Is generative design the key to sustainability?
A conversation with Arthur Harsuvanakit, Autodesk
Joe Foxton from our team has interviewed Arthur Harsuvanakit from Autodesk to help us answer this question.
Over the past weeks, I’ve been deep in an extended conversation with Arthur Harsuvanakit, a Principal Research Scientist at Autodesk who is charged with researching trends in design, and new design technologies. He has uncovered some promising findings about generative design, and its impact on the future of sustainable design, manufacturing, construction, and on the human race at large.
Tell us a bit about yourself
My name is Arthur Harsuvanakit. I am a senior designer for the Autodesk research team. I live in Berkeley, California with my partner and my cat. I’ve been with the company for nine years, and I have a background in product and industrial design.
For those that don’t know, what is Autodesk?
It’s a design software company. I like to think of us as toolmakers. We develop tools for designers and fabricators with a focus on the built environment. Our three biggest industries are Architecture/Construction, Manufacturing / Product Design, and Media & Entertainment. This includes everything from helping designers and architects to build buildings, all the way to designing the phones and computers that we all use every day, to the movies that we watch, and the kind of visual effects and animations that we all enjoy on those screens.
What is your current focus?
In the last five years, I’ve been focused on a technology concept called generative design.
What is generative design?
Basically, the generative design allows designers to explore multiple design solutions through a computer AI system. So the software generates a multitude of solutions based on the goals and constraints of the design problem, then allows the user to search for the best solution in the generated design space.
The essential ingredients are goals and constraints, ie. the inputs. The generative software can create thousands of solutions, but a lot of them may not be viable. So the role of the designer is to intelligently inform the system about the goals they’re trying to achieve.
What are the goals of your research?
Two generative design goals that I’m very focused-in on right now are sustainability and affordability.
Regarding sustainability, I’ve been researching generative design’s ability to help with lifecycle assessment and assessing the environmental impact of our design choices. This includes everything from whether we put a battery in a computer, to the type of insulation in the walls of our homes.
Generative design tools help us understand the environmental impact of those choices, and the difference between one product versus another, or one battery type versus another. So if we have that data, and allow the generative design system to optimize for the best solution for our goals, ideally we come out with a more sustainable design solution, and therefore a more sustainable product.
Another goal of my generative design work is to understand how it can impact affordability. It’s a nebulous and very complex space. Costs are dependent on manufacturing or construction techniques, as well as the market, which is a perpetually moving target. So right now, it’s a lot of data crunching.
I understand. It sounds like generative design challenges current notions about what design can achieve, and what a design solution should consider. It implies a stronger collaboration between the designer and the technology.
Yeah, one of the benefits of being able to quickly explore a broad solution space is that you get to explore the fringes of the space; the things that you may not have considered, or considered as borderline crazy. It challenges your preconceived notions on how to solve a problem.
The most straightforward example is an everyday structure like a chair. Using our generative design software, I worked on a chair with a designer in France. We provided some constraints, like that it has to have four legs (for manufacturability), those legs must touch the ground, and it needs to have a platform that is off the ground. But we didn’t tell the software how to connect the elevated platform to those four points on the ground. So it really challenged our notion of what a structural frame for a chair would look like. It got us out of the linear approach of legs, backing and feet. In a way, it educated us on the physics of the material that we’re working with.
The generative system uses something called finite element analysis. Basically, it uses the physics of the material and the environment incorporating gravity and considering the maximum weight that you would put on a chair, about 300 pounds. It considers the tilting as you move around on the chair, and it uses those forces to move and remove material, considering the load paths of sitting, moving, and even dropping the chair.
So it educates designers and engineers on how nature would create the most optimal structural paths for a chair because the software follows the same laws of nature. Its gravity, its physical forces, and its moment loads and things like that.
Interesting. That raises a question. It’s clear that in order for the generative design to work well, or any AI to work well, it’s dependent on the quality of the inputs, as we know from the well-worn adage, garbage in, garbage out.
How do you make sure that you’ve sufficiently considered all the right inputs? Is this a new job for the designer, to think more about the inputs than the outputs?
Yeah, I wouldn’t say it’s new. I think a good designer is always thinking about constraints and inputs. But generative design definitely emphasizes that much more.
As a product designer, you’re always running through the consequences of your design decisions for things like manufacturability, or finishes, or the downstream impacts of your decisions. This is where generative design can really help.
If you are able to communicate your manufacturing constraints to the generative system, then it is able to know that those constraints are going to be happening downstream, and it can try to bring them upstream. This means that the solutions are just going to be way better.
In that sense, if you’re able to communicate those constraints, and predict things like, for example, assembly-like knowing that you need a screwdriver to come through this form, at a certain angle to access a certain screw — If you know this upfront, then you can communicate to the generative system not to create geometry around this area because I need access. Then it just becomes that much more viable as a solution.
If you don’t think about the assembly, or that the screwdriver will be coming through a certain gap, then the system wouldn’t know either. So then you’d have to backtrack. So generative design forces the designer to think through the steps downstream and bring them upstream.
Let’s consider the challenges of generative design. In a previous chat, you listed the challenges as such:
1. Generative design always needs a curator. A designer will always need to have the last touch.
2. Design always changes because culture always changes. You are always a step behind what we would consider totally new.
3. Some things are hard to quantify. You can never know the full universe of input parameters. Manufacturability is particularly hard to quantify.
Do you have anything to add to these challenges?
I think as we tackle all these challenges, they will, of course, get easier. The thing with AI is that the more data and the more solutions you generate, the better the next round of solutions will be. So the more people use the system, the more the system understands its goal, and the better it gets at understanding what is considered a viable solution.
But also, the programmers developing that tool will further understand what they need to build into that system, as far as manufacturability, as far as supporting aesthetics, as far as supporting that last stretch of taking the solution into production, and into, say, a marketing campaign.
It’s important to consider the question: How do we make it a highly usable design tool, not just a kind of engine that just churns out what I personally might consider viable, but others might consider uninteresting.
We are still working on how to bridge that performance gap between what generative design can currently offer, and what is necessary to allow its solutions to go straight into production. That’s a big problem that we’re working on. I think the gap is rapidly shrinking.
I would love to witness the results of your experiment with the chair and see what the design AI thinks a chair is, given your inputs. Did it end up doing a good job?
It did a pretty good job in terms of challenging our notions of both our constraints and the actual form of the chair. So that project finished a year and a half ago. I can show you the animation of the chair morphing into a viable form over multiple iterations.
I think design will always be an iterative process. And so will generative design. That’s kind of inherent in the process. This is exactly the way it went with the chair; it looked at the boundary conditions or the bounding box of a chair, and it started to remove material where it thought it wasn’t needed. Then it evaluated, and then it removed more, and then re-evaluated, then it moved something, and then it re-evaluated again.
So there’s an inherent feedback loop, an iteration loop in the system, but there’s also a kind of macro-iteration loop with the designer. So the designer will put four points on the floor, asking the AI to connect to these four points. But that’s just an initial setup step. Maybe for the manufacturing technique, you need to have a little bit more information than just four points. Maybe you need a different kind of side-loading conditions or something. So the results actually inform the design what additional information the design needs. Based on the results, it educates the designer on a previously unconsidered limitation, but also a possibility.
So the technology provides an opportunity to iterate on setup in a way that was previously only possible downstream, which is much more costly.
One of the most vexing issues when automating creative processes like design is how to capture the concept of aesthetics. Being a coder-by trade, I know that computers need quantification and quantization.
How can you quantify aesthetics?
Yeah, it’s a challenge, I think we have yet to understand how to fully quantify aesthetics, but there are certain principles in aesthetics that we can communicate, like, proportions. There are certain proportions that just make sense, in terms of the actual product, in terms of the actual design space and the design constraints. So we can encode those things, but also things like logo placement, or where the headlights are placed on a car. Those kinds of constraints and relationships are well defined for brands, and for certain industries, so they can be easily encoded.
There are subtle changes in the shape and placement of elements that will communicate different things. So it can get very hard to understand what is a good reductive process, or change process. But I think, again, the more people use the tool, the more they will understand the range of what is acceptable, and what is not acceptable.
I think part of the challenge right now, for us, in terms of aesthetics, is actually just getting the pipeline working by getting the generative output to be somewhat malleable once it’s out of the system so that it can morph to certain aesthetic constraints.
So say you wanted a modernist aesthetic, you want the form to adjust to the principles of modernism. So you want a very rigid look with hard corners and no-frills. The form can actually just change according to that aesthetic profile. Getting the format down and getting it right, so that it’s very malleable to those kinds of filters is not a trivial task for software.
So ultimately, you still always need that last mile to be done by a designer who can inform on taste, fashion, and culture of the time, and what the client user is essentially looking for from an aesthetic perspective.
Absolutely. I think that generative design systems will always be a tool. It will just be able to encompass more and more design tasks. But in terms of communicating the value of its solutions, that is still up to the designer. Like you said, interpreting the state of the culture in terms of what is the right look to hit, what the design is trying to communicate to the culture, to the buyer, or to the industry… That still is very abstract and needs to be communicated through the designer, not through a data-driven tool. Despite what the hype on the internet says about AI taking creative jobs.
Do you think it will ever become possible for generative design to replace the designer? Or will it just make their job cheaper, faster and… different?
I would say the latter, I would say it will make the designer’s job different and hopefully more interesting because we only know what we allow ourselves to explore, and generative design gives the designer a lot more options to explore.
Hopefully, this tool will challenge us to explore further. For a typical design process, you might go through two to five iterations, driven by cost and fatigue constraints. The generative design allows you to go through a hundred iterations in a single day. So you’ll be able to reach further, in terms of different goals, in terms of the variety of what you’re looking for. In some sense, it may be that you save time, but it may be that you actually spend more time in certain areas, like exploring the finer decision points between one design versus another. You’ll be able to spend more time really weighing the pros and cons of different options, or comparing the costs, or manufacturability of all these aesthetics.
So maybe you will actually spend a lot of time on different things than you would today, like evaluating the hundreds or thousands of different solutions that a generative system can offer. We’re working on tools to filter that time down so that it’s just way easier to traverse a 1000+ solution space. Ultimately it will make design way more engaging and way more creatively fulfilling.
Replacing the designer is not the goal. I think we, as toolmakers, are in service to designers. The design process is a very abstract process. There are a lot of technical aspects to it, and a lot of quantifiable aspects to it, but there’s so much grey area in the design practice. There is human problem solving that is hard to even verbalize. As a designer, you are constantly processing all these things, and you’ve got these unquantifiable constraints in your mind that make us human.
A lot of times, you get struck by a solution, but you have no idea how you arrived at it. If you can’t know how you arrived at it, then it’s even harder to quantify it.
Indeed. At the end of the day, the best designers have this thing called taste that is, again, impossible to adequately describe and quantify, and speaks to innate human abilities and desires. It is simply a human quality that is essential for successful design.
That triggers another thought. Being a coder by trade, I’ve noticed a big trend over the last twenty years away from software development being the domain of the nerdiest amongst us, to essentially being an assembly of pre-built libraries, making it possible for virtually anyone to learn. This has unleashed a tsunami of software that now runs every part of our lives.
I’m wondering if design is on this same trend, where it becomes more democratized by technologies like generative design.
Thinking forward 10, 20, 30 years, will generative design unleash a massive wave of creation into the world? Do you think design will become more accessible, as it becomes more possible for anyone to design?
Yeah, I think it has to. The way we build has a direct connection to impact on the environment. The skills to build in a sustainable way need to become very modular, and very easy to access. We can’t be writing simulation packages, and one-offs constantly, for every kind of sub-industry, right? We have to share those analysis, those engines in a way where we develop a collective platform that tackles much more complex problems, like sustainability and environmental impact.
So, yeah, I think the problems of the next five to ten years require modularity. We need to be able to bring preconceived pieces together in a way that can be iterated on, and taken apart, pulling from different places and different disciplines, in order to develop robust, sustainable solutions
Let’s dive deeper into sustainability. How does generative design apply to sustainability? Can generative design give designers new insights into sustainability? Can those insights be trusted?
Yeah, that’s something we’re actively working on right now. Sustainability concerns are already being built into the tool.
One can equate sustainability to the reduction of material. A solution that uses less material is likely to be more sustainable. Using less material also produces less waste. Using less material is just better for the environment in general.
But that’s not enough. I think for the current problems of design right now, it’s not just about reducing materials, it’s about choosing the right ones. It’s about designing for the recycling industry, or considering that the product might end up in a landfill. You must consider the entire lifecycle.
So, right now, where we’re pulling together the data that exists in the industry in terms of sustainability, the actual material data, so, the impact of extruded steel versus aluminium, or one technique versus another. We are trying to extract that kind of raw data into insights, crunching the vast inventory of materials and the impact attributes that they have into the real-time comparative analysis.
Let’s say I’m designing my keyboard. I know the manufacturers will have to use plastic. I choose one type of plastic, and the system recognizes that plastic and crunches numbers on the fly. If I use this alternative plastic that fits my design criteria and constraints, I would be saving this much on Co2 or this much on acidification, or this much on other impact measures.
In our last chat, you talked about sustainability in the context of building affordable housing.
Yeah, it’s the same type of problem. In a way, buildings are just much larger assemblies than products like a computer or a keyboard. Similar types of problems apply. There are definitely differences between the construction industry, and the product industry, in terms of manufacturing. For example, most houses are built on-site, not in a factory the way products are. So that inherently makes it much more difficult to control the process and to ensure that it’s a lean process that results in minimal waste.
We’re working with all kinds of companies to figure out what is controllable, and what is quantifiable. We’re seeing some affordable housing companies actually building modular homes in factories so that they can be assembled, or stacked to build multifamily housing units. Seeing as they are being built in factories, you can start to really quantify that process to optimize for things like sustainability, impact, cost, and productivity.
Industries are starting to converge and starting to borrow from one another. Optimization tools like generative design are really well placed to carry those insights between different industries, and also just to optimize the process within those industries.
It’s interesting that the most sustainable building materials are often the most aesthetically pleasing. Is there a connection between aesthetics and sustainability?
I think so. We as human beings just respond to natural materials more than synthetic ones, in a kind of visceral way, and in a subconscious health-related way. Natural materials are just less toxic, and you can kind of feel that. Your body will sense it in the air; your body will sense the gassing-off of certain plastics. Synthetic materials often require flame retardants, whereas natural materials may not need that.
An example is using sheep’s wool as insulation. It has antimicrobial properties, and it cleans the air that filters through it, pulling air through the wall. So you actually clean the air inside. In a lot of urban areas, the air is more toxic inside than it is outside.
So you have these health benefits that subconsciously make you feel better about the material. I think that visually, you can understand that certain sustainable materials have a kind of wellness benefit to it, whether it has truly helped or not.
Sometimes sustainability can elevate a synthetic material. Some recycled plastics have speckles or imperfections, and sometimes a designer will celebrate that.
Then there is something to the narrative that a material embodies. A material can visually communicate a story. It may not just be about aesthetics, it might be the story behind why the material was used, the region it was used in, and what the material means in that context.
We’re down to our final question, which is a two-part question.
What is the most exciting thing that your research has revealed? And what is the most disappointing, challenging or difficult thing that it has revealed?
I think the most exciting thing is seeing how industry players, ie. very established designers and architects respond when they first encounter generative design. When they see how it can augment their design process, they’re totally blown away.
In some sense, it’s because they’ve been doing this process for so long, ie. refining their process, the way they work, and the tools that they use to achieve world-class designs. So to see the opportunity for them to change and see that they recognize the potential, wanting to further evolve their process, I think that’s really satisfying.
Designers are always looking for the next thing. Whether it’s the next visual language or the next process language, they always want to challenge themselves. Witnessing the designer’s drive to evolve is exhilarating.
When they see the potential, they become our partners and help us learn. We get to better understand their challenges, and what’s really going on in an industry. That exchange between the builders of tools, and what’s happening on the ground in the industry, helps us converge on bigger goals like sustainability. Having a hand in that is really satisfying, exciting, and surprising.
The biggest challenge from my seat is not getting too caught up in the tool. At the end of the day, it’s the user of the tool that really allows the tool to do what it can do, right? A lot of designers try to break the tool because they want to use it in a particular way. And that’s actually good. It’s not like you’re not using it correctly. We are not dictating the process. We are learning about the constantly changing industry and the constantly changing challenges within the industry. And sometimes tools are not the solution at all. Sometimes they are not the only solution. And I think we have to acknowledge that.
I think there are certain things we can impact with new tools, and there are certain things that are just human nature. Understanding the right time and the right place to inject a new technology is a delicate balancing act.
I think a theme has emerged here. Technology will empower and magnify human ingenuity. It will never replace human ingenuity.
Yeah, I’m definitely in that camp.
Thank you for reading!
This article is based on an interview with Arthur Harsuvanakit, a Principal Research Scientist at Autodesk.
Interviewer: Joe Foxton
Visual curation, layout & final touches: Justyna Cyrankiewicz
- For more inspiration and stories, follow us on our Instagram, Facebook, and here 🏝
- To see our projects, go straight to our website: Future Works🚀
With love, Future Works
