Delve

Episode 12: Generative Design

Could machine learning help us find better designs for cities — even ones that we didn’t know were possible?

By Eric Jaffe

episode_12_equitable_3x2.jpeg

Once the largest office building in the world, the Equitable Building casts considerable shadows over the surrounding area.
(Image: Wikimedia Commons)

Subscribe wherever you listen to podcasts

“City of the Future” is a podcast that explores ideas and innovations that could transform cities. In this episode, hosts Eric Jaffe and Vanessa Quirk discuss how generative design can help cities make better decisions for their communities with writer and historian Molly Wright Steenson, geographer and city planner Evan Lowry, and Sidewalk Labs’ Senior Product Manager Violet Whitney and Senior Design Lead Brian Ho.

...

Eric Jaffe: In the late 19th and early 20th century, the elevator was changing cities around the world.

Vanessa Quirk: Thanks to this innovation, skyscrapers became practical for the first time. And soon, the race to see who could build the biggest building was underway.

Eric Jaffe: In 1912, The Equitable Life Assurance Society of the United States asked their architects to design a New York City headquarters that would be the envy of all the other towers in Manhattan.

Vanessa Quirk: There was just one problem — elevator technology was good, but it wasn’t quite as good as the building owners’ imaginations. So, they told their architects: if you can’t go up, go out.

Eric Jaffe: At over a million square feet, the Equitable Building did become the largest office building in the world when it opened in 1915, but it did not open to great acclaim.

Vanessa Quirk: Here’s what Real Estate Magazine said at the time.

Radio Voice: It’s a monstrous parasite on the veins and arteries of New York City.

Vanessa Quirk: And a New York City report the following year.

Radio Voice: A notable illustration of the evil effect of a building that covers too much land.

Eric Jaffe: To understand what was so evil about it, we decided to go and see this monstrosity for ourselves.

...

Vanessa Quirk: We are in the shadows of the Equitable Building right now.

Eric Jaffe: We’re across the street on Broadway, and it’s covered in shadows. It’s just one big shadow, which actually looks kind of nice because it’s pretty hot out today.

Vanessa Quirk: You can imagine in the early 1900s, when electricity wasn’t such a big thing, that it would have been problematic to have your office completely blocked out with the Equitable Building shadows.

Eric Jaffe: And the shadow’s casting over this little public square that is right beside the Equitable Building because the Equitable Building is so tall.

Vanessa Quirk: It’s not just tall. It’s girthy. It’s very wide, and I really think it takes up this entire city block.

Eric Jaffe: It goes all the way to the other block. And it’s consistently wide and massive, whereas a lot of the other buildings around here are narrow.

Vanessa Quirk: The hullabaloo that was kicked up around this building’s shadow is actually the reason why New York City had its first zoning laws passed.

Eric Jaffe: Is that because others were upset that they got blocked from the sun?

Vanessa Quirk: Exactly. Because of those early zoning laws, now skyscrapers have to do something called a setback. It’s actually easy to see in this building right here next to the Equitable Building, which I’m going to guess was constructed afterwards. See how it tiers up? It’s like a layer cake, where each layer of the cake gets a little bit smaller as it goes up.

Eric Jaffe: Right, I can see a little more sunlight casting over there, so maybe that was the point.

Vanessa Quirk: As the history of the Equitable Building shows, every urban development project has trade-offs, constraints, and unintended consequences.

Eric Jaffe: The goal for architects and developers should be to anticipate as many of the effects a given choice will have as possible.

Vanessa Quirk: But it’s not easy. Even today, designing a building is a very time-consuming, expensive, and fragmented process. If one person makes a change, it takes a great deal of money and effort to know how the repercussions will ripple across the project.

Eric Jaffe: Over time, these costs lead not just to more expensive development projects, but to less affordable cities, too.

Vanessa Quirk: Fortunately, there’s a whole new field that uses machine learning to help architects, real estate developers, and urban planners better understand how their designs might impact communities.

Eric Jaffe: And choose designs that work better for everybody.

Vanessa Quirk: It could even give communities greater voice in shaping their neighborhoods.

Eric Jaffe: Welcome to “City of the Future,” a podcast from Sidewalk Labs.

Vanessa Quirk: Each episode, we explore ideas and innovations that could transform our cities.

Eric Jaffe: We’re your hosts. I’m Eric Jaffe.

Vanessa Quirk: And I’m Vanessa Quirk.

Eric Jaffe: In this episode, we’re talking about an innovation that could help us make better decisions about how to build communities that work for everybody.

Vanessa Quirk: Generative design.

...

Violet Whitney: I was first introduced to generative design back in architecture school.

Vanessa Quirk: That’s our colleague Violet Whitney, an architect and programmer. She’s been thinking about the ways technology can improve urban planning for quite some time now.

Eric Jaffe: For the last few years, she’s been really excited by this emerging field of study called generative design.

Vanessa Quirk: Can you explain what generative design is exactly?

Violet Whitney: Sure. It’s the process of automatically producing many designs — thousands of options based on goals and constraints you feed into a computer.

Vanessa Quirk: Okay, so you can apply generative design to anything. You could use it to automatically produce thousands of designs for a chair, for example, and then see which chair is optimal based on your goals for the chair. You can put into the computer, “World’s best chair, please.”

Violet Whitney: Right, best according to whatever you told the computer was best for you.

episode_12_gendes.jpeg

In Amsterdam, Dutch startup MX3D partnered with Autodesk to create a 3D-printed bridge using generative design. The software was taught to find the most efficient design under specific parameters. (Image: MX3D)

Violet Whitney: At Sidewalk Labs we’ve been wondering: how could you apply generative design to something as complex as the urban planning process?

Eric Jaffe: Violet is part of a team of designers and engineers who have built a product that generates thousands of design options for any given development project. The idea is that this product could help urban development teams find the best designs for a building, a neighborhood, or even a whole district. They call it Delve.

Violet Whitney: Delve uses machine learning to produce radically better urban designs in a way that just wasn’t possible before.

Vanessa Quirk: Over the summer, we asked Violet and her colleague Brian Ho to give us a sneak peek of a beta version of the product before it launched this fall.

Eric Jaffe: Alright, can you describe what we’re looking at here?

Brian Ho: This is what we call Anywheresville.

Violet Whitney: You have a 3D building environment, but we’re looking at it from an above-map view where we see the entire district.

Vanessa Quirk: I see streets, parks, and buildings.

Eric Jaffe: How did Delve produce the neighborhood design we’re looking at here?

Violet Whitney: One of the first things that we always do is to onboard all of the constraints and local rules and spatial guidelines that might exist.

Brian Ho: One of those is often zoning. We can encode maximum heights, and we can do things like setbacks — anything that you can get from municipal open datasets. If there is data that is relevant to a project and it exists in a GIS form, we can go and bring it into the project.

Vanessa Quirk: Okay, so now that we know that our plans won’t violate zoning codes, can we get to the fun part of generating the best designs? How does it work?

Violet Whitney: We’ll start with the stakeholder and whoever’s developing a site, and they’ll have goals and objectives: this much residential space, this much open space, everything within a five-minute walk distance. We’ll put those goals in, and that’s when Delve starts generating the best options for them.

Brian Ho: Some things have trade-offs. You can’t optimize everything all at once. What the tool is essentially doing is negotiating that complexity.

Eric Jaffe: To that point, there is this little scorecard that’s popped up here in the bottom of the screen — you can see the population, you can see how many jobs, you can see how much commercial space there is. I imagine that changes based on the configurations that you’re putting on top of these parcels.

Brian Ho: Yeah, I’ve selected an option here. This is Iteration 163.

Violet Whitney: You can see it has a daylight score of 57%. Whereas this one, Iteration 204, optimizes for density, and it has a daylight score of 32%.

Vanessa Quirk: Oh wow, you can see right away it looks so much denser.

Eric Jaffe: It’s possible that audio is the worst medium for this conversation because when you look at it, it’s just so obvious and clear and striking — the visual differences, the types of scores you’re going to get, and the types of community you’re going to live in.

Vanessa Quirk: Audio challenge accepted, Eric.

...

Sound of birds chirping.

Vanessa Quirk: So in this iteration — let’s call it Sunshine Place — none of the buildings are too tall. They’re kind of far apart from each other, separated by a lot of green space. I imagine kids running around, and maybe there are benches so you can sit down and soak in the sunshine. The birds are really happy here. And because the buildings aren’t too tall, Sunshine Place gets a very high score for daylight access. It doesn’t cast shadows onto other buildings or public spaces.

Eric Jaffe: I like the bird sound effects.

Violet Whitney: We should actually put these sounds in the Delve products, so that if you click an iteration with open space, you hear birds.

Eric Jaffe: I want to optimize for bird chirps.

Vanessa Quirk: Oh my God, I love that.

Eric Jaffe: Alright, I hate to burst your bubble here, but this density score — the amount of residents in this place — is terrible. No developer would be able to make their project economics work with that small of a population.

Vanessa Quirk: Okay, this is a valid point. Also, I want it to be affordable for a lot of different types of households. Let’s look at another option. The other one: let’s call this one Dense City. Get it? Den-sity?

Eric Jaffe: I award you no points, and may God have mercy on your soul.

Vanessa Quirk: Oh, Eric. Well, then you describe it.

Sound of cars and traffic.

Eric Jaffe: Alright, there’s a lot happening here. There are tons of super tall buildings, and they’re crammed close together. It’s almost like Midtown Manhattan. There’s only one park here, and it’s a central park. It’s located in the middle of all these tall buildings. But I’m guessing this park has a shadow problem being surrounded by all of these towers.

Vanessa Quirk: Right, I’m imagining trying to set up a picnic blanket and having to move every few minutes because the shadows keep shifting across the park with you.

Brian Ho: Yeah, it looks like the daylight access here is 32% on average. We’ve dropped 25 percentage points through the distribution and massing alone.

episode_12_shadows.jpeg

Buildings cast shadows over streets, reducing daylight access in the city. (Image: Alexander Rabb / Flickr)

Eric Jaffe: But a lot more people can live here in Dense City compared to Sunshine Place — and they don’t have to walk far to get to stores, or get to transit for their jobs, or get to the park.

Vanessa Quirk: To your earlier point, for the developer, you’re probably more likely to get a return on your investment in this option.

Eric Jaffe: And that’s a very big part of the trade-off.

Vanessa Quirk: But clearly, if we just needed to generate two very different scenarios like these and choose between them, you wouldn’t need a product like Delve. Any architect could create two scenarios. I think what’s different here, Violet, is that Delve uses machine learning to generate thousands of these kinds of scenarios in minutes, right?

Violet Whitney: Yeah. Initially when we started on this project, we were thinking a lot about trade-offs, but now we’re thinking a lot more about how Delve can produce options for a site that we just didn’t even know were possible before. There are a thousand options out there that might be much better to live in. Delve doesn’t just generate all these new options — it actually helps you find the best design for all the different stakeholders.

Eric Jaffe: Let’s pause on that for a second, because we know from history and still today, development is pretty complicated and often fraught. It’s hard to create a building, a neighborhood, or a district that takes into account the needs and priorities of all the different stakeholders involved, whether that’s architects, developers, or community members.

Vanessa Quirk: Right, and then the question becomes: can we use technology to do that — to help us make better, more human designs?

Eric Jaffe: That’s a big question, and it’s one that architects have been asking for quite some time.

...

Molly Wright Steenson: Well, I think there’s an interesting question when you’re looking at generative design about where creativity comes from. Does creativity come from a human, or does creativity come from how humans work with technology in different kinds of ways? Could a computer suggest options to you that never might have come to mind or make you think very differently about a problem?

Eric Jaffe: That’s Molly Wright Steenson, a writer and historian at Carnegie Mellon whose work centers around the intersection of architecture and computing.

Vanessa Quirk: What’s exciting about Molly’s work is that technology today is finally catching up to the visions of the architects of the ’60s and ’70s who were, back then, just dreaming big.

Molly Wright Steenson: The big ideas were there years ago, but the possibilities for actually putting those big ideas into place were not.

Vanessa Quirk: One architect with big ideas was Cedric Price, who we’ve talked about on the show before.

Eric Jaffe: In the ’70s, Price worked with software developers on a project called Generator, which he imagined as a building made up of cubes that, thanks to different computer programs, could be moved around by the people visiting it.

episode_12_generator.jpeg

A working electronic model of Cedric Price’s Generator project. (Image: Cedric Price, View of working electronic model for the Generator project, between 1976 and 1979. Colour electrophotographic print, 16.3 x 23.9 cm. DR1995:0280:651:004:006. Cedric Price fonds, Canadian Centre for Architecture © CCA)

Molly Wright Steenson: The most important program they created was a boredom program. Basically, in the event of Generator’s parts not being moved around, it would start coming up with its own layouts for itself.

Vanessa Quirk: I love that he was so ahead of his time.

Molly Wright Steenson: He’s amazing.

Vanessa Quirk: Because when we talk about generative design today, we talk about simulations, not reconfigurations of a building in real time. The fact that Price was trying to do that is just so neat.

Molly Wright Steenson: Unfortunately, like most of his work, it wasn’t built — the project got cancelled. But what I like so much about Cedric Price’s work is that he wanted to make you think differently about how you use it, what it meant to be designing it, and what it meant to be the architect even coming up with these ideas.

Eric Jaffe: Price wasn’t the only architect of this era who wondered if computers could show architects another way.

Molly Wright Steenson: Christopher Alexander was one of the very first architects to use a computer.

Vanessa Quirk: Alexander, who you may know as the author of A Pattern Language, preached that the best spaces are built by the people who live in them, not professional architects or designers.

Molly Wright Steenson: He felt that designers had biases that made them do design-y things, and that the self-importance of a designer or architect might make them make design decisions just for the flourish and not for what would actually make for good design, or fair design, or just design.

Vanessa Quirk: Here’s Christopher Alexander in 1996, addressing a crowd of computer scientists.

Christopher Alexander: In our time, the production of the environment has largely gone out of the hands of people at large in society, and it’s a serious problem. That’s a debatable matter — some people would say, “What are you talking about? It’s all absolutely fine.” I suppose the architect of this particular room would say that. But actually, it isn’t fine. It’s a hell of a problem. Already, most of the habitable environment is something that has been built in these last 50 years, so the fact that it is not nurturing is a very drastic matter for all of us.

Eric Jaffe: Alexander had the radical idea that we should just ask ordinary people how they feel about a design.

Vanessa Quirk: Questions like: Are you happy here? Do you feel good? Bad? Wonderful? Comfortable? Then, he’d use that feedback as the basis of the design process.

Eric Jaffe: Architects and developers weren’t the biggest fans of Alexander’s idea of sharing space at the drafting table.

Vanessa Quirk: But computer scientists? They loved Alexander. They used his work as inspiration for user-generated spaces on the internet, like Wikipedia. In 1996, a few famous computer scientists invited him to keynote that conference from earlier. Here’s another clip:

Christopher Alexander: Compared with the pattern language of mine that you’ve seen in the book, my colleagues and I have written things that are much more like what you call code. They’re generative processes which are more like sets of instructions that produce things. Because of the complexity of the situation and because of the way software is going, software that was designed to do that could very rapidly take the world by storm.

Eric Jaffe: That sounds a lot like what Delve is trying to use software to do.

Vanessa Quirk: You can see why Alexander felt that computers could one day completely change the way buildings and neighborhoods are designed, using inputs like what people actually want out of their neighborhoods.

Eric Jaffe: Yes, but one thing missing from Alexander is the visualization element. It’s easy to ask someone how they feel about an existing project; it’s really hard to ask people how they feel about a project that doesn’t exist yet.

Vanessa Quirk: According to Molly, Alexander did start thinking about this question of visualization in the ’60s, but he gave it up to think about other areas of research. But in the ’70s, a different group of computer-oriented architects picked up where he left off.

Eric Jaffe: They were all at MIT, and they called themselves the Architecture Machine Group.

Molly Wright Steenson: They went to Aspen, Colorado with a Jeep, put a lot of movie cameras on it, and drove down the streets and gathered footage.

Vanessa Quirk: It’s almost like Aspen got their own Google Maps three decades before the rest of us.

Eric Jaffe: The Architecture Machine Group called it the Aspen Movie Map.

Molly Wright Steenson: You could sit in this Eames Lounge Chair that had joysticks in it, and you’d zoom down the streets of Aspen, Colorado. On a screen to your left, you’d have a satellite map; on a screen to the right, you’d have a touch screen map. It was an idea to get stories and people to come out of the different buildings, and you could interact with them along the way.

Vanessa Quirk: So it’s not a design tool per se. It’s more like a way that anyone can virtually experience a design.

Eric Jaffe: Right, but imagine if you could allow people to have that visual experience of how a place would look and feel before shovels go into the ground. What could that mean for the planning process?

Evan Lowry: As planners, we’re thinking and looking at 2D site plans all the time, so we have a picture in our head of what a building in a particular location would look like. The residents don’t think about it too much.

Vanessa Quirk: That’s Evan Lowry. He’s a geographer and planner for the city of Charlotte, North Carolina.

Eric Jaffe: One of the things Evan is responsible for is keeping track of his city’s growth — and it’s been growing fast.

Evan Lowry: It’s been around 15,000 people a year for the last 10 years.

Vanessa Quirk: To help model out what Charlotte could look like in 20 years time, the city is using visualization software.

episode_12_charlotte.jpeg

An aerial view of Charlotte, North Carolina (Image: Clay Banks / Unsplash)

Eric Jaffe: Could you explain why the ability to visually experience these models is so important? Is it a game changer for you?

Evan Lowry: I think it’s more of a game changer for the residents. If they can pull up a website and turn a layer off and on, and see how it relates to their own house and their own neighborhood, that will definitely be a game changer in community engagement.

Vanessa Quirk: Charlotte hasn’t rolled out their visualization tool to the community just yet, but when they do, it will allow community members to give their feedback online.

Evan Lowry: Before COVID, the residents would be expected to come to a meeting at the government center or somewhere in the community. Now, the push is towards more virtual engagement, which I think has the opportunity to reach more people.

Eric Jaffe: Evan thinks that this kind of participation wouldn’t just be more accessible. It would be more meaningful too.

Evan Lowry: You would be able to make your own assumptions and comments based on what it would look like if you were standing there and could see it. In the future, I’d like to see it become a more interactive process in design. The residents will be able to sketch out their own vision of what the development will look like. I think that would be very powerful.

...

Vanessa Quirk: I can understand now why generative design has the potential to really transform development, because it’s exactly what Christopher Alexander was dreaming about: a technology that lets you incorporate the needs and feelings of lots of different kinds of people, not just architects.

Eric Jaffe: Like the Aspen Movie Map, we can potentially experience a project virtually, so we know what it would feel like to live in a place.

Vanessa Quirk: Not only that, but like Price wanted, generative design allows for all these new possibilities, too.

Eric Jaffe: Let’s get to practicalities here. Of course community members should have the best tools to provide more meaningful feedback into a development project, like they’re trying to do in Charlotte. But how can generative design help with the very messy task that falls to cities and development teams of weighing all these competing priorities?

Vanessa Quirk: At the end of the day, how do you make a decision on which plan to move forward with?

Brian Ho: You’re asking the million variant question: how do we go from many to one?

Eric Jaffe: It’s a question Brian and Violet hope Delve could help cities and development teams answer.

Brian Ho: The idea is not that you click a button, get a design, print it, and walk away. It’s really the opposite. We want to augment and give superpowers to designers, planners, and developers who are out there working with communities and trying to build better places.

Violet Whitney: We’re thinking about future iterations of Delve, where community comments can actually become constraints. Designs would be held accountable to the community’s needs, and that would ensure that when we put out designs, a community can say this actually meets a check mark — it has this much daylight, or it’s this walkable.

Vanessa Quirk: Or, it has this many birds chirping!

Violet Whitney: Exactly. If we can achieve that, then we can measure how successful a project is according to how well it meets people’s needs or improves their quality of life.

Vanessa Quirk: And if you do that before a shovel even hits the ground, the community can really understand what this project will mean for them and how it will work for them before it becomes a reality.

Violet Whitney: That will really change the way we plan not just buildings and neighborhoods, but cities, too.

Eric Jaffe: Brian and Violet are hoping that Delve makes it possible to bring more voices into the development process.

Vanessa Quirk: If it does, it will show that generative design can help us make better and more human designs, buildings, and neighborhoods.

Eric Jaffe: Bringing our cities out of the shadows, and just maybe into a brighter future.

...

Eric Jaffe: Thank you for listening to City of the Future, a podcast from Sidewalk Labs. Your hosts are Vanessa Quirk and me, Eric Jaffe. We are produced by Benjamin Walker and Andrew Callaway.

Vanessa Quirk: Mix by Zach McNees. Special thanks to Violet Whitney, Brian Ho, Molly Wright Steenson, Evan Lowry, and Andrew Callaway for his special appearance as the mid-Atlantic old-timey guy.

EJ: Our art is by the great Tim Kau. Our music is composed by Adaam James Levin-Areddy. If you want to hear more of Adaam’s work, you can check out his band, Lost Amsterdam.

Vanessa Quirk: To learn more about Sidewalk Labs, visit our website, where you can subscribe to our newsletter at the bottom of the page. You can also follow us on Instagram.

Eric Jaffe: See you in the future!

Vanessa Quirk: Bye!