 lots of pictures, although if you're not really sure where Newcastle is in England, you might just want to have a quick sneak look at Google Maps because I realized I didn't put an overview map in my slides. So the first thing I'm going to talk about is how we conceptualize coverage when we talk about sensor networks and why it matters, right, because I think that a lot of times when we talk about coverage or smart cities or bias, it's implicit what sorts of measurements of coverage or representation would help me about. And then I'll transition to talking more particularly about how we could support decision making. So what tools could we put in individuals' hands that could help them make decisions about where sensors should be placed in a city. So getting started, thinking about coverage when we talk about sensor networks and why I think it's important to spend a little bit of time on this, right. So lots of people have talked about smart cities, their tendency to polarize, the ways in which spatial inequality matters. Knowledge is produced for some parts of the city, but not for others. This can lead to enhanced spatial inequality. It means some voices are never heard. It means that some people don't have access to levers or avenues to power or even understanding sort of what might be happening in their neighborhood. And there's no reason to expect that any of this is going to change in the future, right. Now, one thing that I would say about the quotes that I have on this slide is that if you follow the news or if you read on this subject, we talk a lot about surveillance, bias, AI, machine learning, the ways in which some people get lost or decisions get made that either reinforce privilege or disadvantage. But the point that I'll be making throughout this talk is that this is implicitly spatial in the case of smart city technologies where you're making decisions about where and what to measure throughout a city, okay. And so to me, coming at this from more of a demography or a population standpoint, I'm really interested in who is where in the city and whether or not we're measuring information for them, okay. So on that note, right, if I think about surveillance technologies, I think that we have sort of especially these days sort of a, I don't want to say knee jerk is strong, but a tendency to assume that surveillance is bad, right. And so this was my effort to, with a Venn diagram, differentiate between different kinds of surveillance technologies, right. We have the ones that we could think of that are monitoring people. And here I'm thinking about gunshot movement throughout the city, people's behavior, right, following people with CCTV. And we could think about then technologies that measure stuff that affects people. And this is the stuff I really care about, right. This is air quality sensors in the case of the examples I'm going to talk about today. And I would suggest that there's a difference because where we might over measure for noise, for example, or traffic, which I see is sitting at the intersection of these two things, where you might mean to measure how congested the flows are through a neighborhood, but inadvertently you're measuring how people are behaving in space, right. The same for noise, which on the one hand could be a nuisance, right. It's a, it can affect health, it can affect well-being. On the other hand, noise is a natural outcome of people going about their daily business and we can very quickly sit sort of in the gaps between caring about what is good for people and actually surveilling people in a risky way. So the point that I really want to make with this slide is that I think it matters where we place air quality sensors because we're more likely to not measure than to over measure, right. So we really want to make sure that we're capturing the groups that are important. And we can think about sensor deserts as areas of the city where we're not measuring, right. So where we don't have an air quality sensor. And you've probably seen this graphic before, this idea of survivorship, bias and airplanes and deciding where we should reinforce vulnerable pieces of aircraft before they go into battle and realizing that of course it's not the places where they've come back and they've been hit by a bolus because they made it back, right. The part of the airplane we really care about, we don't observe because those airplanes never came back, right. The point for cities is a little bit different, right. If we think about where cities have been placed in, in where sensors have been placed in the city, it's not so much a question of, of survivorship as it is a question of policy development and understanding of how, how a city works, right. So if we don't have a sensor in a, well, I mean, let me step back a little bit. If we only have sensors in particular parts of the city, right, we might ask ourselves the question, are we only measuring places where we have good air quality so that we don't produce any knowledge for poor air quality places, or are we only measuring the air quality in the neighborhoods that we really care about, right, in which case we, it's not even that we think we're ignoring the poor air quality, it's that for some parts of the city, for some groups of the population, we simply don't know what's happening, whether it's good or it's bad. So we think of sensor deserts as parts of the city that are not covered by existing sensor networks, places for which we're not producing knowledge. In this case, we don't know whether they have good air quality or poor air quality. And if we were to think about this from a modeling standpoint, so generating surfaces of air quality, you might think of these places where there are not sensors as being areas where you have a larger uncertainty bandwidth. So we've estimated what the air quality is, but we're less sure about what it is relative to places where we do have the sensors. And when we think about sensor deserts, we could think about how we work through placing sensors inside a city, right? One way of thinking about this is that you want to place it at source, you want to place it at the congested intersections, you place it at industry, so that you're measuring the polluters as, as the air pollution enters, enters the environment. Another way of thinking about it, though, is from a vulnerability perspective, right? Why do we care about poor air quality? Well, we care about it because it's hazardous to young people's health, to old people's health, to those who are more infirm. So maybe what we want to be able to say is something about exposure to poor air quality, right? And that would suggest that we want to disperse our sensors throughout the city, where people are, especially where vulnerable populations are. There's another point to this, though, which is that why do cities place sensors? Why do cities want to be smart? Why do they have sensor networks? Why do they spend money on these things? Part of it is for the data, part of it is to be able to measure air quality, but part of it is to be able to say that they're a smart city, right? So there's a little bit about city branding. Some of it is also about political representation, right? So in this case, you might think of a sensor desert as sort of a material desert, that is, you have neighborhoods or parts of the city that aren't covered by a sensor, and they don't feel themselves to be represented by, by the current technologies, right? This is a little bit different from places that are falling outside sensor catchment areas. So if you look at the bottom of the tables at utilitarian or technical sensor deserts, these, and this is what I'll talk about in the second part of the talk, are a little bit trickier because we have to make decisions about how sensors function, right? What is their catchment area, right? What is their watershed? For what distances do we actually get valid measurements of air quality? And it turns out that this is pretty tricky. Okay, so now I'll give an example or two. So the first sort of bigger example that I have is from the Urban Observatory, which is here in Newcastle. And it's the UK's densest air quality network, but has all sorts of sensors. So it has traffic and air quality, it has smart building observations. It has noise, it has temperature, it has precipitation, some traffic measures, lots of different variables, all freely available to have a look at if you want. And it represents a really large investment on the part of our research infrastructure here in the UK. So this is what Newcastle looks like. It's in the north of England, about an hour south of Edinburgh on the east coast. And we sit on a river, the River Tyne, which is why we're in Newcastle upon Tyne, just to the north of the river. And so you can see on the left here, some of the sensors we have air quality people, whether beehives, water quality. And across the entire region, we have 196 of these air quality sensors, 144 of which are in Newcastle. And this is the way that they're spread across the city. And you can see that down where the largest cluster is, that's the center of city. So that's downtown. And then we have a quarter running up north. And then we have some sensors sprinkled throughout the rest of the city. So the real question here for me is, what kinds of neighborhoods are these in? What kinds of people are covered by this existing sensor network? Can we predict the type of neighborhood that you have if we know that you have a sensor? Right? It's a very small share of neighborhood. So these are called lower super output areas. They're small sort of the equivalent of a census tract or a neighborhood. And what we first do is compare them to what's called the index of multiple deprivation. Right? So this is produced by the national government produces this for 38, almost 33,000 these small areas across England. And it's a multi dimensional index. So it covers everything from employment to health to crime to sort of measures of built environment. And it ranks every one of these small areas from the least deprived, the most deprived in the country. Right? So when I show you the most deprived, the red areas for Newcastle, these aren't just the most deprived areas in the city, but actually in the country, right? Okay, so if we look at Newcastle, the black circles with the numbers just show the number of sensors. So you can see in the center of the city are where we have large clusters of sensors. And this makes sense if you think about this from a traffic congestion point of view, right? This is where a lot of movement is happening with buses and cars. And we want to be able to track that kind of air quality. There's a lot of commuting that goes over the river. Every day, for example. It's a little bit difficult to see here. But one point that I would make is that if you look at the darker red areas that are sort of shaded out, it's actually a lot of the more deprived neighborhoods in the city that are not covered at all by a sensor network. And a lot of the least deprived and these blue, right, these are the least deprived neighborhoods in the entire country. So not just in the city. Alright, and if we compare this across different types of sensors, right, you can see that in fact, for most of the sensors, the less deprived neighborhoods are more likely to have sensors and the more deprived neighborhoods are less likely to. So it's air quality, but also holds for noise. In this case, holds for weather sensors for traffic for people. And so we would then suggest that there are some parts of the city where we don't have, where we do have deserts and where we don't have representation for groups that we think we should be representing. So another way of thinking about this is that of the most deprived neighborhoods in New Castle, five of them have large roads. So these a roads are large, almost highways that run through neighborhoods. They're also deprived. So they've got the they've got the pollution source. They have the vulnerability, right, as assessed by this index of multiple deprivation, and they have no sensor at all. If we would look at other aspects of vulnerability, so so children tend to be more vulnerable to poor air quality. We see that because of the way we the way young population is distributed in New Castle. And because most sensors are in the center of the city where there aren't a lot of people living, where you end up is not actually capturing four place of residents, so much of the young population. And we see something similar for the elderly population. So this is for those age 75, and up, right? Yeah. And just to give another sort of counter example, this is brief, this is from the array of things in Chicago, right? And just, this is probably the most famous sort of smart city sensor network in the world. And certainly, if you follow this, then you would be aware of it in the US. And just to point out that on their website, when they describe sort of where they'll locate sensors, they note that node locations are chosen with input from researchers, neighborhood groups, city departments and community members. So sometimes what you find with these sensor networks is that there's a deliberate sort of a deliberate plan of either uniformly distributing sensors across the city, or ensuring that every neighborhood is able to say that it has a sensor. So there's this representation, but the political aspect of knowing that that when you're walking down the street, your air quality is being measured. And what we see for Chicago, and we've done this by median income, and also by the share of the black population, that it's much more equitably distributed across the city. And this to me is is interesting partly become an American and I've lived in Chicago, but also that Chicago is one of the most segregated cities in the US. Right. And so if you were to think about where we might be most likely to observe these sorts of spatial inequalities, this would be a city, right. But in fact, what we find is that the sensors are pretty evenly distributed. So there's probably something about the political process, the decision making process that is explicitly geared towards equity or inclusion that leads to a more even distribution of sensors across types of neighborhoods. So I'm going pretty quickly. So how can we support decision making where sensor placement is concerned? So here we switch gears a little bit. And the first part of the presentation, I show you where after the fact sensors have been placed in the city and then we assess how fair it appears to be. But ideally, if you're thinking about cities that have to work through the process of deciding whether this is a way that they want to capture information for the city before a sensor is ever placed before the investment is ever made, we ask, wow, what sort of investment do you want to make? And what is the goal of having a sensor network? What is the coverage that you're looking for? Is it to let every neighborhood know that they can check online to see what the air quality is? Is it to capture where you think the most affected areas of the city are? Is it to capture vulnerable populations? And to us, this is important because not so much because we can deliver optimal solutions. Although sort of the end goal is to be able to say this would be the best set of locations for a given number of sensors. But actually that we believe that working through the process of assessing trade-offs, that having to do that work allows policymakers to understand better what is they're actually doing. And I think it's probably better for the average person, citizens and city inhabitants too, to be able to see why it is that they observe the pattern of sensors in their city, that this is actually a complicated not straightforward process. So one of them is this monitoring priorities that I spoke of earlier, that you have on the one hand, whether you want to place sensors only at the source or if you want to cover more broad demographics. And this is going to be tricky because even with 144 sensors in Newcastle, you could see that there were big swaths of the city, especially in terms of just geographical area, if not population, that aren't covered by any sensor whatsoever. And sensors are going to be valid for relatively small areas. They're very sensitive to the built environment. So the height of buildings, narrowness of streets, that sort of thing. So even though we're going to show examples for sensors that have sort of coverage over a larger area, about half a kilometer, even that is sort of pushing it, even that is being generous. So in fact, even though I sort of posit this as two different trade-offs, the fact is that you're going to be limited in your investment that you can make and you're going to have to probably choose some combination of both without actually having a perfect solution. And then there are other trade-offs. So sensors are expensive. They require upkeep. And you have to have a place to put them. So I didn't put that in here. But you also have to have the right to hang them somewhere. So often on a lamp post or on a building. So you have trade-offs that are that you're probably not going to have as many sensors as you want it because they're expensive. You also have the challenge of both demographic and spatial coverage. So you could aim to cover the total population, but because people sort over space, we know they sort by income level and income. We know they sort by race and ethnicity. We know that they sort by age. That is, some kinds of people are more likely to live in some parts of the city than others. It means that it's going to be very difficult with a fixed budget and a limited number of sensors to cover everyone. So no one's going to come out of this with complete representation. And you have the trade-off too that depending what sort of city you're representing and what your role is and how your city runs, that you may be wanting to place sensors where people can see them, right? If it's taxpayer dollars that are paid for this, they want them to be visible. And you may also want to spread them out across the city just to help neighborhoods feel like they're participating in this process. So as I said, these are two hopefully tools that will be intuitive that can help visualize and work through these trade-offs and outcomes before you locate sensors. And I don't show it. I'm not going to show you any drafts of sort of the decision support tool but more focus on the unidimensional location of the sensors themselves with just one constraint or one category that you're looking to maximize. Okay, so I laid out all of these different ways that we could measure coverage, right? You could place a sensor in every neighborhood. You could focus on groups. You could focus on source. But there's the tricky technical bit, which is this bit of, well, what are we going to call coverage? How are we going to estimate that for a sensor, right? So if we have our neighborhood and we have our sensor network, we're going to have to choose that sort of bandwidth that we're considering to be coverage. And that already is a pretty big assumption, right? But with the decision support tool, as long as you can sort of provide decision makers with an idea of what it is you're getting with, right? A larger bandwidth isn't giving you exact measurements for air quality. A larger bandwidth is sort of saying, okay, this is the representation. This is the larger neighborhood that we're going to cover. Whereas a smaller bandwidth is going to give you more accurate measures for a smaller area. And then you're really saying, well, these are this is where we know we can get good measurements for air quality. And that's what we're going to call coverage. And that's about neighborhoods saying we have an air quality sensor. All right, so one approach is thinking about what is the best or sort of optimal allocation of a given number of sensors given a particular goal. So we use what's called a greedy algorithm. The sort of bandwidth that we're using theta is set to 500 meters. And so what this means is that these geographies that we use for this piece of the analysis are what are called output areas. They're very small. They have about 125 households on average. So you're easily with this 500 meters capturing your output area and then a little piece of the fraction of the areas around you. So you'll locate your first sensor in the area where you capture the largest share of whatever piece of the population you want to capture. And then the next the next sensor is going to be located somewhere where you get the next best given that you've sort of shaved off fractions of neighborhoods surrounding the sensor. If that makes sense. And we place our sensors at these sort of output area centroids. So a more realistic way of thinking about this would be if you actually had the infrastructure that you knew you would place the sensors on. Right. So you have your potential sites. But we just go with with geometric centroids. And so this essentially means that when you place a sensor you cover as I said that entire output area so that entire polygon and then a little piece of the polygon around. So this is what it looks like just to illustrate for the total population of Newcastle. And the way the algorithm works is that it runs from it places the first sensor and then it runs up to 60. But after 30 which is what I show on the right side of the slide. You're at 40 40 percent of the total population and at 60 you've only made it up to 53. Right. So with 60 sensors you're only covering about half of the population with this bandwidth. So this means you really want to be able to justify the choice of you know so how you're conceptualizing coverage in this case. But you can see that what it does if you can remember sort of what that outline was of Newcastle to start that the sensors are going to locate primarily in the center of the city and then sort of look their way out because this is where population density is the highest. Right. But if you remember those maps that I showed you for the young population for the older population that's not what the distribution of these subgroups looks like. Right. So this this would give you good numbers if you're talking about total population. But as soon as you start to break it down by demographic subgroup right in this case age but it could be race and ethnicity. It could be health status for example of what the distribution is of more healthy and less healthy people across the city. You're going to run into a little bit more trouble covering more than one group of plants. Okay. So as I said people sort across space. These visuals give you a sense of population density of different subgroups. Right. So it's highest for the total population in the center of the city. And then as with most cities as you sort of move out of the city polygons get larger population density decreases there aren't as many people. So if you place a sensor out on the fringes of the city you're not going to cover as many people. Right. You're going to get bigger bang for the buck by placing sensors closer into the city. But because the geography is a little bit different for children and for the older population and it's a little bit tricky to see here but you can see in the middle panel that for children 16 and under they're mostly on the west side of the downtown area or anyway there are a lot of children there and then they sort of spring spread out in a ring around the center city. The older population sort of has a sort of broader more dispersed geography but again not in the center of the city. So here we're just going to watch this run for a minute and we won't run all the way up to 60 sensors because it takes a couple of minutes but you can sort of see how it's placing the sensors and sort of compare how quickly the geographies diverge. Right. So they start off especially for the total population and for children 16 and under fairly close into the city but you'll notice for both the youth and the older population that there's that sort of hollow in the middle of the city. In fact where for the urban observatory we saw lots of sensors because that's where there's a lot of traffic and that's where you have lots of workers. So the density of sort of people passing through in the city center is really high. Right. And you can see with the older population. I'm sort of pointing out my screen but you can see that it's sort of we more quickly start to move out to the fringes of the city. And I'll just pause this when we hit 30 sensors and we can sort of see that. Oops. Let me just go back. I just wanted to show how now we've lost it. But what I want to show you is that there's not a market difference between the different subgroups in terms of the share of the population that you capture. Right. So when you stop at 30 it's about 40 percent of all of those groups give or take a couple of percentage points. Okay. So we had this challenge rate that people sort across space. But the other challenge is that people aren't stationary in space. Right. And this becomes doubly tricky now. Right. So if I had given this talk in April at Columbia which is when I had hoped originally to be presenting some of this I would have said hey everything I've shown you so far is residential population. This is where people are at night. Right. But it's not where people are during the day because during the day the children go to school and the adults go to work. Right. And this is going to vary across subgroups. Right. So some parts of the city will empty out completely during the day. Right. But others won't. Some will have people who stay home and you can think about differences between say children in the older population where you might be more likely to keep your residential population during the day if you have an older age structure than if you have a younger age structure. Right. But now with the pandemic and we call it lockdown here I think it's even more complicated because now people are back at home and some people are at home and some people aren't and probably they've changed the way they're moving from from origin to destination. So this means that if you're thinking about sensor networks from an exposure vulnerability perspective it's a much more complicated challenge than I presented to you so far because if you care about sort of exposure from an epidemiological perspective how many hours of how of a day do different people spend exposed to poor air quality when you actually need to know how people are moving through the city and we need to know something about them so that we can subdivide them into groups. OK. So what I'm going to give you is just the simple case where we look at where people go during the day if they go to work. So we look at working population. OK. And and I don't present this here but in sort of a in sort of a multi criteria decision support tool you can also play sensors for fixed locations like nursing homes hospitals schools. So this is one way if you know that you have children dispersed across an entire city but they're all going to converge in a few locations for a substantial part of the day you want to measure their air quality at school and you can get pretty good coverage in that way. Right. So really if you were a decision maker a policymaker what you might do is sort of take your N number of sensors and distribute them across different categories. You put some at schools. You put maybe some at hospitals or nursing homes. You might put some in the center city to capture the traffic. You might put some out in the neighborhoods so that people know that they're being counted. Right. OK. So we have this dilemma of people not being stationary in space. So here if we look at place of work. All right. So everyone comes in from the outlying areas of Newcastle and you can see I think already just one of the quickest snapshots of the results is that workers cluster pretty highly in space in Newcastle. Right. So they come into the city center and then there are a few other pockets of workplaces and this means that with just 10 sensors we can cover about half of the of the working of the sort of daytime place of work population. And by the time we get the 30 sensors we're at about 70 percent and some of this would be double counting right so that on these sort of outlying areas you get the you get the working population but you're getting other subpopulations too. So this is why even I'm only presenting this on one dimension what you really want is to be able to stack various dimensions of people so that you can locate your sensors and cover more than one group at once if that makes sense. OK. And so just a quick sort of comparison between the urban and the territory which I showed you before and if we place 60 sensors across the city of Newcastle and just to highlight that you do get some differences. And even though with 60 sensors we're only covering half of the Newcastle population so there's still half of the residential population that we don't get. We do cover more people and more places than the sort of ad hoc location procedure of locating the urban and the territory sensors. But again, this was a little bit different of a process for them because part of it was as I said before locating where you have heavily trafficked areas. Part of it was the visibility of the city center and part of it was actually neighbors neighborhoods asking to have sensors. And so you can imagine that if the process is that you have to opt in. And I don't show this but we actually have a piece in the recently there's a recent paper to come out that's on one of the slides in transactions in transactions and geography. I think that's what the journal is called. But you do it for schools too. They had a similar process for schools where schools could opt in to have a sensor. And the minute you have this opt-in process you privilege some groups over others because some know how to navigate these processes better. Some have the time to navigate the processes. Some don't. And so you see these sort of inequalities perpetuated in terms of what kinds of neighborhoods get sensors. And that's why we see a little bit this geography that we observe in Newcastle. But you could see if you remember the map that I had in the first half of the presentation where I showed you the most deprived neighborhoods in Newcastle that do not have sensors. They sort of at the bottom these yellow squares we do much better at getting these more deprived neighborhoods just by sort of agnostic locating sensors to maximize population coverage. So a little bit of an algorithm is a good thing, right? Helps us do a slightly better job than we might do if we were just being human. So a few conclusions, right? So one is just to is just to reiterate that sensors serve a wide variety of purposes in an urban context. So I tend to think about in terms of maximizing coverage, making sure that we could say something about every person and every subgroup across the geography of the city. But that, of course, isn't what cities are always thinking. Sometimes they're thinking that this is a game that they have to play because every other city is doing this. Sometimes it's to meet air quality regulations. So you have to monitor at source what your air quality is. Sometimes it is a sort of political process of elected ward members wanting to be able to say that they have sensors in their neighborhood, right? And so equally, these sensor deserts, I think, are meaningful in multiple ways. One is just the sort of epidemiological perspective, right? If we really want to know who's being exposed to poor air quality, then somehow we need the sensor infrastructure to be able to measure that, right? And that means not only thinking about coverage for sensors, but also where people are in space and how they're moving. So it becomes a complicated question pretty quickly, but I think an interesting one. But we can also think about sensor deserts in terms of representation and production of knowledge. So who do we care about producing knowledge for? Who have we forgotten in these processes? And so sometimes, again, I think that turning to a slightly more agnostic algorithm might help you get better geographical coverage. It's not to say that it's perfect and free of bias, but it is one way to sort of help ensure that you get this sort of fair distribution. And I think representation is hard to get right. You know, I was trying to, a few slides ago, sort of walk through everything that you might be wanting to think about if you or a city manager trying to decide how many sensors to buy and where to put them, you would initially just have a challenge of sorting out what kind of sensor to buy and what kind of upkeep you're going to have to pay for. But then you would have a decision of where to put them and it's a sort of multi-criteria decision making process. So I think in fairness, this is really difficult to get right. And the best thing you can do is sort of work through ways that we catch ourselves, right? That we can sort of make sure that we reinforce fairness where possible. And I have this feeling that anyway, in conversation, when I talk about this research, a lot of times what happens is that either an engineer or a statistician or a data scientist will say, it doesn't really matter where we put the sensors because we're just going to model air quality from that, right? All we want to do is estimate and fit sort of a surface across the city that's going to tell us where the air quality is good and where it's less good. And then we'll know where to target resources, right? My response to that would be the uncertainty piece that, of course, you're modeling a surface, but you're going to be more sure about your values in some places than you are in others, right? And I believe that you certainly wouldn't want your uncertainty range to be too large, right? And in order to keep it within some sort of manageable bandwidth, that's probably going to be more sensors that are placed very strategically. So the sensor network location is a really important part of that puzzle to get right at the outset. But the other response that I would have would be if it doesn't really matter where we put the sensors, why is it that we would observe for some cities then that the sensors are in the nicer parts of the city? And if it doesn't matter, we should just put the sensors in the poor parts of the city and then we still get good air quality models and we're covering the more vulnerable populations. So that would be my argument. And then again, that decision support really involves this sort of multi-criteria purchase. And so what we're working on is not only the sort of the the weightings for the algorithm to be able to to sort of weight, say young population and older population or pollution sources, but also to help decision makers work through that process of deciding how much money you're going to invest, how much your sensors are going to cost, how many sensors you can afford and then how you would like to locate them with sort of intuitive intuitive ways to sort of work through those choices and observe almost in real time how your network changes and how your coverage changes. Right. So it becomes a little bit more of a game and it's not that our goal is to actually give people a print out at the end that says here's where you put your sensors, go hang them up, but rather to show them how sensitive different decisions, how different how sensitive different decisions are to the choices that you make. And I think that's helpful for all of us. So just a few takeaways. One is sort of keeping in mind the challenges of spatial inequality that as I noted at the beginning with sort of the quotes in the literature on smart cities and inequality that we're very aware of the ways in which some some groups and some people benefit from technology and some are left behind. Right. Some of this is about digital exclusion. Some of it is lack of access to technology. Some of it is lack of knowledge about technology. Some of it is the the usual story of inequality, but that underlying this in the case of smart cities and sensors is this idea of sort of spatial inequality and the distribution of the population and the way different subgroups move within a city or state put within a city. And that somehow we need to be thinking about that when we think about why we would invest in smart city technologies and what we hope to get from it. Right. And I think having frank conversations about what the goal actually is is helpful. Right. And just to go back one last time to the sort of multiple meanings of coverage because it's tricky from a technical standpoint. That's where we're sort of still working. But some of the interesting bits actually have been the conversations that we had about how we're going to measure coverage. What do we mean when we say that you have a sensor or you don't have a sensor or that we know something about where you live and that decision support tools can really sort of help support working through these conceptual differences as well. And then not that this I don't mean to sound unoptimistic but that even with effort even sort of working through the different trade-offs and being clear-eyed about how we make these choices our outcomes are unlikely to be perfect. If you look back at those slides with the total population and the young and old population you can see that you can choose one group maybe that you would like to have good coverage for but it might be tricky unless you have a sensor unless you have sort of ubiquitous sensor distribution which with time might be somewhere that we had but it's certainly not where we are now. You're going to have some groups that you're not covering. And that's it. Thank you. Thanks so much for your talk, Dr. Franklin. So I think we'll move on to the Q&A portion of the talk. So just everyone should I stop sharing or do I leave my slides up? Whatever you prefer. I can stop sharing then people can see my face a little bit better. So yeah, just everyone. Oops, I think I missed it. Anyone in the audience? To please just type your questions into the Q&A and then we'll go through those questions. So I think the first question comes from Stefan who asked might there be some surprising advantages of being unmapped? Certainly measuring and mapping the spatial distribution of who provides public services and goods or looking at factors like air quality or traffic can provide for better public knowledge. But are sensor deserts in some ways and with certain issues also potential spaces of liberation precisely because we do not know what is happening? Yes, but now I'm going to go back to I'm going to go back to my slides and I won't, let's see. This was sort of the point that I wanted to have. Oh, sorry, it's a sort of... Well, what I wanted to show you was a slide towards the beginning with the Venn diagram. Let me see if I can do this. So I won't do the whole slide show but this was the point that I was trying to get at here and I probably wasn't super coherent. But this idea that we tend... I think that this is actually how we tend to think about it that there are important benefits of not being measured and that certain groups get to benefit from that. So this is whether you can publicly drink or sit outside on your front porch or be loud in your backyard or play in the street. These are all things where some groups benefit from being able to just move through their space and their day without impedance. But I think that there are some forms of measurement and air quality sensors are a particular case of this where we risk the sort of spill over of the benefit of not being measured, sort of spilling over, I think, into measuring air quality where we know how important it is. And we also know that, you know, if you read anything from the environmental justice literature that there are just particular groups that don't get measured, right? And that maybe one argument that we can make is that we just didn't know but there's also deliberately not knowing so that you don't have to do anything about it, right? And I worry more with air quality about that case where because we haven't placed a sensor, we simply were agnostic to what your air quality might be in your neighborhood. And I think we can probably think of lots of other stuff that affects people, right? The quality of your sidewalks, whether or not you've got crosswalks, whether you can sort of roll, you know, whether you can use a wheelchair or a stroller unimpeded through your neighborhood, things that we might like to be able to measure about the neighborhood where it wouldn't be beneficial if we hadn't captured that for you. But yeah, I mean, it's an important point but that was sort of why I had this slide to sort of, I think assert this difference between the two types of surveillance. Thank you. Our next question comes from David who asked, do you see an opportunity or issues related to co-locating sensors with 5G deployment? And David, I don't know if you want to add on anything to that question. Yeah, tell me what I need to know about 5G deployment. David, are you still on the call or well, you might have. Well, unfortunately, I don't know anything really about 5G deployment except that it doesn't cause COVID. Probably not very helpful in this case, yeah. All right, yeah, why don't we move to the next question then. So the next question is from Wenfei who said, I have a similar question to David. Anthony Townsend talks about civic tech and smart technologies as buggy and brittle. That is, they're vulnerable to break or are prone especially over time and especially as technologies become obsolete. What are some of the long-term considerations about using these sensors over time and what do we know about the temporal consistencies of the sensor measurements? Okay, so the sensors as they exist are very troublesome to work with. So if you wanted to get frustrated, you go download all the data for the air quality sensors and then you see that they're sort of calibrated to themselves where they're not calibrated with each other, for example, right? And they break frequently. So they're not smart. I mean, this is like, if you were to imagine like the best way to capture air quality for an area, this probably wouldn't be the thing that you would invent. And this is not my area of expertise. So I'm happy to have someone weigh in. I think that this is an example where sort of the hype of smart cities really wins, right? It just sounds like a really cool way to measure these things. And we didn't have good models for air quality before, right? It seems to me that we're very quickly going to move to some new phase of measuring air quality. And again, I'm not an expert at this, but I think that satellites become more sensitive and the ways that we can model air quality from space or remotely become much more sensitive, probably more accurate, that's much better because what we want for air quality is a surface. And that's not what we get with sensors. What we get with sensors are point measurements and what we really want is a continuous surface. So I think very quickly we'll sort of either move to something more remote or the idea of sort of more mobile sensors, right? Which don't get us away from the problem of calibration and breaking easily, but are probably going to be much cheaper and can be deployed as people move through space, right? So you might, I think within next year is more cheaply and more broadly be able to measure people's exposure as they go about their daily business, which is really the question that we're sort of after, right? So that's probably what I expect to see happen. And in conversation with people at the Urban Observatory early on, when we said, look, the geography of the sensors is so strange. And they said, well, you know, they're really, really expensive and there are some parts of the city where if we put them up, they don't last, they're gone. And there are other parts of the city where we can put them up and they stay. And not only do they stay, but people want to come and do things with the sensor. They bring their kids to look at the sensor, they download the measurements. And so what you get is this sort of inequality that's reinforced just because of the cost of the technology. Yeah. All right, could I follow up? I can see Wenfei raising her hand. Hi. Hi. So I actually, I had another question, but it was talking more about kind of like spatial. So I guess my first question was kind of like talking about temporal consistency. And then the second question is more kind of about the spatial consistency. Yeah, so my sense is that kind of these sensors are a little bit inconsistent, but I wonder if I'm like a modeling perspective. How do you, do you kind of account for or study the kind of the inconsistencies or the uncertainties kind of across time or maybe kind of like spatially? Like do you kind of account for that in your models and or maybe if you were to, how would you go about doing that? So one thing that you might notice is that we don't actually use any air quality measurements. We only use sensor locations and not the information that the sensors are generating. And one of the first things that we had to do when the project started was to think about, well, how are we going to, the whole project is about sensor deserts and about coverage, right? And then I realized, well, we don't actually have a good way to measure coverage. And so in the paper that's just come out with Caitlin, we really just look at whether you have a sensor in your neighborhood, right? And like which neighborhoods get sensors and which ones don't. Now we're sort of looking at this sort of, you think there's sort of an optimization problem and how that might be helpful. But what we really thought would be interesting would just be to take an air quality model and look at the uncertainty levels. And then you could just look at the geography of high levels of uncertainty. And the statisticians said, yes, sounds really interesting, but the model's taking a long time to calibrate. So it's really hard to measure this stuff, right? But I think that would do, that is an avenue that we'll pursue. Cause to me it seems really, to me it seems almost intuitive that the real deserts are what the modelers can't capture well. Because obviously if a sensor is a point, most of, you can't, points don't take up any area. You take like, you know, you could never cover the entire city, but with a model you might be able to do better. So yeah, that probably doesn't entirely answer the spatial question, but it's an important one, which is that it's just really difficult to get good, reliable, high confidence measures for all parts of the city at a high spatial resolution. I'm not seeing any more questions in the chat, but maybe I'll follow up with one of my own questions if anyone else wants to take the time to type in their questions. So earlier in your talk, Rachel, you talked a little bit about how people sort themselves differently over space, and how there's sometimes this tension between where population density is and where collections of people that potentially could benefit, cultures of people with certain demographic characters, except potentially could benefit most from sensors or kind of not always located in the same area. So I was curious whether you thought there could be a case for, there potentially to be a random allocation of sensors across space. Yeah, I mean, you could, so another thing that we have thought about with the sort of supporting decisions would just be, especially maybe where the sensors are already located, like how much better is it than a random allocation in terms of coverage to just throw your sensors out there? Like if you are really deliberate and really careful and you weigh all of these different criteria and I tell you, hey, you're still gonna miss some groups, why not just throw darts at a board and see what happens and then assess sort of what that coverage would be. And a nice thing about doing that, I think, and again, I'm not a statistician, but the nice thing about doing that would be sort of, you could build up a distribution that gives you a sense of what your coverage tends to be for a given number of sensors, right? So if you're only going to invest in 60 sensors, what's the best you can possibly do if you just keep throwing 60 darts at the board and you look at how many people you're covering, then you sort of have a sense of how sensitive, given the way your city is structured and where people are located, what the built environment looks like, you have a sense of what that ceiling is of how good you could actually make things, because 100% of the population is probably too difficult to look for, right? And I think there's also, I fall back and in the end, we fell back on just what neighborhoods have sensors because I think that's actually a pretty good rule of thumb, actually, not if they're massive neighborhoods, but lots of the smart city projects have actually taken the approach of, we've got 16 wards, we'll start with 16 sensors, and once we've got 16 sensors, one in every ward, then we'll think about where to put additional sensors. And I tend to feel like that's probably a pretty good rule of thumb because then, if you're at a city council meeting and somebody wants to know what's going on, you can say, well, we've got a sensor here and this is what it's showing us, sort of somebody who doesn't go to city council meetings and is not a policymaker, this is what I would do if I were put in charge. Got it, thank you. So I guess I'll just, I'm not seeing any more questions. So unless anyone would like to weigh in with a final question, I think we'll wrap it up here. But on behalf of GSAP and the urban planning program, I want to thank you for the talk today, Rachel. It was really fascinating. I know I personally learned a lot more about sensors than I have known about before. So thank you so much for taking the time. We really appreciate it. Thank you for having me and thank you to everyone for having to watch me sort of talk and gesticulate. I wish that I could see your faces as I was talking and sort of know where I'm on track and off track, but thank you. Sure, appreciate it. Thank you so much, Rachel. My pleasure. Thank you for your time. Thank you, everyone.