the image mod takes the image kind of tools were kind of like taken over, and I said, okay, let’s make a shoe, let’s put on the table, because somebody’s going to step up and they’re going to tell that, you know, like they made issue using AI. So let’s do it, let’s keep it to ourselves. We did this it’s been like 18 months, we never talked about it, we never showed it. And then, you know, we employed again, I’m not going to go into details genitive AI techniques, image blending, takes the image. But then, you know, somebody takes this and use employ some, like, bitmap processing modeling techniques, and then we finally made a you know it’s a zc COR print, so it’s not a flexible model, which we had before for other ml models. But essentially, this was this is kind of like showing exactly in the company, saying that AI is coming. It’s okay, you know, we made it, so it’s done. And don’t take it take it with gr of grin of salt, if anybody else does that. Okay, so that was the point. Now I’m showing it because it’s done recently.

Okay, so computational design at New Balance didn’t exist I started 8 years ago. I got the first title, and then I said we need a group, you know. It was kind of like constant value proof like proving the preparing this, like proof of Concepts, and proving the value of computational design, why it’s important, how you know, how can it help solve problems. And then, you know, in my journey, actually, I built competitional design teams for all these companies compe sunf, Fox, being the architecture, new buildings two years ago, Samsung. And then I’m also often referred as Troublemaker, because instead of kind of like selling the stuff to you, I kind of like punch it in the face, like, data. And so, you know, computational design and then show, okay, you know what, these are Smart Systems, but you have to be smarter to kind of like do something really about it. So this is my troubl making slide, just to put everything into kind of like context.

5:21 One thing really important for me is, like, why are we doing something. And, you know, what what what is, like, what is the added value, what what kind of value are we adding, are we doing it for, like, changing the world, like making it more sustainable, being remembered, and so on and so forth. So that’s the kind of like what’s going to happen with human population okay, so at some point it ends. So just so you know, so the human population estimated estimated Peak is 11.2 billion people, we only have kind of like 76 years to get there, if I’m if my math is right, now, right.

6:00 And when you talk about AI, actually AI started in 40s, so it’s been 84 years. So we are closer to our Peak than they kind of like you know, like seminal work in AI, just imagine that. Okay, so then it’s going to kind of like start declining. And, you know, also the the the second note about troublemaking notes: usually everybody gets forgotten by their grandchild, gets have kids that, like, by Third Generation. Okay, so just keep in mind so we have to enjoy, you know, what we doing a little bit.

6:35 So this is not a plan it’s kind of gen it’s it’s the AI rendering of a plan that I draw, like, years ago. And this is not a facade same thing. What is interesting is, like, what’s happening here first. Okay, why are we kind of like captivated with AI, or numbers, data, which I’m going to talk about too, because of this interesting kind of like transition. I wasn’t doing any facade work here, actually I was working, trying to generate, like, plans. But because the data set is super biased towards, you know, generating building facades, it kind of like started shifting towards that. And, you know, you can like, we can talk about this, right? So is it interesting? Well, the transition kind of looks interesting, so it was interesting for me, for half a day, and from not plan to another, a facade is an interesting transition. So we get excited when we see that, but not everybody gets excited when they see that. Okay, so what is the distinction should we get excited, or should we not get excited is the main question here. And let’s ask why.

7:32 So here is one I’m talking about the model. The model motivates The Architects to start working with realistic images at the very beginning of the design process. So this is kind of like proposing a rendering system which generates the images that you want to see sounds familiar? Have you seen any AI Architects on LinkedIn making crazy buildings? Okay, so this is me in 2006 this is my thesis, which was, you know, at MIT, trying to render the images that you wanted to see, okay? So which was super challenging I didn’t know coding, so I had to teach myself coding first. Bitmac processing, trying to numerically evaluate images and give Fitness scores to them I I mean, I’m An Architect, by the way, you know, like I wasn’t a programmer at the time. So it was a big challenge of, like, learning. And then 5 years ago, I was like, oh my God, this is like machine learning this, which I didn’t realize at the time, so it was interesting.

8:29 So when you do that and this is not me, like, I’m not talking about myself, I’m talking about experience of getting lost, which is probably happening to you too when you look about gen genitive design. It has a very I’m talking about specifically the image making tools it has, like, a 2 and a half year history, okay. Diffusion models and so on I’m not talking about esans, which is older. Kyle Steinfeld, you know, an older colleague, also friend from MIT years he wrote this, like I, you know, as as I was looking, I found this article which is super interesting clever little tricks. It’s talking about the history of this image making Venture which started in 40s and bringing it to this day. And what is interesting is that he published this by using references that are not scholarly, okay? So he was apparently he was taking a risk, but the article is kind of pretty strong. It says the connectionist AI in 40s use Network to simulate intelligent Behavior it’s, you know then somebody’s talking about a perceptron, it classification algorithm, first invented in 1943. So those are the drawings of, like, neural networks from that time it’s the research, I cannot understand the math. So I’m not that kind of like deep, but if you look at the Warren MW clock paper from 43, you’re going to see that.

9:44 So why are you maybe not also excited you understand how things work. About, like, computationally defining Aesthetics has been a problem since the initiation of computers people try to kind of like find ways to make computerss say what is good, what is bad. This is from George Stein and James Gibbs, the founder of shape grammars. In so this is the book from 78 there’s a paper that’s called Ai and Aesthetics, it’s called architectural intelligence and Aesthetics, right? So they are defining the system so there’s the initial conditions, there’s a receptor which kind of like sees stuff, and then there’s a synthesis algorithm, and then there’s an affector which makes stuff in the environment, right? So they’re kind of like hands, and so on.

10:26 75, it’s pretty impressive he’s also my kind of dissertation, you know, PhD advisor. So it’s talking about receptors I’m going to skip just, the same time, are used for object recognition, right? And then effectors are used for physical outcomes. So one kind of like takes in, one kind of evaluates, the other one does, and so on and so forth. So these Concepts have been there for a long time, but they end the paper saying the ability to specify criticism, or a design algorithm so the whole purpose is the system is supposed to make art or criticize art, and so on the ability to formalize a wide range of perceptual and cognitive skills, and a wide range of knowledge, and thus can extremely be can be extremely difficult to develop, at the time. So that’s what they’re saying, so it’s it’s interesting.

11:12 So those years, like 70s and ‘ 80s, like, people start kind of like diverging to two parts one is, you know, one side follows digital, like computation as we know it shape grammars, for instance, kind of like steers away. So, you know, at this point, steiny and Gibs, they say: we’re going to talk about computation, but our bits will be shapes. So let’s say x equal to three, right, like a variable so here you don’t have symbols, like you don’t have an X, you don’t have like what you see is X, is what you see. It’s just like two squares, right? So if you start kind of like taking that shape which is two squares or four triangles, in this case, you have, like, two different descriptions for the same thing, which is computationally a challenge. But you can still Define find it, right? So if you have two descriptions for one thing, you can still, you know, create these two states. But then you can also discri you can come from different sides and discretize it in this way, which is super weird. So two two different things, you know, you start this discretizing, and they become the same thing again when they’re totally kind of like split like, how how would you make that work computationally? So these are kind of like incompatible descriptions of the same thing, which is a big Challenge.

12:24 And this is human Vision so that’s that’s really interesting, because I’m not going to go into computer vision that much, but the way in which the computer see is not the way in which we kind of like cognitively see and understand things, right? So we had to kind of like solve it in another way. So let me keep discretizing these pieces into other parts so I’m not going to talk about what they are, but depending on how you do things, you can get, like, different results, different, you know, kind of like let’s say number of different solutions.

12:52 And then the problem is, like, this this is what we do, right? Somebody I mean somebody comes to you, and then, as an artist, throws a pipe, or, like, an you know, an engineering problem appears which is which never existed, right? Like, suddenly you have to kind of like solve for that, which is exactly the previous presentation that that we just saw. So how do we deal with this inexhaustible problems?

13:17 So one thing about AI I’m going to talk about AI a little bit is that there’s this thing about H H TL, right? Human in the loop anybody, do you know that? What what’s what, okay, somebody, that’s great. So the the promise is that AI does some stuff and it fails, you know, which we saw, and then the human steps in, says, okay, it had to be this way, so you correct it and AI learns from you, wonderful.

13:40 And then there are papers about creativity so one says the gold standard for creativity assessment is the consensual assessment technique. So we are going to get together and agree on something in which human rates human rers us meticulously evaluate ideas and products of the B bases of our of their own judgment and expertise. So now we have to decide on this I’m G to ask, Art or not, for instance, because we have a consensual census it’s art, I’m sorry, it’s done, so you can’t decide it already happened in 1917. toam Fontan, it’s art, right?

14:19 So probably, if this was a new piece I was showing you this is my art piece probably consensually you would tell me to get, like, you know, off the stage, off of the stage. And it keeps happening again. Troublemaker humors this is the comedian, it’s an art piece that was presented in 2016 in Miami Beach it’s called yeah, it’s it’s called com comum, pretty much. So interestingly, again, you cannot compute these things.

14:41 This is, you know, it’s anomaly it’s kind of like outside the average, you know, the statistical average, but you can computational talk about them, like, in mathematical terms.

14:53 So one thing about shape grammars is that, because I can take any shape in this case I take the fountain as the shape, it goes to Fountain so that arrow is a computation, which means Fountain goes to Fountain, and it’s art. Or, you know, you can get a banana and tape, which are like two different objects in two different layers, and you take the banana from the table, put it on the wall so it’s the te transformation of banana, and probably you squish squish the banana a little bit, so Banana get deformed, so it’s like banana Prime now it’s not the same banana you picked up, all right? And then you put the which is the T of the tape so that’s the computation, that’s how you compute art. And then somebody ate it this is all real, by the way, it’s not a joke so Banana goes to Nothing, is another computation, it disappears. And somebody already paid 120,000 for that artwork, so that goes to zero, okay, so that’s another competition. So what so I show this people, and people ask me why are you still employed. It’s not a joke.

15:55 So let let me let let me try to kind of like bring it back with da. We did this interview just before I came one question was, what have you explored since the last time? Well, AI I looked into it, I didn’t want to say AI, it wasn’t interesting, and it wasn’t really internally it wasn’t interesting either. What was interesting was the stuff that we were doing before that, like a multiobjective kind of like evolutionary algorithms. I happen to look into octopus, you know, in this case McNeil, rhina, grasshopper, thank you again. And then, you know, I call, like, Robert’s name, I messaged them, and so on. The the wonderful thing about this is you’re trying to optimize against two goals here, and there’s a point that you can never achieve, which is wonderful, right? You’re kind like trying to get there, and, like, you’re never getting there it’s it’s so real, it’s like life. So we use these things to to kind of like, you know, of course, optimize things, and digitally, but at the same time you have to kind of like test the ideas physically to. So it makes sense.

16:56 This is also all founded back in 75 so John Holland writes, you know, about multiobject evolutionary algorithms in ‘ 75 so he draws this, like, pattern recognition device nobody reads it, nobody, like, you know, they they they publish, and probably he was getting, you know, he says, oh. And then in ‘ 92 he publishes a book, and in the book he says, well, every year we were selling selling like 100 copies, maybe 200 copies of the book check the citation number, and that’s only the 92 paper, not including the previous one. So he got, I think, 40 to 70,000 citations for that work, so it was a little bit early.

17:36 So this is this is this is the so what Pro so what kind of like problem at at the time, if we are kind of like converging towards the same thing together, probably most of us are going to fail, it’s not interesting, okay. So if you’re trying to kind of like steer away from the the crowd, and, you know, if you are wrong constantly, I would but still, you know, suggest going that direction, there might be something there, if you’re internally feeling, you know, there might be an answer elsewhere. So being being starting early is good, it’s kind of like painful, you’re alone there.

18:09 So gentic components anybody knows about who knows about gentic components? There’s one person there, raising his hand. I’m going to I’m going to talk say tell who you are. So this was like 2004 and five, and constantly we were saying this is micro station from UK we’re saying, you know, the symbolic view, we want to pull wires and plug them, that’s how we want to design, that was the feedback we were giving, and they, like, no no, it’s representation, like you have it’s hard to do. And then, you know, what happened, right? Grasshopper came in, like, 2007, so that was it. So we were like, okay, we were talking about this.

18:42 So 2006 I started doing, you know, these this was a building that I had to reconstruct from the points, because it was parametrically designed, and the parametric the model was corrupt, and I was hired, so they as asked me to rebuild the building from points. Then I had to learn R A Visual Basic, so that was pretty much it. This is a drawing by Bradley renberg of the CEO of anthropology, because he was a curious student, and I got a chance to kind of be his man while teaching gen of component. So this was the architecture kind of architecture he was doing as a student. Then he said I’m going to become a CEO, I guess, so that’s good.

19:18 We had access as kpf to object machine maches, 3D printers, Z Corp, really early. What I’m when I look look beh look back, we were kind of like redesigning all the structures for printability, and we were challenging the tolerance of the machines at the time. So, you know, like, just just looking back, oh, we were designing for, actually, like, the manufactur they’re not mechanical parts. So I don’t want to I don’t want to demean what you the crazy complexity that you’re dealing with, I’m just saying, like, there’s a portion of computational design Architects that, you know, dealt with these problems without knowing what they were doing, which was the interesting part, I guess.

19:56 Buil intelligence system so there’s a red dot Circle there, it says self-aware facade system if it ends up generating the same pattern, it kills itself and regenerates pattern again. So those are the kind of like stuff that we were doing, and for TPMS and lates. So when I joined I joined new in 16 but they were doing already, you know, like luses, and by 2020 we were already done. So we were kind of like so when I designed this, I knew that it was printable with the material I kind of more or less know the, you know, let’s say physical properties and the field, which just came by experience, right?

20:37 So I just want to highlight that again it’s not about me, because we ended up printing 1,300 parts. We got a kind of like mental sensation of how the thing works with that resin, with that printing method, and with the, you know, kind of like the lates we had. So if I had like, if if we try to try to train AI today, probably we won’t be able to, because we don’t have enough data to to kind of like get that inside, which is an interesting problem.

20:58 Before AI there was ml you remember that, right, machine learning. It’s interestingly the I I’m not talking about us again, maybe in general public, you know the technology comes, it becomes a kind of like super hype, and then it kind of disappears, right? So everybody was talking about machine learning, SC scans, and so on, and then now everything is AI. Couple of people said machine learning, so thank you for that you know, one is an umbrella term, obviously everybody knows that. But it’s it’s how interest what is interesting is that the way we talk about things also changes the way we think about them obviously, chat GPT, right, the way you describe to it, it tries to think, but it can’t.

21:34 So one thing we were doing is kind of like blending these image methods, and, you know, from, like, Gans, and then trying to blend it but pressure data, and so on and so forth. So one thing I do is I I I create some other problems before I Sol all of them. So one was, for instance, I was really not super happy with the way pressure data was being kind of like represented, right? So at some point I said, okay, just see which is the conventional ways what you’re seeing there, right, usually kind of flat ground. But this is not 100% real it’s kind of like all the data streams are coming from elsewhere, the moving foot is an interpolation, outso and insult doesn’t belong to that scan, and so on.

22:15 But this was my proof of concept, of kind of like showing hey, we can appro we can look at this problem in a different way, not only talk about what’s happening under the foot, but in and up towards the foot, and then maybe think about, like, the legs, muscles, and so on and so forth there, right? And then you can also deal with the what’s happening outside, and we were talking about, like, VR at the time. So I kind of like blend all these together.

22:41 When I saw Moon rabbit’s presentation yesterday that’s one of the ones I really love, because I haven’t seen anything close to kind of representing data in 3D that way, so congrats for that, which made me put this, so that’s good. The next thing yeah and also, like, when I deal with this, you know, one thing I started going to so I was hired to work on data to design streams for newels, and the first thing I said, there’s no data anywhere, right.

23:10 So that was one of the things I said, and that is to say again, just making us aware that it’s not like there’s there’s a data and we get collected it’s we projected, right? So this is pretty kind of like albertine way of you don’t set a frame and see the world, actually you project from your eyes. I want to see data under your foot, or on the building, so I can make sense of it, and computationally I can work on that, right? So these are kind of like great great questions.

23:37 This was a question again, like, you know, how can you bring art and architecture and Footwear design and so on so together and, you know, again, like, referring to back to my how do you convert this to Capital was a question. And, well, I I like being on the critical side, because hopefully you’re getting something from this, you know, questions I I want you to leave with questions, right? So that’s my only goal here.

23:58 So just just to demonstrate that what we did was this is a this is a 6m wide, 2 m High mural that is kind of like plastered at our in Innovation corner, now, at New Balance. So I kind of like push for this to be designed where is math? So math wrote the code for this, just just for generating this mural. So you you kind of put the image you could you dump images, you dump the shape, and just generate this 600 DPI printable 10 meter mural. So if you need anything like that, we can help with it just to show that, hey, this is the bread of things you can do, how can we make meaning by using that.

24:38 One thing I’m trying to say is I’m coming to the end, so I’m not going to hold you too much, I’m almost there. Computation, engineering, efficiency like making everything efficient, making everything data agnostic, making everything with the rules of the fragmented world if we are kind of like feeling meanings are, you know, getting lost as we compute things or compete with others, you you’re kind of like losing quite a bit there. So it’s always great to step back and say, Hey, I want to use computation, and I want to compete with people how can I preserve those meanings? I think that’s that’s very important.

25:17 One way to kind of like if you know what you’re dealing with, that helps the the the yellowish books are two good books that I suggested in my interview, which is which are about computation George Dyson, tourist Cathedral, and what computers still cannot do by Hubert raus those are wonderful. And the other two books are really useful the bluish ones are useful for understanding of meaning and information, like, how like, what is meaning, what is information, how they differ differ, and how can you again learn about those and bring it back to computation. I found them pretty useful.

25:48 If that’s too much, I prepared these little pills and wrote about them, so this is my medium account you can find on medium everything framed here is about AI, information, Pro information processing model of the Mind, how human braak brain works, how creativity Works, how this connects the computation, and so on and so forth. So I keep writing also it’s it’s it’s my journey of, like, finding answers and getting lost again.

26:17 One thing about AI is that the problem is generic tools, methods, or thinking want yield good design. So that’s one of the again, suggestions, maybe I’m making you may agree or disagree or if you’re looking for an average solution, or close to the age solution, good enough solution, maybe that’s good enough. But, like, some people like shooting for that red dot, which is elsewhere, right? And the problem is almost all all AI applications are generic, like, because because it’s it’s a statistical kind of like average of what you can do with that.

26:44 And I was super excited when I read Professor sigman’s interview, because he mentions about extrapolation and how AI is really Limited in extrapolating data. I was like, okay, like an engineer is saying what I’m saying as well, so I should use that for myself, so thank you. So, yeah, the goal is the Red Dot, the tool is AI. I’m wishing you good luck, right? So that’s pretty much that I don’t think you can land there, you can, but just using AI as a tool, you know, with your all all the other weapons you have.

27:15 Also, one thing is, when when we do, you know, computation, again like when we talk about efficiency and so on, we we get lost so much in the colorful pictures and details on how good the systems work, we negate ourselves like, stop doing it, right? So, like, talk about the tools and U let’s say processes, numbers, data. But I think the insight and input matter matters so much that there’s, like, there’s learning that we have to kind of like, you know, do by hearing from other people.

27:46 So I was writing some of this is becoming a chapter in a book and then I was writing about this, and I’m like, I have to I have to highlight the human the loop like, that’s like, you know, I started reading papers about it, because the human steps in, and, you know, tells it, this is the right way to do it, also. And then I was diagramming it out, it didn’t really work. I’m like, no, this is wrong human is the loop, it’s like there’s no human in the like, there are no Loops without humans, I mean you are the loop, right? So even even when you’re training in AI I had to write this, I was like, okay, human is Loop when you are training AI, there are no Loops that humans are not a part of, like if I mean, if I cannot think about a loop, then it’s not mine. I’m not the it’s, you know, that that’s sold any Loop that I can think of, I think we’re a part of.

28:35 And the other thing is, to all your credit again like, this was super inspiring for me, thank you very much for again, I saw all the presentations but one, unfortunately it was, yeah, my I was mind blown, you have already imagined or engineered what is attributed to AI today. This is external happening Eng generative AI for language models, for IM to 3D, which has come I’m going to talk about that later, not today, don’t worry about it. This is the last slide, I’m done not today, but, yeah, again, give give credit to yourself and just Embrace AI just is just another tool. So this was, for instance, just modeled and rendered, you know now ai is creating this kind of stuff, and everyone oh my God, you know, have you seen mid Journey? Yes.

29:21 Just to highlight that this is the last slide, last animation how you communicate this at a at at a company like New Balance, 10,000 people, now, is very important. Ai, and I happen to be, you know, like everybody came, okay, you are the AI guy. I’m like, I’m not the AI guy, I’m, you know, I’m computational designer, so, okay, it’s close enough, right? So you have to do something about, okay, I’m sure.

29:39 So it was more like picking the right tool, which is a AI rendering Tool that that uses designer sketch so instead of it, also use prompts, right? So you can use all this, like, text, the image, real time rendering, almost. But at the same time it’s very embodied, right, so, like, the designer doesn’t have to changes now, I have to describe it what, with words like it doesn’t work, right? So how you retain that, and also use this tool.

30:12 So what we did is we made people non-designers submit sketches, so we as all the non-designers to submit sketches picked one, and then I spent a day making this video, also just rendering, interpreting that sketch as a design, and just we we gave that video back to those people. I’m I’m just going to leave you with that, and thank you in advance. Let me get the sound right sound is oh, okay, all right, I’ll I’ll just I’ll just open it here. Thank you.