Podcast: 15-Minute Cities Via Urban Design AI

author avatar
Podcast: 15-Minute Cities Via Urban Design AI

In this episode, we uncover the fascinating synergy between artificial intelligence (AI) and the urban design landscape to enable the mystical and sought after 15-minute city!

In this episode, we uncover the fascinating synergy between artificial intelligence (AI) and the urban design landscape to enable the mystical and sought after 15-minute city!

This podcast is sponsored by Mouser Electronics


(2:50) - From Pixels to Pavement: AI's Impact on Urban Design

This episode was brought to you by Mouser, our favorite place to get electronics parts for any project, whether it be a hobby at home or a prototype for work. Click HERE to learn more about how future cities will leverage the internet of things and AI for better residential experiences!


What's going on, folks? Welcome back to the Next Byte podcast. And let me ask you something. Do you ever dream of a 15-minute city? Do you hate the fact that most of your city is taken up by roads for cars that you really hate even more? Well, AI is here to save the day, because in today's episode, we're going to talk about how AI from Tsinghua University is going to resolve all of our city that is planned around cars problems. So, let's buckle up and get into it, or get out of the car.

I'm Daniel, and I'm Farbod. And this is the NextByte Podcast. Every week, we explore interesting and impactful tech and engineering content from Wevolver.com and deliver it to you in bite sized episodes that are easy to understand, regardless of your background. 

Farbod: All right, we're gonna be talking about urbanization, but before we get started with today's episode, let's talk about today's sponsor, Mouser Electronics. Now you might be wondering, how does Mouser, the world's biggest electronic distributor, one of the world's biggest electronic distributors, what do they know about urbanization? How do they even factor into this conversation? Well folks, we've told you before, but we're gonna tell you again, right? Mouser, by being connected to all these folks in academia, industry, they have a lot of insight about how technology is being used in different aspects of our lives. So, they have this technical resource that is focusing on cities are becoming bigger, urbanization is happening, and they talk about the role that technology is gonna play here, whether it's impacting sustainability, transportation, AI. And if it's something you're interested in, you should definitely check this article out and just get that foundational knowledge set before you listen to today's episode, which hint, hint is related.

Daniel: Well, I'll just say, I've got the article open, so I'm taking a look at it now, but one of the things that I think is really cool is it's kind of a primer on all the different ways that artificial intelligence could be included in a smart city. So, you mentioned some of it, intelligent transportation systems, they talk about using AI as a part of information systems to make sure we've got smart logistics, et cetera. This is awesome because one of the ways that you can use AI and improving the whole urban environment is in urban planning. And that's not mentioned in the Mouser article. So, I kind of think that this is kind of a big extension of the discourse here, right? We can talk about all the ways that AI is already being used in cities to improve them. And Mouser's got that covered in the article that we're linking in the show notes. And then the extension of that is the future of using AI, not just to improve our transportation systems, not just to improve our renewable energy systems, not just to improve our logistics, but also using AI to plan the city itself from the ground up. And that's kind of the awesome segue into the article for today.

Farbod: So yeah, let's talk about today's article. This research, it was published in Nature. It's coming from Tsinghua University, which I don't think we've ever talked about before. But fun fact, it's known as China's MIT.

Daniel: MIT is known as America's Tsinghua.

Farbod: Even better. Yeah. So, you guys know what a fan we are of MIT, of America. So, we've got to start covering more stuff from Tsinghua, because this research really piqued my interest. It's not usually a topic that we talk about, like urban planning. It's not a topic I even really care about for the most part. Outside of this really hot topic, which is the 15-minute city. So, you and I, we live in suburbs. Suburbs are very popular in this side of America. But the major downside is that if you want to go anywhere, and I mean anywhere, even remotely close to your house, you're probably driving there. So, the obvious answer is, well, just go live in a city. For example, for us, that would be Washington, D.C. Well, the cities aren't super friendly either. Most of the city is just made up of roads because they only, whenever they were planning the city, they cared about cars greatly. Because there's so many cars traveling around and it's really dense, you have all this carbon dioxide. And fun fact, the carbon dioxide that comes from your vehicles account for about 70% of the carbon emissions inside of a city. And some of the things that this research talks about, a little bit of the background they're giving here, they're saying when you make everything so road and car dependent, you actually increase the inequality of access to these essential services like schools, like hospitals, et cetera, et cetera. And the goal of the 15-minute city is, is what if we could make everything accessible to you within 15-minutes, wherever you are, whether it's 15 minutes of walking or biking, right? And what's interesting to me is that, and I'm on Twitter too much, but the discourse around this topic is really interesting. Some people are super for it, some people are kind of against it, but I feel like those that are city dwellers are really into this idea because they know the struggles of current city planning.

Daniel: Well, that's what I was gonna say because I've got a couple of friends who live in New York City. One of these guys, Tim, I don't know if you listen to the podcast or not, but Tim, every time you post something, it seems like it's trying to convince people that we need to make more walkable and more bikeable cities. And one of his major complaints is like, cities are full of people. Not many of those people actually own cars, yet the way that we urban plan in cities right now, is highly focused around accommodating cars. We've got space that we use for parking lots, parking garages. Most of the space is used for roads. We make the roads wider, which makes cars travel faster, which makes it even less safe for cyclists and pedestrians. It's a painful hurt cycle. And speaking for my friend Tim, who lives in New York City and doesn't own a car and doesn't drive there, it makes it harder for people that aren't in cars to be able to live successfully and thrive in these cities. So, I appreciate what you're saying here, this gap in urban planning, this gap in the way that cities have been planned to date, and there's a potential to improve it by focusing on making everything convenient, etc. The challenge there being, urban planning is really, really time consuming, and when you try to challenge these conventional assumptions and try and make a world in which anyone can access everything they need within 15 minutes, regardless of their mode of transportation, becomes even harder. So, it becomes even more time consuming, it becomes even more challenging. And essentially, traditional methods of planning cities are mainly based on just that tradition, right? So, they use heuristics, they use manual processes, they use rules of thumb rather than optimizing for actual, you know, an optimized design. So, what we end up with cities that are built based off of advice from a book, but not built with the best, latest and greatest technology to try and optimize what the outcome is for the people that actually live there.

Farbod: Dude, you hit the nail on the head as always. And like, I feel like we've hinted at it, but let me explicitly say it. For the most part, a lot of current cities suck and there is a great level of interest in changing it to make it this ideal 15-minute version of a city. So, urban planners are gonna have a lot of work on their hands in the coming years to either develop new cities that match this criteria or kind of reinvent existing ones to be more accommodating, right? And just like you said, there's a lot of limitations here. Urban planning is very time consuming, it's iterative, it's for the most part manual. It relies on a lot of rules of thumb that you might need to throw out the window if we're gonna do this or you might need to have a whole lot of experience to know the drawbacks of, let's say you're trying to optimize for ease of access to X while also reducing the road and so many other parameters at the same time, you might need to have a lot of experience to know what these various drawbacks are.

Daniel: Well, I will say, no offense to my friend Tim, right? It's really, really easy as a keyboard warrior to go on your Instagram and say, hey, we should really replan cities to look like this, this and this and optimize for this, this and this. I can say this to Tim because he is an AI engineer and he probably could develop this tool, but not many people are willing to understand the work that it takes to actually plan a city around these optimized tools instead of using convention, which are what we've used today out of necessity because we didn't have literally the processing power, the human processing power, we didn't have it to be able to plan a city in a timely manner to where we're able to actually develop things on time and scale a city to a scale in which it keeps up with the population growth. But enter AI, we've now, it seems like, based on this research that we're talking about today, we've now potentially got the opportunity to use AI agents that are trained over millions of iterations to plan a city in a way that's not only optimal, that not only are the outcomes optimal for the people who are gonna live there, we can say with a pretty high degree of confidence, it's better than humans at planning cities. It's much better than humans at planning cities. So, can you kind of walk us through what it actually is that they did?

Farbod: Yeah, yeah, yeah. And before I get into it, by the way, it looks like these urban planning folks were very aware of the benefits that competition could have in their field as early as the 1970s because they started using Excel models. And as things got better and better, they even started using AI. So, there are AI products for doing some level of urban planning. But the major downside has always been that a lot of these tools do not address the specific layout of like, this is where you put a hospital, this is where you put a school, and this is where you put the roads. Apparently from what I read in this article, they were saying we've been able to do, for example, building layouts, but not so much as like, here's an entire layout for the city. That's always been the drawback because it requires that human touch. Like, of course, at its core, it's computation of like getting something quickly, but that insight of, okay, like what's actually good and not, that's, like you said, it's just like, you know, hint, hint, we have an episode coming up about blacksmithing, a little bit of knowledge that you need to have, and that magic cannot just be captured by simple Excel.

Daniel: Well, and I agree with you there, and one of the nuances that I wanna point out that I had gleaned from the article the first time I read it is, not only do urban planners have to provide very specific inputs around what should be where, which spaces should be green areas versus which space should be developed, they also can't say, we have to have a grocery store here, we have to have a hospital here. Based on the policy of wherever they might be city planning, maybe they can only suggest a certain type of zone. They can zone this as residential building, and then they've got to provide the correct incentives for developers to actually come in and build those residential buildings there. So, in addition to like not being able or tools not being able to specify for them, hey, a hospital should go here, an apartment building should go here, a grocery store could go here. The actual inputs that they have to the way that a city ends up being built is mostly around zoning rather than them sitting there with their Legos building like, oh, I'm gonna make this an apartment, I'm gonna make this a hospital. A lot of places where there's like a public market with enterprises involved, one of the only things that they can do in that economy is like provide some input, some zoning as to what the building should be, and then someone else has got to come build it. So that makes it even more complex when they don't control the actual outcomes either.

Farbod: Exactly. But yeah, let's, let's start getting into the sauce of it all, right? Like what, what did these folks do? We talked about it. There's AI tools before, but they came up with an AI tool. So, what makes theirs so much more special and so much more effective than what we've seen before, right? Now, this article was definitely a little intimidating, I'm not gonna lie, but I will do my best to try to digest this sauce as well as I can. Reel me in if I'm going too far into the weeds.

Daniel: I will, I promise.

Farbod: Thank you, thank you. All right, so the best way I can go about this is by saying their sauce is a two-parter, right? So, they're using deep reinforcement learning. We've talked about it before on this podcast. It's where you have a machine learning model that says we will train you by telling you when you did good and punishing you when you did bad, which allows you to converge on an ideal solution much more effectively. The reason that this is so necessary is because they said, hey, when we look at a modern space, a modern city, we can have 4,000 to the power of 100 potential combinations that we would have to…

Daniel: So, lots and lots of combinations.

Farbod: Exactly. So, an exhaustive search where you would put every combination together and then analyze them is just not practical whatsoever defeats the purpose of efficiency that we're going for here. The way this deep reinforcement learning model breaks down is you have one value network and the entire role of that value network is saying, here is my standards for the 15-minute city, how well does what you're coming up with live up to that? And then you have two policy networks, one for how you're making use of the land, the other one for what areas are you using for the road and how are you using the road. Both of those propose like, hey, here's how I want to, I wanna put a road here, and then the other one's like, I wanna put schools here, and then they talk to the value one, they're like, how well did we do? And it provides some feedback on how well they did. The other side of this is the graph neural network. That's like that real saucy sauciness of the sauce, where they're taking the topography of the entire the landscape that they're trying to do the urban planning on. And they call it the encoder. And it really acts as like a transition layer between all of this stuff that's happening. So, it gives context of like, you said this, but what does it actually translate to in the real world? And then the other ones can interpret it using this like middle layer language.

Daniel: Yeah, it's a happy medium between the actual plan out in the real world.

Farbod: And it's abstracted.

Daniel: Yeah, it's really abstract world where this AI agent is operating in the one that's got, you know, it's been trained over millions of different simulations, all these different potential combinations. The graph neural network is the middle ground where those two can meet and this AI agent can interact with the plan in the real world. The way I like to view it as it's like, that's like the virtual brain that does all the city planning and the graph neural network. It it's taking into account all the input from all the different possible combinations that the AI agent is considering. And then it's also combining that with the actual urban environment to understand what's feasible, what's not. And it creates a bunch of different nodes on this graph that signifies, each node signifies a different urban element. So, like one of these nodes could be a road. One of these nodes could be a park. And then the edges between the nodes show how those are connected or related.

Farbod: Exactly. And it reminds me of those, I don't know if in chemistry you guys ever made like the elements with those sticks and spheres. It looks a lot like that where every sphere is an element like a school or a hospital. And then those connectors resemble the use of the land itself. As you can see folks, if you're on the audio, I mean, the videos, had the podcast, we have a little guest, Kishmish is joining in. She's a big fan of urban planning. So, she just had to step in for this one.

Daniel: Yeah.

Farbod: But yeah, so I'm a big visual person. It was really cool for me to see that. And that really helped me make more sense of how these, again, two different parts of the sauce are working together. But of course, this is all high level abstract. We want to know the impact. We want to know what the thing that these guys came up with is actually resulting in. So, you want to get into it?

Daniel: Yeah, I'll take it away in terms of at least efficiency this is what I’m saying at a high level I don't want to sensationalize this too much, but if we like where to put humans up against AI and say, all right, churn out this city plan that successfully achieves a 15-minute city where I can go from, you know, my grocery store to my school, to my bank, to my hospital, et cetera, everything that I need to live successfully in a city. Can I achieve that in 15 minutes? This takes hours and hours and hours hundreds of thousands of hours. Maybe a thousand of hours to possibly achieve this depending on the constraints. This AI model can turn out multiple viable spatial plans in only mere milliseconds so again not to sensationalize it too much, but the human versus AI aspect, especially in terms of just pure amount of time required to turn out a plan that works, AI is blowing humans out of the water and I think it's pretty incredible to be able to achieve something that truly humans didn't have enough brain power to think about to be able to solve on their own. Or it would take lots and lots of humans a long, long time to solve this. And AI model can now do multiple versions of that in mere milliseconds, which is crazy.

Farbod: And to extend what you're saying, they made a point of it in the paper to say, not only can we do that, but by the way, if you have a model that's trained well enough against a single scenario it can then pretty much become scenario agnostic. For example, you use a plan to optimize Washington, D.C. You can take that and do it with Chicago, New York City, LA. Like, that's pretty powerful and that's pretty fascinating to think about. And...

Daniel: Well, it's not a one trick pony, right? Which is...

Farbod: Definitely not.

Daniel: In many applications for AI, that is the case, right? Maybe they're traditional or many other competitive AI models get really, really good at developing a 15-minute city model of Washington DC. And then you go take it and try to reoptimize for a new, brand-new city or for Tokyo or for Berlin. And it breaks. It's pretty incredible that this team from Tsinghua University in Beijing was able to do something that not only worked for Beijing, but works for a future city, works for other existing cities. Right. It shows the robustness of their AI model. Not a lot of AI models are that robust.

Farbod: Right, right, and that's for developing new urban plans. But here are some other cool statistics that I think they're worth talking about. In the article, they talked about how this specific AI improves spatial efficiency, so that's how well you make use of the existing amount of land that you have, by approximately 50% over other leading algorithms.

Daniel: Which are probably also better than humans.

Farbod: Exactly.

Daniel: But in terms of their competition versus other competitive algorithms, they're 50% better.

Farbod: And when you, this was also for like new spaces, but when you look at existing spaces that you want to renovate, rejuvenate, give a little love back into, make it 15-minute walkable, it improves accessibility by approximately 20% in comparison to human experts. So, even in an area where like the AI didn't optimize everything to be perfect from the ground up, it can still make it 20% better than the best humans that we know of could, right?

Daniel: Yeah, I mean, that's incredible.

Farbod: It's a no-brainer to use this tool, right? But what I want the main takeaway for our audience to be, because you know, you said it yourself, you don't want to sensationalize this. We are not trying to say that AI should replace urban planners. If anything, these folks in this paper said the best use of this tool is in collaboration with urban planners.

Daniel: Yeah, we talk about this term a lot in terms of robotics, right? We talk about collaborative robotics, cobots. This is an opportunity for collaborative AI. And at first, I was like, man, if I'm reading this and I'm studying civil engineering and I have aspirations to become a city planner, I'm gonna be super disappointed and be like, man, AI is taking my job, but this truly isn't the case, right? AI is not adept enough and we don't trust it enough and it truly isn't the proper use of the technology. Coming from the people who developed it, they say don't turn AI loose and let it plan your city and blindly go develop that plan. No, they say, let the humans focus on the conceptual part of the planning, let the humans focus on the high-level part of the planning. And then when it comes down to the really nitty gritty, right. Running through all the different permutations, trying putting hospital in location A, does this work? Put the hospital in location B, does this work? Instead let's put the hospital and location C, does this work? Instead of doing all that manual planning and plotting and graphing that a current city planner would have to do, let the AI do that for you. And that to me now changes the outlook from me saying, man, if I'm studying to become a city planner, I'm really disappointed too. Man, if I'm starting to be a city planner, I'm really excited because it sounds like a lot of the nuisance part of my job, a lot of the really nitty gritty part of my job is gonna be resolved by this AI tool. And then city planners get to focus on what they wanna do best, which is optimizing for service, optimizing for ecology, optimizing to reduce traffic, optimizing for equity and equality in terms of outcomes for people who live in different neighborhoods. Someone who lives in a certain neighborhood shouldn't be subject to the fact that they're in a food desert. Exactly. Because of the way that the city was planned. That's something that a city planner wants to fix, but they don't necessarily have the time or the resources or the energy to be able to solve that problem. Now you've got an AI tool that's potentially going to be hitting the market soon that these teams can use to actually achieve these outcomes and make those plans more viable. So, I think again, we keep saying it, we don't want to sensationalize this too much. This isn't really humans versus AI, even though it's fun to measure the humans versus the AI, it's really, let's see what humans can do with AI to improve city plans.

Farbod: Exactly, it's benchmarking against what we currently do to see what the potential is. And one thing that was really interesting to me is early on in this paper when they were proposing why they decided to use deep reinforcement learning, they were like, look, we've already seen that this approach is really, really worthwhile when you look at chemical synthesis, right? We've done episodes on this before where we talk about the new drug development processes. Used to have a chemist do like, oh this, trial that, trial this, trial that. Now they can come up with adversarial neural networks where they can fight against each other, have certain policies that keeps motivating them to go after the right stuff. And it doesn't mean that you're getting rid of the chemist, right? The chemist is still very essential, but that very just monotonous process of like trial and error can now be offloaded to something else. And that human touch that big level conceptual thinking, that's the big value add that they're bringing into here.

Daniel: Yeah, and I mean, it makes me think of you and I both took courses in college where we studied finite element analysis. And one of the things that my professor did, I think I had a different one than you did, but one of the things my professor did to make us appreciate the value of our finite element analysis tools is for the first half of the semester, like leading up to the first midterm. We had to do all of our FEA by hand, like notebook and paper. He's like, analyze the situation. You have to do all the math manually.

Farbod: That's brutal.

Daniel: It was still exciting when I was doing it. I was like, man, this is awesome. I can do all the math, just me and my calculator, my notepad and my pencil. I can do all the math to analyze a certain scenario and understand without ever performing a physical test in the real world, I can understand whether my design's gonna fail or not, based on math. That was super cool. But was even more mind boggling for me was when the second half of the semester, he's like, all right, now I'm gonna teach you Ansys. And we were able to use the state-of-the-art technology that industry professionals use. And I was able to solve this problem that literally took me days to solve on a notepad before. I could run it and it would solve in less than 10 minutes. Like this is what I feel like we're achieving a similar level, if not even a better level of convenience, of technology unlocking potential outcomes. Like imagine how much better mechanical design got once simulation tools were available. Yeah. I can only imagine how much better city planning is going to get now that these AI tools that help solve cities, which were unsolvable before to make city plan real, true, viable, successful city plans available to any city planner anywhere when they want to use it.

Farbod: No, that's a great analogy, man. I totally agree with you. I'm going to do a quick TLDR wrap up of what we talked about. So, in a nutshell, cities pretty much suck. Most people have been begging for 15-minute cities and if you don't know what that is, imagine living in a city where everything is a 15-minute walk or bike ride away from you. Hospitals, schools, core businesses, things like that. But of course, most cities today are taken up by roads because they're just designed around cars. This has led to what's gonna be a coming revolution of urban planning. Now, it's a big ask for our urban planners because they have to optimize for multiple parameters at once and that can be really difficult even for the experts. So, in comes AI. These folks over at Tsinghua University, which is the MIT of China, have proposed an AI, which is not necessarily new, but the implementation of it that they've proposed results in efficiencies that we really haven't seen before. Because in comparison to other leading algorithms. It makes use of the space about 50% more. And in comparison, of renovating existing spaces, it does it 20% better than human experts could ever do. Now, you might think, well, does this mean AI is gonna replace urban planners? Absolutely not. These folks have actually suggested that the best implementation of this AI is in collaboration with urban planners, because it can offload that very monotonous computation side, which again, you're optimizing for a lot of parameters, and then the urban planner can focus on the high-level conceptual stuff. So super exciting. I look forward to living very close to a 15-minute city in the not-so-distant future.

Daniel: Nailed it dude. And yeah, I'm with you. If this technology makes it out of the real world, which I hope this team from Tsinghua University is able to do, get this technology into the hands of city planners. Start making a difference right away. Cause remember, it's not just good for brand new cities being built from the ground up. It also helps to optimize currently existing cities with proper renovations to make sure that new renovation plans in cities are also helping improve them, not make them worse.

Farbod: Imagine being able to go through DC without breaking a sweat.

Daniel: Or New York City or Boston. I mean, I think the US cities are maybe some of the worst in the world in terms of some of these markers, especially traffic and food deserts, et cetera. But I think that all around the world, there's a lot of people who can benefit from this technology.

Farbod: Agreed, agreed. Before I forget, is there anyone that we're supposed to be shouting out this episode?

Daniel: Yes. A new reviewer, 5-star review from Dinosaur Bagel. Thank you, Dinosaur Bagel. We love dinosaurs and we love bagels, so we love you. But also, thank you for your review. Dinosaur Bagel said, “This is the best podcast, great hosts, easy to understand, and they talk about complex topics. Can't recommend more highly. This is the top tech podcast, always worth a listen.” Thank you, Dinosaur Bagel from the USA. We appreciate you.

Farbod: Those are high praises and because you got bagel in your name, I'm going to celebrate this weekend with a bagel. So, thank you.

Daniel: There we go.

Farbod: Yeah. All right, folks. Thank you so much for listening. And as always, we'll catch you in the next one.

Daniel: Peace.

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.

To learn more about this show, please visit our shows page. By following the page, you will get automatic updates by email when a new show is published. Be sure to give us a follow and review on Apple podcasts, Spotify, and most of your favorite podcast platforms!


The Next Byte: We're two engineers on a mission to simplify complex science & technology, making it easy to understand. In each episode of our show, we dive into world-changing tech (such as AI, robotics, 3D printing, IoT, & much more), all while keeping it entertaining & engaging along the way.


The Next Byte Newsletter

Fuel your tech-savvy curiosity with “byte” sized digests of tech breakthroughs.

More by The Next Byte

The Next Byte Podcast is hosted by two young engineers - Daniel and Farbod - who select the most interesting tech/engineering content on Wevolver.com and deliver it in bite-sized episodes that are easy to understand regardless of your background. If you'd like to stay up to date with our latest ep...