Generative Design: What Is It and How Will It Transform Our Industry?

PARDIS MIRMALEK: All right. So can everybody hear me fine? So, hi, everyone. Thanks for coming after probably a sleepless night last night with partying around Las Vegas. My name is Pardis, and we're going to be talking about Generative Design today.

So I'm a Design Technology Leader, Woods Bagot. And I started my journey from the east coast, got my masters of architecture in Princeton. Then I moved to New York and started working in a couple of startups and architecture companies. And then got super quickly frustrated with the inefficiencies that we have in our industry.

That brought me into some interest in tech and trying to get things more efficient. And then-- and it didn't even stop there. It brought me to a journey to go to get an artificial intelligence from MIT in order to be able to combine all of these and kind of be able to make our industry more efficient and more-- less of chasing our own tails, basically. And, yeah, so today we're going to be talking about how that combination and Generative Design could potentially revolutionize our industry. There's a stage right here.

So, historically, architects have been-- all of us have been doing the design in a way that we go through rounds of design iterations manually, and then 3D model it. And then go through taking a sketch to a software program like Rhino, do a 3D visualization of it, and then bring it into something like Revit, and do a documentation set and making sure these are all aligned. And then client has an opinion, and then we go design operation two, and design operation three. And then over and over and over again, sometimes going back to what we design in-- I don't know-- design operation one.

But the journey of design instead of doing that, the goal is to set the vision of the project and set the goals that we're looking for, and then find a way to translate that to the machine. And the machine that-- a massive computing power be the one who generates those design iterations. And us being able to explore those, and us being the agent that ultimately picks the design, the duration, that is the most optimal and works the best.

So today, we're going to be talking a little bit about what is Generative Design, specifically, targeted towards AC. What are the foundations of it? What are the tools and technologies and examples out there so that we can get a sense of how do we compare to the rest of the industries? Specifically, manufacturing, that seems to be a little bit further ahead. And by that, I mean, a lot further ahead. And what is our role versus the machine's role in this new ecosystem. And how should we potentially change the way that these structured teams and collaborate it-- collaborate across different companies-- and even the culture within our own company-- to be able to adopt this new technology.

So what is Generative Design? If you do some research, you will see that there are a lot of definitions out there, not necessarily always consistent with each other. But the one that resonated with me the most is this one. That Generative Design allows computers to explore solutions and creative partnership with designers. Now, think about the word partnership. And we're going to come back to that later in the presentation.

So as briefly discussed, Generative Design is less about having limited options that we manually come up with, and it's more about having all possible scenarios that we, or the machine, could potentially imagine that all the criteria that we define. It's less about going back and forth. When we come up with a design, checking whether it works or whether it is manufactureable, or fits within the budget, but having them prevalidated.

So when the machine runs into design iterations, us knowing that all of those actually, in fact, are constructable, are manufactureable, do fit within the budget. So it's less of chasing our own tail in that sense. And a lot of times, we go and design something for an option one, and then somebody has an opinion about, oh, well, how do we make this lead gold, and then there's a design iteration two.

And then there is some other budget concern that gets us to design iteration three. And then five weeks later, we have the client meeting. And the client says, but I kind of did like that design iteration one a little better. And that's fine. We usually smile and say, that's OK, let's go back to design iteration one. And there's not a problem there.

But the concern there is that we did all of that without really learning anything from it. We wasted our budget, our time, without really being able to gain any insight from all the studies that we did. But Generative Design, we're able to learn from all of that process, get that information, and be able to get insights from it. And kind of bring it into the loop and evolve the system and the process and the later generations of the design from it. So there's a good example of that in the Hackrod project that we're going to discuss a little bit later in the presentation.

What is not Generative Design? There are a lot of terms, like typology optimization, or lattice optimization, or evolutionary algorithms, that are not necessarily Generative Design. Could be tied into Generative Design in certain ways, but they're not necessarily specifically defined as Generative Design.

Now, think about the word CAD. They've been throwing this word around for decades now, and it supposedly means computer-aided design. But if you think about it, we have never truly used computers as an agent that aid us in design. We have usually a partner in a design firm sketching a conceptual design on a piece of paper. Then that's handed over to some sort of grasshopper, a guru, or an intern. And then there's a team of architects staying up late and developing that in Rhino. And then it goes to Revit. So mostly what we have been doing so far is computer-aided documentation.

But with Generative Design for the very first time, you're able to use CAD in its true sense and how it actually means, and using the machine as a partner in the design and as an agent that actually does, in fact, help us and aid us in the design process. Now, let's go through Tools and Technologies out there. And this is by no means a comprehensive list or anything that he can hold against me. But it's something that is a quick summary for those of you who are not very familiar as far as what are the tools and technologies out there across manufacturing and AUC. As far as those of you who do know, to just get a good review of what has been out there so that then we can compare why are we a little bit behind other industries.

So everything started with project Dreamcatcher. Dreamcatcher was a Generative Design tool that started as a research project in Autodesk labs. And every year that I was coming to AU, probably similar to many of the others, was really excited about this project. And whenever we would ask Autodesk when would this project get released, the answer would be next summer. And that next summer never came, and project Dreamcatcher went into sleep. But then later, it got released in something called Fusion.

But the general premise was the same thing, that we define what criteria we're looking for, and then the computer generates various design iterations that fits within that criteria. And then you're able to explore what actually fits our needs, or how could we potentially make the process better. And then it was supposed to be fabrication ready and manufacturability aware.

This brings us to be able to explore design iterations that are not necessarily something that we could have come up with, we could have imagined, because we have always been limited with our own rules of thumb, with our own imagination. And we couldn't possibly imagine something that could have looked like this.

Next, it came Fusion 360 that got released its Degenerative Design this August. Fusion 360 is basically a industrial design product design tool. Believe me, I've tried to use it in architecture. It doesn't really work.

And so how it works is that there is a Generate tab, and it could potentially look a little bit different-- different versions. But what it does is that it opens up to Degenerative Design capabilities of the software. And then you can define certain goals. It will tell you, based on how many design iterations you want, it will give you them and you can explore from-- let's say you're looking for weights and stiffness and different materials that could manufacture that piece with, and explore which one works for you the best. And in here, we're looking for a piece of an electric bike that, as you can look at it, it just is not something that you can potentially come up with with just traditional methods of design.

Probably my personal favorite is Project Fractal, because the way that it works is something that is very capable as far as 3D modeling that we use in architectural design. Project Fractal is something that works with-- how many people are familiar with Dynamo? That's a very good number.

So as you know, Dynamo is some sort of computational visual programming software that works with Revit. It has a standalone version. And if you want to use Fractal, only the standalone version works with it. The, basically, challenges are that there is no Python that you can use. There is no custom nodes that could be used, because none of them could be pushed to the web. So you're left with DesignScript and all of the out-of-the-box nodes in Dynamo.

But once you kind of go and overcome this obstacles, you can design something, send it to the cloud, and have project Fractal run into different design iterations. It looks pretty cool. It takes some time for it to come up with the design, but it definitely takes less time than it will take us.

So then you can kind of navigate through the 3D space and kind of like slide-- different sliders and define-- depending on what goals you have defined, see how they look like and filter amongst hundreds and thousands of design iterations. What you exactly want to look at. Like I want to look at the height between this much and that much, or the tower can [INAUDIBLE] that much.

And this is something that is really helpful as far as presentation and collaboration. And just imagine instead of bringing a huge laptop, or a printed set, or anything in that traditional nature, sending your clients a link like that and just look at them just lose their mind. And a side note, we did a-- who was at the design slam yesterday? Good.

So they did design a stadium in 20 minutes yesterday. And this event called Dynamo Design Slam. And this is the project that was used. And this was the project that won. But not this project, but the one that was designed by Project Fractal. So maybe if you have time, I can show you that later.

But essentially, as mentioned, it was the Dynamo Studio, there is a [INAUDIBLE] web functionality that then allows you to send that graph to the web and be able to use it that way. This is a project that we did in Project Fractal. This is a tower in Cyprus. And, originally, it was designed kind of more traditionally on a sketch, then in Grasshopper, then in Rhino, then we had to bring it to Revit because of the standards on what the client was looking for.

But the problem was even through Flux, the curvature was not transitioning correctly to Dynamo and Rivets. And it was easier for us to rebuild it in Dynamo rather than try to transition it from Grasshopper. And Fractal enabled us to kind of go through various design iterations and be able to see how the facade rationalization works, and how does different curves connect or not connect, and work or not.

And there are a lot of other pieces of software that could be tied into Generative Design. There's the space layout or previously known as Space Plan Generator. That is a package within Dynamo that could be used for Generative Design. It was a collaboration between Autodesk and [INAUDIBLE]. But it's a great tool. But the issue with that is that it's mostly a layout. It's not a 3D way of looking at buildings that is essentially the nature of the way that we do our work.

There is Thornton Tomasetti's Design Explorer. If you look at it, they kind of have the same nature of how they look like all of them, that you have different design iterations. And instead of having one design that we say this is the design that we came up with, and then two months later completely change it-- change it. And then three months later, completely change it again. You have a range, and we kind of start big, and then kind of limits its to the point that we get to the most optimal result.

And, obviously, there are a lot of custom tools, Generative Design is a really hot topic that every company almost that I've been looking at that has an interest in technology is really, really highly focusing on it. [INAUDIBLE] is one of them that is highly and intensively investing in Generative Design. And we have our own custom tools that do design-- enables us to explore design iterations. This is the same tower that I showed you earlier in Fractal and Cyprus.

As you can see how much it has changed the way that it looks like, that was one of these four towers. And we started like this but with kind of going through layout generations and explorations and ended up there. But there is a point of having all of this. Because it seems like across various industries, we are focusing a lot on Generative Design because of how much potential people across various industries see in it.

Now, let's go through some examples I want to start with something that is not architecture. So across various product manufacturings and, specifically, shoe. There's a lot of investment on Generative Design from Nike to Under Armor to Adidas. They are really focusing on using this technology. Why are they doing that? So if you think about it, and think about professional athletes and the people who run, this is exactly what they're looking for.

When you're looking for an Olympian that runs, what they're looking for is the shoe that is the lightest and the stiffest. But the challenge is that the stiffer the shoe, the more it will weigh. So there is a huge trade-off there. And that's why Generative Design is the perfect tool for them. Because then they can see and explore what are the points of the shoe that needs to be stiff, and we can put the material in there. And what are the parts that we can take the material away and make the shoe as light as it can be.

So this is the design that they come up with. And how much does that make a difference? Well, on a 100 meter dash, a shoe as light as this, could bring down the time of an athlete running that 2.1 second. Does that matter? It doesn't matter to me. It doesn't probably matter to any of you. But for an athlete-- for a professional athlete, it matters as much as the fact that it can turn into-- a fourth place person to a gold medalist.

So you can only imagine how much of an investment companies like Nike or Under Armour or Adidas, in collaboration with [INAUDIBLE], have been putting on this. One of the greatest examples is the bionic partition. And I believe we have the innovation manager worked on this project from Airbus, and the audience with us. It's Bastian. Say hi to Bastian. Hi, Bastian.

So this project was really interesting because, traditionally, this partition in Airbus 320 have been designed by a rule of thumb that, you know, ends up in a certain weight, certain stiffness. But The Living worked in collaboration with Autodesk, and they came up with using their custom tools on project Dreamcatcher and various other numbers of tools, to come up with a design that takes the same nature of what we're looking for in Generative Design and seeing, how can we take material away from certain parts of the Airbus partition?

And what that ended up with is a partition that is, basically, about half as light. It's 45% bring down the weight, and is as stiff as it needs to be. And why does that matter? So if you think about it, when a plane flies, it takes a lot of fuel. And that fuel is directly related to how much it weighs. That one partition, being half as light, it will bring down the fuel that is needed for an airplane, about 3,000 kilograms lower to what it needs to be. And that, essentially, is 160 metric ton of carbon emissions per year.

So since global warming is real, this is something that could be potentially very, very helpful, and something that companies do a lot of investment on. This is a project that I mentioned at the beginning of the presentation that has the potential-- shows the potential of Generative Design for us to bring in kind of creative feedback-- feedback loop and bring in data to be able to make the process better.

This is the Hackrod project. What they did is that they took a chassis of a car and went through the same process. Took away parts of the material that was not needed, but making something that is equally as stiff. And their vision is to use AI to create cars, and you be able to create your own car, kind of like playing a game or-- like a computer game.

They did the same thing, which is kind of similar to what carbon did, what Airbus did. But something more interesting is that then they made a prototype. They drove it in California desert, and they took-- they put some sensors in different parts of the car-- apparently, some parts of the driver's brain-- and they were able to bring back that information and gain insight from it, and make their process better for them to be able to make their product better.

AU last year was designed, the exhibit hall through Generative Design. It was kind of interesting because, traditionally, these exhibit halls are very symmetrical and very-- the way that you would expect them to be. But last year I was very excited by the way that this project was designed. I think you could kind of see the difference in navigating through the space and see how much more interesting it gets-- the space to be, and how much easier it is for you to navigate.

But, essentially, The Living used the same technology and the same tools to design the exhibit hall. I don't know what they did for this year's exhibit hall, but I was more impressed with the last one. The general premise is the same thing. We go through pregenerative design, and that's setting the goals. And then Generative Design that the software gives us iterations and evolves. And then post generative design that these are the agents who select what works best.

And this is the way that their software looks like. And you can explore and see different design iterations in it. And another really good example is the Autodesk MaRS office in Toronto. That is, again, another work with The Living. Something really interesting here that I want to mention is the first step is that they define what they're looking for. Because if you think about it, architecture is not manufacturing in the sense that majority of the time they are looking for something that weighs as less as possible, is as stiff as it can be, and is as cheap as it can be.

You have different goals that we look for. And then, therefore, the first step of the process, unlike manufacturing, for architecture is to be able to define those goals and see what they mean. And kind of define the scale of them and be able to translate them into the machine. And that is what they did.

So if you think about all of these examples, and all of the tools that we discussed, which was a kind of brief but a good enough summary, none of the softwares that we discussed is a commercially released software for Generative Design and architecture. Even the ones that we have is either like Project Fractal that we don't really have necessarily the capability to use Python, or custom nodes on. Or 2D layouts generating packages.

There's even a refinery that got released officially two days ago that I still have to explore and dive in to see how much potential it has. But, essentially, we still haven't had that much capability within our own industry to do Generative Design. Even in the examples as impressive as some of them are, will show that we haven't yet-- we are on the baby steps of figuring out how to do Generative Design in a realistic scale and architecture.

So why is that? When I think about it, I think the reason that architecture seems to be behind manufacturing and product design and Generative Design is mainly two things. One, is the variety of goals, and the other is the diversity of teams. Let me tell you what I mean by that.

So think about all of the examples that we discussed in manufacturing. The shoe, the Airbus partition, they had car project, everything that they're looking for is consistent. They're consistently looking for the weight, the stiffness, and the cost. All of them.

But now think about everything that we look for in architecture across various sectors to various projects. They look for visibility. We'll look for day lighting. We'll look for our path of travel. We look for noise. We look for a view. We look for so many things. And even we come up with our own terms, like, The Living define this term "buzz" for designing the 2017 exhibit hall in layout, which means that how much traffic and interaction is the same people at a certain point.

Or Woods Bagot as defining something called in betweeness, which means in a workplace, how close or far I am to my own collaborator. So if you think about it, it is really challenging for us to define specifically what we're looking for. And if the goal is to translate this to the machine and code and in programming language, this only can happen when it is actually very clear and very specifically defined.

Now, not only we have those variety of goals, but also we have our own interpretation. If I ask any of you how much visibility is acceptable in an architecture project, probably we have as many opinions as the number of you here. Or if you think about what is the noise level that is acceptable in a hospital? Or what is a good view? You probably come up that there's as various multiple options as there are people out on the planet.

And that is what creates so much challenge for us to be able to define certain consistent set of goals, like weight, stiffness, and cost, to be able to tackle Generative Design. So what needs to happen, I believe, is to for us instead of starting everything from scratch in our individual companies, setting goals, trying to define what I mean by visibility, what do you mean by visibility. And then trying to find custom tools and build custom tools, and kind of translate that to the machine, come up with a system.

Let's think about, for example, an open source project. And everybody instead of doing it from scratch and individually, contributing to that open source project, and then come up with a standard. Kind of similar to when BIM started. Everybody was excited about it. Every client was asking for BIM but they didn't know what it means. And then we started coming up with local and national standards for us to be able to understand each other and what we specifically mean when we say BIM. What is a low D? What are we looking for?

So we should go through the same process of defining on a granular level what are we looking for in Generative Design? Now, the next challenge is diversity of teams. Let's go back to the examples of manufacturing. The shoe, the bionic partition, and the car if you think about it, the teams behind each of these projects were just one company. This year was the Nike. The bionic partition was the Airbus. And the Hackrod project was the Hackrod company.

Now, think about any architectural project that you have worked on in the past however many years. And now think about the teams that are behind that and the sophistication of the collaboration and the flow charts of the hierarchy of teams. You will probably come up with a spaghetti of very, very complex flow charts of how they collaborated, and how we have different politics. We have different skills. We have to define certain people to do certain things. And then we have our own, obviously, local enterprises and minorities that tie into the system, too, which makes it hugely sophisticated and complicated.

As kind of a funny example, this is a real project. It's a real flowchart. And it's a half a million square feet project. This is the collaboration between the teams on this project. And it doesn't even fit in one page. And it's obvious that you can see how sophisticated and highly-- kind of like spread apart it is. And if you see that circle, it's kind of funny, you don't really see what it says. I'm going to blow it up. But it says discrete works scopes, ramp level toilet rooms by others. This is the level of sophistication and detail that we go into defining certain parts and scopes about architecture projects and giving it into certain different types of people.

But that is super challenging because, obviously, that is the way that our industry is defined right now. But the problem is that is we are not working under one umbrella of vision or business or IT. So if anybody decides to do Generative Design, there is a whole dozen of other teams that don't even know what it means, might not even have the tools, and might not be able to do it.

So when that is the case, they're not going to be able to use Generative Design effectively. I mean, think about if you start a project in [? 11/2016. ?] And two years later, the client says, let's upgrade the model to 2018. This is a real story. It took us about eight months to, on the same project, decide if you want to do that or not. Because everybody has to assess, go through their leadership, asks about their IT whether they can do it, they cannot do it. How long does it take?

So this is, essentially, the same nature of challenge that unless we're working on the same umbrella of business and vision and IT strategy, we're not going to be able to take on and adopt a new technology. This is me trying to simplify that spaghetti. But I think I could somehow turn the spaghetti into an octopus. I don't know if I made it any simpler or more complicated.

But, essentially, it's the team structure on that project. It's a design project. There's a contractor, there's an architect. We have our own set of consultants. The contractor has their own set of, basically, subcontractors. And I think maybe something that would help us take on Generative Design more easily is to bring everybody together, think about more having consultants in-house. A lot of these companies are really small companies that are one or two people working on that massive project.

So if we're able to have a lighting designer in-house or assign each consultant or an acoustic in-house, then it will enable us to make decisions more quickly and more effectively. This is why companies like Katerra are super successful and making such a huge impact, because they're not looking at architecture like the traditional way that it's been looked at. But like they are looking at it like a Toyota company and like the way that they look at the design to build, to supply change, and how they can make it more effective is more through the lens of manufacturing rather than architecture.

So what is our role in this new ecosystem? As we went through all the examples and all of the processes, it's unlike before. It's less about us drawing the design and then trying to make sure it matches the documentation, and then changing our mind, and then going back to another design. And then as much as time and budget allows us to do that. It's more about setting the vision and setting the goals, and then finding a way to translate that into a way that the machine would understand. And it will give us those design iterations, and us being the one that picks the final result.

So who is the designer? Is it us or is it the machine? So this is usually-- when I get this question, it's usually from people who are not very happy with the way technology is going anyways. And they're usually very, for justifiable reasons, concerned that their job and thinking, well, there would be a time that the machine will take over and there's this terminator post apocalyptic moment that AI and Revit and Dynamo took over the human. And I don't know, just skeletons laying around.

So it's something like that. We even have codes like this that artificial intelligence is our biggest existential threat. Does anybody know who said that?

AUDIENCE: [INAUDIBLE]

PARDIS MIRMALEK: No. Hawking is against the [INAUDIBLE], but this is not from. Anybody?

AUDIENCE: Elon Musk?

PARDIS MIRMALEK: Yes. What's your name, sir?

AUDIENCE: Matthew.

PARDIS MIRMALEK: Matthew. Matthew is right. This is from Elon Musk. And you can only imagine if a guy who was trying to create a colony on Mars says something like that and is so suspicious of the technology, what you can expect from and a partner at an architecture firm.

So will the machine take over? I'm here to tell you probably we don't have to be as afraid as we are. In the program and the certificate that I got from MIT and AI, I learned that there is a specific distinction between narrow AI and general AI. General AI, or human level AI, is the post apocalyptic terminator kind that the machine has its own feelings, it comes under the vision, and then decides maybe to potentially take over the planet and maybe keeps us as pets.

But now AI is specific tasks that we ask the machine to do. Like image recognition, or self-driving cars, or even a simple auto correct on your computer. So we don't have to be that afraid probably. But a little bit afraid. Because people have been predicting for human level AI to happen in the next 20 years. That's scary. It's also wrong.

People have been predicting for human level AI to happen in the next 20 years for 60 years. So it's been over a cent-- half a century that we have been thinking, oh, my god, AI will take over. The planet as we know it is over. But, you know, time went by, everything was fine. And as far as we're concerned right now, human level AI is not going to happen in the next decades. And we don't have to be worried about that at all.

So this is where it's more about a collective intelligence. It's less about a threat when it comes to the technology, or AI, or Generative Design, and is more about, as mentioned, us looking at them as our own partners, as somebody that helps us with the design. And us being able to do what each of us are good at and have a collective intelligence that brings the level of work to the next level.

It's less about the machine versus us, and it's more about how could we build that partnership and use this collective intelligence to raise our own bar? So I want to leave you with a very good example of that. The year was 1997. IBM Blue has created this highly sophisticated, most intelligent computer of its time called the Deep Blue.

The Deep Blue was extremely sophisticated, very, very intelligent. And as any other computer of its time, the way to measure that intelligence was for it to play chess. So Deep Blue was put in front of a world champion in chess, Garry Kasparov, and they play chess. And for the very first time, the machine was able to beat a world champion in chess-- in chess tournament nature.

This was hugely publicized. Kasparov was devastated. And there were numerous articles and documentaries. One specifically called The Man vs. The Machine. And that's horrible. But then time passed, Kasparov got over it, and then he started realizing the potential, some sort of a partnership. And the potential that the computer could bring to his industry.

So he started creating a game that is called Advanced Chess, Centaur Chess, and Cyborg Chess. A lot of you probably know about Kasparov being beat by the computer, but you probably didn't know about Cyborg Chess. Kasparov came of it, again, that instead of human playing against a computer, it instead partners up at a computer, and then plays against another team of the same nature-- of one human and one computer.

And people around the world have been playing Cyborg Chess and have said that it raises the level of chess to an exciting level that was not possible before. This is a piece of the magazine from New Scientist August 1972 that says how much there's an interesting possibility that arises from such a partnership, and how much if a human wants to play, we can only imagine five moves ahead, 10 moves ahead, but not a lot more.

But with this game, we can say I want to make this move. And then the computer explores what happens 500,000 moves down ahead. And it is able to give us with its massive computing power a massive number of options that we can have, and us to explore how does that work for us, and which one works best? And us being the final person that makes that decision. Then that's super similar to what we discussed about Generative Design.

So the gist of that article is that an interesting possibility is the introduction of a new brand of consultation that the partnership is between man and the machine. And that is where I hope we will be within our industry and, hopefully, in the near future. Thank you

[AUDIENCE APPLAUDS]

So at this point, if there are any questions, you can ask me or you can come say hello and introduce yourself and give me feedback.

AUDIENCE: [INAUDIBLE]

PARDIS MIRMALEK: I'm supposed to repeat their questions because they're recording. So you mentioned if it is possible to get a recording? I will upload the material on the Autodesk website, yes.