Generative Design – Is it the next stage in the evolution of design?

BRUCE BLAHO: Bruce Blaho, thanks for coming out this morning. So I'm with HP, a workstation group in Fort Collins, Colorado. So I'm a chief technologist for workstations focused on new technologies, new innovations, new ways to grow the business. In particular right now, I'm focused on AR, VR, machine learning, and generative design.

FRANCESCO IORIO: Thanks, Bruce. Hello, everyone. My name is Francesco Iorio. I work in the Office of the CTO in Autodesk. I am part of the Autodesk research organization. And I lead a team which deals with computational science. So I am partly to blame for a fair amount of this, as I started actually the generative design effort at Autodesk that's called Project Dreamcatcher, and that now is actually spearheading a fair amount of the generative machine learning activities that we see here. So sorry and thank you.

[LAUGHTER]

SCOTT HAMILTON: Hello, my name is Scott Hamilton. I am an industry strategist at Dell, in the workstation group as well. So I focus on the workstation business and our partnerships and strategy for the engineering, manufacturing, and AEC industries. And I'm actually a longtime employee of Autodesk. I was there for 10 years before I was at Dell. Nice.

SCOTT RUPPERT: Nice. Good afternoon, I'm Scott Ruppert. I'm the workstation portfolio solutions manager for Lenovo. So across our range of workstation, desktop, and mobile. I really focus on emerging technologies and where that intersects the professional space. So really the same as these guys, right? A/R and VR, artificial intelligence, deep learning, and how that applies to our professional workstation customers.

DEBRA: OK. So yeah, a list of questions. These are some really popular questions that most people have about generative design. But the key is we really want this to be interactive. And you can see how interactive I am, right? So don't be like me. But if you have any questions, they may have the answer, just please jump right in. Or you can save it to the end for our Questions and Answers at the end.

OK, so number one, we're going to start with Frio. First question, how do you think the vision of generative design will change the world of design?

FRANCESCO IORIO: Good question. So OK, well, this is clearly a kind of tricky question, right? As we heard this morning, Andrew said there has been a lot of tension between the negative aspects of automation and the positive aspects of automation, right? So what I strongly believe, and this is one of the main reasons why this project and this process actually started, is that throughout human history, tools have really kind of shaped human civilization. So most of the time, the discovery of new tools and new technologies have advanced the human culture, human possibilities time and time again. So again, we mentioned, of course, some main eras of our history in the presentation.

So I believe that we are now at the point where the technologies that are behind materials that are behind manufacturing that are behind what is actually now possible with electronics, for example, and others, for all those opportunities, the tools themselves are now becoming the limiting factor. So as in the creativity of people, actually, is being harmed, actually, by the fact that these tools were designed many, many years ago for very different manufacturing options that are now not obsolete, but essentially, they have been surpassed in several capacities in their capabilities.

So when we now think about how to give people the opportunity to really exploit these new generations of technologies and materials, there needs to be again a leap forward that changes the way we interact with computer systems in this case. So the reason why I could really trust this is that the leap consists mainly in taking the design tools not as glorified drawing boards, if you will, that have been actually for the past 40 years, including the tools that we sell. But rather, actually giving the possibility to people to really have a dialogue, to really have a partnership with these design systems.

And so that, again, to go back to the original point, that is not to replace human designers or engineers. But this is really to enhance their capabilities, to remove some of the biases that we have inherently because of our training, our education, our past, our legacy, the books, word of mouth. So some opportunities have actually come to life hundreds or thousands of years ago. And then they have largely remained the same, which tends to form a culture around it, right? It's a bias.

So these types of design systems actually have less of that. So they can explore possibilities and opportunities, combinations of factors, of materials that we wouldn't think of, necessarily, or that are beyond our imagination, not necessarily our imagination, unless you're a genius, and even then.

So to end, the reason why I really, truly believe that this is the next generation of design, or that will enable the next generation of design is because it will allow us to progressively move from, again, pure design as an expression of a solution that we already have in our minds, and that we're using the tools only to document it, in fact, to an era where it is about decision making. It is about that deeper understanding of the problems that we're trying to solve and having instruments that allow us to explore a much broader set of opportunities, and then allow us to actually express our power in terms of decision making, even more so than in pure drawing or drafting or design itself.

DEBRA: Anybody want to add anything or any questions on that? OK, question number two is for Scott Hamilton. In generative design, with IA, your final design could have a compilation of multiple data points. In this scenario, who do you think is liable for the full, final design?

SCOTT HAMILTON: I have to ask the lawyers about that. I don't know.

DEBRA: I'm sure they'd have plenty to say.

SCOTT HAMILTON: I think that introduces some interesting questions, because you can now have people collaborating. Are you going to blame it on the Dreamcatcher software? I don't know. I don't know. So I mean, I think the true answer is it depends on who's doing the design.

Obviously, if you're an engineering firm, and you're designing a product, and you're utilizing that technology, you ultimately take the responsibility for the product that you create, because going back to you're using the tool-- it becomes a tool. And it helps you make decisions. But you're still responsible for those decisions.

It's like I remember when I was younger, the adage was with the calculator, garbage in, garbage out, right? So you have to be responsible for the garbage going in or the right stuff going in, making sure that you're not just taking a computer's word for the end result. You still have to be a savvy engineer and understand the result that's coming out of that, and making decisions on whether you feel that's appropriate or not, whether it's because potentially you gave it bad information. It still came up with a valid answer. Or maybe it didn't come up with a valid answer in the first place.

So I think, ultimately, the engineer and the company that's responsible they've got to take the full responsibility for that design but I'm sure it will bring some interesting legal discussions when we get full force into using the technology.

DEBRA: I know we were talking when we created these questions. Would there be a requirement for new regulatory guidelines and rules? And that would be interesting. Anybody have any thoughts on that? Because I'm sure there will be, right? Because any time that there's some new technology and something that we've never done before, and a lot of times, our forefathers who created the laws didn't necessarily think we were going to be doing AI one day, or generative design. So--

SCOTT HAMILTON: Yeah, I think we'll definitely see that happen. In fact, I've already seen one evolution of that happen if you just look at all the products that we have out there to do computational fluid dynamics finite element analysis and everything there are companies now that are using that technology. Rather than building physical prototypes and doing real world tests, they're using a lot of that technology to get FDA certifications, and all kinds of stuff. So there's already been one evolution of that, certainly. And no doubt, there'll be another one for generative design.

BRUCE BLAHO: I'd like to pile on if I could with that. I think in addition to the design correctness and who's responsible for that, the other thing that was a real eye opener for me as HP got into 3D printing was on the security aspect. You have a whole new realm of security liability. So for us as manufacturer of a 3D printer, what if somebody hacks the data after when the job is submitted to us or when it's resident on the printer or at some point. What if somebody gets in and hacks that?

And only now, instead of causing a denial of service, causing cloud service to go down for an hour. It causes a plane to crash. Some people have already demonstrated-- I think I saw some YouTube videos that some folks had hacked parts on a quadcopter, that part looks great. It looks perfect. It functions. And after about 20 minutes of operation, it fails, and then crashes the drone. So I think that's the other dimension. It's like, wow, you need end-to-end security. It's a whole new challenge in terms of end-to-end security from when you did the design to when it's submitted to wherever you're going to print it or manufacture it, and on the machine. So lots of fun.

AUDIENCE: [INAUDIBLE]

DEBRA: State your question. I'll give this back. Bruce is next, and his question is, what other products are needed to make generative design a viable product? And we're going to combine that with, would we be using the human creativity, the ability for humans to create, once machines are creating all of our designs?

BRUCE BLAHO: Thanks, Debra. Yeah, this one, when I first agreed to be on the panel, the first thing that sprung to mind for me was that it's a real fusion of multiple technologies and multiple trends. So generative design is really-- to me, the first thing I think of is it's, I think, completely married to the whole additive manufacturing 3D print. We're looking at next generation 3D printers. We'll not only do multiple materials. But you can do voxel level control. You can vary the strength, the flexibility, the color, the electrical conductivity, the capacitance.

One day, we sat down and tried to figure out how much is the data going to explode in the future. And you can easily rattle off dozens of parameters per voxel that will be varied someday in future printers. And the immediate inescapable conclusion is there's no way a human could, nor would they want to try to manage that level.

And you can do very clever things in additive manufacturing and with generative design to do things like gradients on strength. And you can basically use that material control as a new design element, as opposed to geometry or other things. But how on earth could you ever do that? And the answer is generative design. I can tell you that every company making 3D printers is banking on generative design.

There's a lot you can do today without manufacturing, with the existing tools. And that's always how digital revolutions go. In the first early generations, you do wonderful new things. But you have all the old tools. You're solving all the same old problems But with this great new tool. But when it really blossoms and gets its wings is when you've got the new tools that come along. So I think we're all, yeah, very grateful to Frio and his compadres for leading the charge on that.

And then the second part of the question, and it gets to the keynote, I just wanted to comment-- and Frio touched on this already-- this one, to me, I don't feel at all as far as stifling human creativity. I think it's just the opposite. Hopefully, what I just said a minute ago about it opens up a whole bunch of new design tools. Things like machine learning and generative design mostly, I think, will get you out of the drudgery, and I think let you be more creative.

One, it can take over some of the jobs that you probably didn't like. Who enjoys the details of drawing lines and filling things in. A lot of the drudgery can be automated. That's a good thing. You get to spend more time on your creativity. And second, and it opens up whole new tools. Like I say, there's whole new ways that you can create and let you go through the cycle much faster, essentially. I think it's just the opposite. It's a really good thing for human designers.

DEBRA: OK, Scott Ruppert, you're up next. What are the most compelling enterprise use cases for generative design with AI?

SCOTT RUPPERT: Oh, wow, all of them. Actually, I think as I move into that question, I was going to segue off of what Bruce was saying, I really think as hardware providers, we all ought to be sitting up and taking notice when Frio was saying that the tools aren't keeping up with your creativity, with the ideas. And so I really think it behooves us as hardware manufacturers, as integrators of that technology, to stay close in step with Autodesk, make sure we're making the tools available and easier to use.

And then I think that opens the door to a lot of different enterprises. I think Andrew did a great job of answering my question for me this morning in the keynote. The Van Wijnen story was amazing. I've got a little story here coming up of another architecture VIM, kind of smaller scale, of generative design.

So personally, I'm seeing it grow there first, I think. But I think there's so much opportunity in really any kind of-- I don't want to just say design but-- any type of design, any manufacturing, automotive, aerospace, but especially architecture and VIM. I don't know if you guys want to pile on any examples there.

FRANCESCO IORIO: Sure. No, absolutely. I mean, I couldn't agree more. So I think that there are some areas where generative design and the combination-- I don't really make a massive distinction between AI for design and generative design much. Like for me, it is a collection of tools to again to augment human capabilities, really, to surpass our limitations.

So but this combination is very clearly applicable to some domains and less immediate in the short term. But I think, overall, while in the short term, actually, aerospace or-- enterprises where the quest for performance, like the do more there with less. So that the more it's key to a particular industry, and the more we will likely see adoption of these technologies.

But a very concrete example. In architecture, for example, is that architecture is a field where, as opposed to what we use as an example, the space of all shapes, so essentially the combination of all the materials you can think of in a volume that defines an object or a mechanical system, largely the way we create VIN models, for example, for buildings. Again, I'm not in construction. So excuse me if I'm actually saying something actually that's completely inappropriate.

But the universe of objects that compose a building is relatively more finite than essentially the molecules or the voxels that can be tweaked in a multimaterial 3D printer to make a multifunctional part or system, so if you will. So let's say it's a slightly more tractable problem in the short term. So even though we have invested a fair amount of energy and effort in the mechanical engineering side, like with Dreamcatcher, so that's because it was a harder problem. So we wanted to get ahead there before actually reverting back to architecture. But architecture and construction are definitely fields where we'll see application of generative design very soon, very, very soon.

BRUCE BLAHO: I'll pile on a couple of things I thought of while you were speaking, Frio. As far as-- you guys have been working on generative design for many, many years. It predates, really, the emergence of deep learning. This really just happened in the last five years. One of the things that really struck me then is how the Autodesk team, the Dreamcatcher team is applying deep learning inside of their generative design tools. So you don't necessarily see it out front and center.

I give two examples, two of my favorite use cases. One is at the front end and is maybe more obvious. The generating of multiple design options, like the AC example of the neighborhood planning example this morning in the keynote, where it's like given these design constraints, here's all these different options.

The other thing that's maybe a little bit more buried, but is one of my favorite applications of deep learning is in taking things that used to be very time-consuming and lots of computation in getting a quick approximation. So in the case of generative design, doing physical simulation. So usually simulations take a long, long time. And it's hard to integrate that into you have a simulation driven design. It's very difficult, because it's so time-consuming.

Well, now companies like Autodesk are using deep learning to basically approximate the answer that you're going to get to the simulation, whether it's find an element analysis or computational fluid dynamics. You can get a quick-- I won't say quick and dirty. I'll say a quick and reasonable, a quick and good enough answer to know is this thing going to function from an engineering point of view to iterate on the design multiple times. And then, of course, at the end, go ahead. Do the full on simulation. But I think that is so empowering. That's one of my favorite use cases.

DEBRA: OK, we will go to our next question. Frio, there are multiple approaches to design. Why have you chosen generative, as opposed to democratization?

FRANCESCO IORIO: So I thought this question was actually interesting when I read it, because the way I conceive generative now-- we conceive generative design-- is probably one of the ultimate forms of democratization. So consider the power that people, essentially anyone using advanced generative design tools will have when virtually behind the screen, you have an army of engineers, of designers, of architects, of structural analysts, of fluid dynamicists, of electromagnetic analysts, et cetera.

So the knowledge that we are embedding into these tools and that will ever expand-- so as we move forward, we actually increase the capabilities and the knowledge by means of machine learning. For example. So these tools actually will learn to solve these problems better and better over time.

So these capabilities will be available to essentially everyone. So something that currently is maybe only the domain of enormous organizations with extremely powerful engineering studios, if you will. So it will be at least to some extent achievable by most people. And that's kind of an interesting forum.

So for me, it was not a choice of whether to choose generative design, essentially, as a form of expression versus essentially democratization. But it was actually, rather, trying to join them, so to understand essentially the possibilities that generative design unlocks, again, is kind of an ultimate instrument that will be available to everybody.

AUDIENCE: It's a way of democratization.

FRANCESCO IORIO: Yes, yes, exactly, exactly, exactly.

AUDIENCE: I have a question that kind of relates to this in a sense, next to it, which is--

DEBRA: Hold on. Hold on.

AUDIENCE: --if we can touch on the difference between I want the generative tool to help me think. I don't want it to think for me. There's a difference.

FRANCESCO IORIO: So there's multiple levels of thinking, right? So generative design shifts your thinking from thinking all the way into the solution into thinking all the way into the problem. So with a generative design tool, you're not expressing a solution that you already come up with in your head or that you read somewhere. So you're trying to define the problem that you're trying to solve in a way that a computer system can help you, to assist, that can assist you to solve it.

So some of the thinking, yes, will be done actually by the computer system itself. So some of the solutions that you will see you probably may never have thought of. But the point is that that is what gives you the intuition. So the purpose of the generative design tools is not just to present you with the solution that perhaps you don't understand. But it's to give you the insights into how does the problem behave. So what are the opportunities? What are the challenges of solving that particular problem?

So it's only a fraction of the thinking. And what we're hoping is that it is mostly the fraction of the thinking that you wouldn't have thought of anyway. So the other part is still a symbiosis. And in fact, actually to kind of riff a little bit about what we were talking about before, one application of machine learning that will come in the future of generative design is in the very relationship between yourself and the generative design tools.

So currently, the only way we can communicate these problems is by keyboard, mouse, and a display monitor. So we believe that maybe other avenues of communication so that unlock a different type of bandwidth of communicating your thought processes, rather then click on a menu. So maybe click on a menu is the best thing there is for a particular type of information that you want to convey or that the computer actually wants to convey back to you. But maybe not.

Because again, we're actually not in the business of drawing lines anymore. But is in the business of discussing a problem, discussing a problem. So that actually leads to a whole different category of interfaces, if you will, that machine learning will power.

AUDIENCE: It's going to be a discussion between you and your team.

FRANCESCO IORIO: Absolutely.

AUDIENCE: It opens up the thinking.

FRANCESCO IORIO: Yeah, between you, your team, and a virtual team that sits behind the screen.

SCOTT HAMILTON: Yeah, and I guess part of that is you have to redefine the problem from the get-go to make sure that the machine understands the real problem, which might cause you to think about it in a completely different way than you would if you were just solving it yourself, because I guess you inherently know what you know. And now you have to communicate, like anything.

It's like being a teacher. To be a good teacher, you have to be really good at the subject to--

AUDIENCE: [INAUDIBLE]

SCOTT HAMILTON: Yeah, exactly.

FRANCESCO IORIO: I hope that answers--

AUDIENCE: Yeah, yeah.

BRUCE BLAHO: So we need a webcam that reads your emotion that would know if you like the design or not.

SCOTT HAMILTON: (LAUGHING) Exactly. If you have a smile, it's all good.

DEBRA: OK, next question is for Mr. Scott Hamilton. How would you recommend implementing generative design with IA, would it be one of my favorite things, a workstation, a server? What would it be?

SCOTT HAMILTON: Well, you could guess I have a preference, right? Well, I look at this it's very much like the discussion that I think has been occurring over the last few years about virtualization technology. There are great uses of it. And there are some drawbacks, right?

I mean, first of all, if you look at a service-oriented approach like Autodesk is doing with Dreamcatcher, you almost have unlimited compute resources that you can access. And it's very scalable. So that has a lot of benefits in itself. You can solve a small problem. And if all of a sudden that problem gets really big, you know you've got the resources to attack that problem. So that's the plus, I think, for the service side of the argument.

Of course, as Bruce mentioned earlier, security is a huge concern. And a lot of Dell customers, we've been offering virtualization technology for a long time. And we were a little bit surprised. We thought that was going to take off a lot faster than it really has in reality. And really, the main concern has been security.

And we've all heard the stories about Sony getting hacked and all the people-- it's just every day now. And so I think people's level of uncomfortableness or acceptance with security has completely changed. So that's certainly one of the benefits of having the workstation or the compute resources local, at your desk or within your own firewall. And that's where workstations can provide a huge benefit there.

And then, of course, you've got to look at the different type of problems. I think there's a balance between, how much data do you have to feed the problem? And how can you get all that data into the cloud versus providing it to a local workstation? We have a great example of a deep learning application that's running in our booth, if you guys want to come by and see it. It's called real-time style transfer.

And that deep learning application is taking a real-time video feed that could be just a video camera. It can be a still image. It can be what's happening on your computer screen. And it actually converts it into different artist formats, like Monet or Van Gogh or-- well, I learned when I was in Amsterdam, it's actually Van "Gog" is his real name, not Van "Go," like we call him.

But anyways, so that's doing it in real time. And the amount of data that you're feeding into that at any given second is tremendous. And so having it in the cloud is maybe not the ideal solution in that problem. Now there are many other problems where, again, I think the cloud is a good solution or as a service.

SCOTT RUPPERT: If I may just add to that, I think my take on what you're saying is that it doesn't have to be an "or." It's an "and." So it's know start at the workstation. Scale to the data center. Scale to the cloud. What popped into my mind is none of us are in a one size fits all business, right? So I think we need to work closely with you, with Autodesk, understand the problem, understand the challenges, and help fit the hardware. Make sure the tools meet what you need them to do.

FRANCESCO IORIO: Great summation. Thanks.

SCOTT RUPPERT: Thanks. You set me up for it. Thanks.

DEBRA: Any questions? We have-- oh, go ahead. Hold on. I should've worn my roller skates. What was that?

SCOTT RUPPERT: We need one more mic just floating around.

AUDIENCE: In the field of architecture and construction, what software platforms have you seen successes in generative design? And what are some potentials you see coming down the pipe?

FRANCESCO IORIO: Is it?

SCOTT RUPPERT: Yeah.

FRANCESCO IORIO: OK. All right. OK, so [CHUCKLES] OK. What are we seeing, actually, in the field? So what platforms. OK, so what we've seen so far is mostly individual organizations that essentially built their own tool chains, if you will. So the experiments that we've seen successfully deployed are very ad hoc. They fall-- they're molded around a process that is very bespoke for a particular organization, which is perfectly fine.

The challenge with that, of course, is that Autodesk is in the business of making general availability software. So it's somewhat harder for us to create workflows of that kind unless we give a massive amount of our customers' ability. So I think, again, in the future, people will be able to leverage several practices of our generative design stack by using Forge.

So the idea would be that rather than essentially rolling your own stack that just works locally on your cluster, for example, and that follows a very bespoke model, and that has also, built in, all the tools that you need to perform general design, which is sometimes prohibitive for a small, medium organization to have.

Again, the idea behind Forge is specifically you only create the pieces of your vertical that you need for your business or to create the vertical that you can then resell, if you will. But most of the heavy lifting-- the analyses, the generation, the machine learning or at least the parts or the portions that are in common between several workflows that will be exploitable on Forge.

So it's challenging because, again, we've seen it numerous times. And these processes are used for necessity. They're so bespoke that the heart to transplant. So they're successful in that particular organization. But it's very difficult to actually transplant them somewhere else.

AUDIENCE: I beg your pardon for my English. Actually, I'm Italian. And yeah, I would like to know more about what kind of framework, let's say, with deep learning can interact with Forge, because we try also TensorFlow, Keras, Caffe, Theano, PyTorch. There all different kind of framework.

But the most important question is, what kind of data? Because we talk about the generative design, actually. But I would like to call like learning design, because we need some information that the machine actually can make from a good prediction, can learn from a good prediction.

So the information inside the model should evolve accordingly with the customers and also the architects. So in this way, people can automate and see how they're doing during the process of design. So how we can achieve? Because Forge is a great platform, let's say. And so Dynamo is not. Actually, it has bottlenecks with our own Python, with NumPy. So which one data-- or what kind of data do you suggest to apply for this kind of prediction?

FRANCESCO IORIO: OK. That's a very good question. So I think this is actually something that we're struggling with ourselves. So you can imagine creating a generative design system as a service, actually, will eventually create lots of data. And how we manage it, how we ask our prospective customers to share all of it with us, some of it, none of it is very tricky, because that will affect our ability to learn from it. And it will also affect our ability to further share it with everybody else who wanted to run their own predictions, so their own models, or create, actually, their mathematical models.

So deep learning is a very tricky business, or at least-- the technology is improving, but still, to actually create complex, non-linear models, it takes a gigantic amount of data, a really disproportionate amount of data, like enormous so to actually have some meaningful models. So whether that is something that Forge itself actually can provide, I actually don't know. It's a business decision, actually, from Autodesk's point of view.

So something that I would be amazed if it didn't happen would be that for companies, organizations using Forge to essentially create their own internal verticals and to create their own systems, software design, of course, they will have to be able to reuse all the data that they produce by means of leveraging generative design to do whatever they want. And so having TensorFlow in the cloud or having it access the data, actually, in whatever matter, it is just a technical issue.

So I don't have a preference. For a stack, I use TensorFlow most of the time. But that's beside the point. What I mean is that the data will be accessible by some kind of standardized means. And then what processes then you use to model whichever property you want actually out of your data and then reuse it.

It will be either a problem of improving your vertical progressively, essentially creating your application in a way that what you learn from the data that you generate is then reused directly in your system. Or, essentially, you wait for that to emerge from the platform itself. So Dreamcatcher, all of those generative designs and such, whatever, will be the names of the products that will make-- actually, for both architecture and mechanical engineering, will learn from the data that they produced, and from the data that the customers actually feed.

So how to access that directly? I don't know. That's a very good question. As in like, if you want to access directly what we have learned or we're learning from it, then you may require some specific APIs, writing Forge, for example. Or if you actually want to feed some pre-trained models, or you want to train your models yourself, actually, on that data. I think they're just different technical challenges. But none of this actually is impossible to achieve.

So of course, again, the bigger issue actually will be what data will be available, how much of it, and from where and for what purpose. So we are going to face the exact same issue [LAUGHING] ourselves. No, it's a very good question.

AUDIENCE: [INAUDIBLE]

FRANCESCO IORIO: Yes, of course.

BRUCE BLAHO: I wanted to just add just a little bit to that. I think deep learning systems, networks tend to be embedded in these products. So the generative design, they're multiple, I had mentioned before. There's one network that's helping the upfront design choices. There's another one that's helping with the simulation work.

I suspect that what's going to happen over time is that as much as people have taken CAD systems today and added their own IP and their own modules that are particular to their line of business, the same thing is going to happen here. So I suspect that some of you will be creating your own deep learning networks that will augment and use. You're going to build on what Frio's team has done and probably create your own, in which case, then the answer to your question is going to be, well, it depends. The data that you need is going to depend on what kind of problems are you trying to solve. And I suspect those are the companies that are going to really take off, that can figure that out the soonest.

DEBRA: So we have one final question. That belongs to Scott Ruppert. But before he does that, because he has something he's going to set up after his question. You can talk into our example. We really wanted to have a real life example for you. So after Scott finishes, we have a video, quick. It's like a minute and 10 seconds of an actual real life example of generative design.

So Scott's last question is, what industries do you see impacted first, automotive design, architecture, construction, or manufacturing?

SCOTT RUPPERT: Hmm, all of them. Can I use that answer again? I don't want to lead the witness, actually. I kind of have my-- we talked a little bit about seeing it growing in all of those industries. But I'm more curious to hear from you. You're all here for a reason. So what industries do you see it coming, want it to be coming? Any that weren't on that list? Certainly AEC, or aero, architecture, construction, certainly product design. But in my mind, it's larger aero, auto, consumer goods, other manufacturing practices. What else?

DEBRA: Any thoughts on where you see it coming first, what you're seeing out there in the real world?

AUDIENCE: Manufacturing, because you go there different industries, like aerospace, medical.

SCOTT HAMILTON: Sure.

AUDIENCE: Oil.

SCOTT RUPPERT: So really just any industry, just across manufacturing, across that spectrum. Right.

AUDIENCE: I don't know the relationships between generative design and the [INAUDIBLE].

SCOTT RUPPERT: Sort of as we merge the design into the make, right?

SCOTT HAMILTON: Yeah.

AUDIENCE: [INAUDIBLE]

FRANCESCO IORIO: [INAUDIBLE]. No, that's a really good point. So one of the-- the whole purposes of generative design is to make and design closer at the release, right? So that the generative design systems know what it means to produce, to construct something. So that when a design is proposed. it's never just a daydream. It is something that can be built, right?

So that's the whole point. So where you're not only exploiting the capabilities and the characteristics of a particular manufacturing process, or several of them, in fact. But you're also guaranteed that you can actually make it. So you're not just drawing something that then your manufacturing engineer will not be pleased.

So it's a blend. It's actually a blend. So it's much harder, actually, for us-- it will be much harder for us to deal with intangibles like style. But manufacturability is just this left, right, and center. We really want to blend the concept, the process of designing with the process of making in one.

AUDIENCE: [INAUDIBLE]

FRANCESCO IORIO: So OK, so let's put it like this. We're there for some manufacturing processes. So for additive, we're there, as in, like OK, let's say with the widely available manufacturing processes, like laser centering, metal cladding, stereolithography, in general. So then, yeah, I mean, we're there as in-- let's put it like this. We are there not in a fully multiscale level. But we're getting there. So the handling of the microarchitected material at the nanometer scale, all the way up to a mechanical system scale, we're working on it. [CHUCKLES]

But to design a course-- let's say it's going to come with large-scale parts or interconnected mechanical systems, actually, with additive. Yeah, we're very close or there already. So we're also there with machining, so with essentially working centers, so like three 5-axis milling machines and layers. So the Dreamcatcher, like prototype and the product, actually, is we're working, as Andrew was saying. So we have ways to guarantee that a part that we create is machinable, and to make it actually faster or slower to machine, et cetera. But it is going to click and push button for real. So it goes into your Delcam in this case, and it gets machined.

So others, like casting, we're very close. Injection molding, we're getting close. We have expertise, like Moldflow in-house, so for the analysis. So composites, it's a little bit farther away, but we're working on it. As you can see, actually, on the show floor, we have the big boat, part that is actually made of advanced polymers. And they're carbon reinforced. So they deposit on on the fly. So we can generatively make those types of parts.

And then other technologies, like sheet metal, so we're working on that as well. Will it ever see the light of day? I don't know. But we're working on a wide range of options for production things. And so tailoring it so that people can have a choice. So I can see if you need to create 10,000 of something or 1 or a million. There's trade-offs, of course.

DEBRA: Any questions?

AUDIENCE: Look at soft goods or knitting or anything like that?

FRANCESCO IORIO: I'm sorry.

AUDIENCE: Soft goods, knitting, any of those? They do seem like logical applications, because they're things that interact with a living organism, which generative fits pretty well.

FRANCESCO IORIO: No, that's an extremely good question. So soft goods, not let's say in general, not in a very abstract point of view. Textiles, yes. So we are looking actually into textiles. So as in the micro-architectural properties of knitted structures, right? Of fibers, right? So I can't super elaborate, but parts of the world that we're doing on composites is in that direction, so as in the architecture of the knots, the layering, that 3D stacking, that sort of stuff, and the automated printing of them. So, yes, yes, that's one we're looking at, yeah.

AUDIENCE: I was wondering if you could talk about the rate of adoption with generative design and when you expect it to be as ubiquitous as some of the traditional engineering tools like Autodesk and SolidWorks and all the other software out there.

[INTERPOSING VOICES]

BRUCE BLAHO: I think that the tools are getting there. The thing that I see that really usually slows down adoption, especially with, say the bigger enterprises and bigger customers is that they're so invested in the old tools. It's mostly just the inertia. And like I said before, a big enterprise company might have more software engineers developing add-ons than Autodesk had on the product initially. So it's really tough to switch that over. So I think it's going to be-- probably, you'll see the smaller companies. And the more innovative folks will probably adopt it quicker. I think it'll be a little while. It'll be harder for the General Motors and Boeings of the world to move over just from inertia.

AUDIENCE: It's a huge [INAUDIBLE].

BRUCE BLAHO: Yeah, exactly.

AUDIENCE: [INAUDIBLE]

BRUCE BLAHO: That's right. Does to send in the investment these companies have. Yeah, it's just-- and if thousands of people trained in the old system, how do you move them?

DEBRA: Well, like any technology, right? Whenever we switch to a new technology, there's always a lot of funding.

BRUCE BLAHO: Yeah, the switching costs, I think, are particularly high in this industry.

FRANCESCO IORIO: And we're hoping [INAUDIBLE].

BRUCE BLAHO: OK. [INAUDIBLE].

AUDIENCE: So I'm working in the education team at Autodesk. So I work with the next generation of customers. And they are very keen to work also with additive manufacturing. So the educators care about design for additive manufacturing courses. And everyone who knows about Dreamcatcher pings me, hey, I would like to get access to Dreamcatcher. So how do you see Dreamcatcher and the generative design in the education? So because I don't think that 100 students shouldn't push the button for Dreamcatcher to get many Cloud Credits here.

FRANCESCO IORIO: I should've brought the business guy. I'm the wrong guy to ask. But I don't think it should be any different from the policies that we have with other products, as in like if we give away to students the ability to do a rendering in the cloud for free, or at least do some-- we should do that with generative design as well. That's just me, right?

So the reason why we can't have people access Dreamcatcher now is that the research group that builds this prototype is so small that we read like overwhelmed instantaneously, within a fraction of a second, actually, from requests. But the product is coming. So the product, actually, that will be released relatively soon. So we'll answer those questions, as in that it will be available to everyone just like all the other products in our portfolio. So from that moment on, people will be able to use it just like they use Netfabb, just like they use Fusion, or others.

So it's interesting when you say that we should promote adoption by means of training, as in like starting to have people think differently. So it's a complementary experience to the traditional CAD, if you will. And so the product will be just a glimpse into the future of what generative design will bring, because we physically have only so much time to make it and to release it into the market. So we will not do everything so clearly.

But it will be an opportunity for people to understand our perspective. And if we invest into training, they can actually see beyond that, beyond what they can just purchase immediately and get on the mental process of changing their relationship with the design tools, hopefully. I could not agree more that the investment and training is essential, I think.

DEBRA: So we have about three minutes. But I think your questions are really important. So I want to try to make a call. If you have some more questions, we'll take those. Or we'll see the video. But I don't want to miss any of your questions. And the guys will be here afterwards for a little bit anyway. But do we have any more questions? The video is only about a minute. Scott, you want to set the scene for us?

SCOTT HAMILTON: Yeah, well, let's do this. How about you mute the sound? It's just a music background. And then I'll just narrate as we go. But we've been working with Autodesk, with a company called MX3D for several years now. Many of you might be familiar. They've been here at AU for at least the past three years, I think. They were an early-- heard it say an early adopter, early user of Dreamcatcher.