One Ring to Surveil Them All: Hacking Amazon Ring to Map Neighborhood Surveillance
10:56PM Jul 26, 2020
Hello and welcome back to the hope 2020 conference, we have with us today delivering the One Ring to surveil all hacking Amazon ring to map neighborhood surveillance with Dan cellucci, who is a PhD student at the Media Lab at MIT in Boston, and he's here right with us. And we're gonna get right into your welcome. And we're going to go right into his video. All right, stay with us and we'll see for q&a right up. Hi,
I'm Dan falacci, and I'm a PhD student at the Media Lab at MIT. And for the past year or so, I've been looking at thinking about and collecting data on surveillance cameras that people have been installing on their front doors. And to start.
To give you an idea of why I think this is
a problem and sort of the breadth of the whole thing. I just want to ask everyone a few questions they're rhetorical so just keep them in your head. Do you live in a city in the United States, in particular, do you take regular walks around your neighborhood. And have you ever knocked on a friend's door, and notice one of these sort of weird rectangles with a camera on the top. And if the answer to really any of those questions is yes, and you live in the United States, then chances are your video has been taken, without your knowledge or consent by the fastest growing private surveillance network in America. This, you know, to give you an idea of the breath I'm going to zoom out a little bit and show you the route that I took to get to the train every morning before locked down. Now this is kind of boring at first, but when I walk on this route and I really pay attention and look for those rectangles. Here's how many of these cameras that I pass, and they're all owned by the same parent company, again, and they're all recording me on that walk because they're motion activated and their cameras right there sitting on the front porch of houses and not just recording a porch, they often record the entire street and. So, on this one mile route the longest stretch of time that I'm not being surveilled is only about a 10th of a mile. And this is a route that I walk at least two times a day when I'm commuting regularly. And to zoom out even further and give you a peek at what we've been able to do with data that we've collected from the platform. Here's a map of all the cameras that I've been able to detect in my metro area, Boston, so you can see this is sort of a wide problem. And so, obviously, this parent company here is, Amazon, who bought ring, the company that first popularized these devices in 2017. And, yeah, the same Amazon that has a horrible track record on selling face recognition to cops and the same Amazon that's also the wealthiest company in the world. Today I want to tell you about why the spread of these cameras is a quietly growing threat to civil liberties nationwide share how he was able to map out part of this growing surveillance network. And I want to touch on some of the ongoing problem projects that this mapping has enabled. So, first off the bat, I want to show you what ring in the neighbors app looks like so this whole thing is grounded in some reality. Some of you might already know all this so just bear with me a little bit. So this is the ring neighbors app for my neighborhood. When you sign up, you give them your home address, and it uses that to define a relevant area around your home. You can edit it, make it bigger, smaller etc. And then that area defines the kinds of posts that you see on your main feed, and that feed is really like any other news feed like Facebook or Twitter. But the posts here are automatic crime alerts posts from rings editorial team or pretty commonly posts from your neighbors with videos recorded from their cameras, or just text that they posted to the app. So this post that I saw yesterday is a great example of what ring seems to be often used for. It's a video taken from a doorbell literally down the street from my house. And you'll notice sort of initially that it's not capturing the person's doorstep, like I mentioned earlier, although rings policy, really clearly states that you have to orient your cameras to only record space in your property, most cameras capture wide swaths of public street, it's sort of impossible not to when you have a doorbell just facing straight out of your house.
The second thing I want to point out about this is that this is a post, not just about, it's about carjackers, but the posted video shows two kids hanging out on the street rollerblading and doing tricks in what really clearly seems to be their car. The poster is basically to speculating that them doing rollerblading tricks is a cover for them breaking into cars, but, you know, there's no theft captured on camera to me there's no suspicious activity happening. And the video here isn't high quality enough to strictly identify the kids, even with Amazon's recognition, but immediately because this is posted on the app, and this, because like I'll go on to explain in a moment. This is tightly integrated with existing police networks. These two kids could suddenly be suspects in what seems to be a neighborhood problem, even though they've just been recorded hanging out at night. Likely without even knowing it. So why is this a problem in the first place, this is you know this is a camera sitting on someone's front porch. And this is a privacy and security oriented audience so many of you might already be familiar with ring, how it operates and why it seems problematic. But for those who aren't really I want to spend a little time going over the basics of ring, how it operates. Why it really does seem problematic and highlight their really recent excellent journalistic coverage, over the past couple of years. That's really exposed ring as a police surveillance network that people are just inviting into their neighborhoods. So, in case you missed it, Amazon doesn't just market these cameras to regular civilians. I characterize the spread of ring as a private surveillance network. But, as is often the case with surveillance technologies in the United States. This isn't really quite right it doesn't capture the whole picture. So as of this week, ring has business partnerships with over 1300 police agencies in the United States and they really go all over. This is a map that ring has seemingly been keeping up to date detailing all the agencies that it has ongoing relationships with, and this number is absolutely growing every week because of ring has spent the past three years systematically introducing their surveillance network to law enforcement agencies across the country. They throw parties for police they offer discounts on ring devices, sometimes they just simply gift the cameras to departments on the order of thousands of dollars. For more information about that pattern the materials and what what those relationships look like. I really recommend you check out Carolyn Haskins excellent reporting series on that topic in motherboard. from reporting like Caroline's, and from advertising and press materials put up by ring itself. We've learned a little bit about what those relationships look like, and for civil liberties advocates it's really not good. The big sale that ring makes to police agencies, is the promise of using video recorded by citizens to make arrests track suspects, and monitor crime. They do this through a special police portal that allows them to see where all the cameras in their district are located request video from users and monitor other data streams like 311 calls or crime reporting. Sometimes ring systems are really tightly integrated with cops Computer Aided dispatch leads, and they frequently have sensitive information about 911 calls and other reports in some us counties, when you make a 911 call your address, your name, and sometimes details about your call all gets stored on rings servers. Now, they don't push that data out to their client app so normal people can't get access to it. But that doesn't mean that they're not storing it somewhere for safekeeping or for training data.
Because the main way of getting video from the ring network for cops is through the consent of camera owners ring also offers training to police officers, they've developed and shared really specific strategies and language for requesting video from citizens, with the goal of optimizing how often cops can get a request approved. And this includes, like I said, Really specifically what language to use when asking customers for video, and how to post and engage with their community on social media as multiple journalists have found police agencies, often use the specific language that ring shares taken all together, the scripts, the social media templates and content and the training that provides police departments looks suspiciously like ring both training police and how to do their jobs and unashamedly asking police departments to use their influence to market these devices, and the neighbor's platform to their constituents. There are Facebook posts and Instagram posts from cops that literally say on the image that's being shared, like buyer ring camera today. In terms of the actual videos, and cording to ring policy police can request video recorded from any device, up to 60 days back either if that video is stored on device or it's stored on a server somewhere, it's kind of hard to tell my change by the camera type. And if a camera owner refuses, that request, police can still subpoena the videos, same time timespan up to 60 days back. If cops have probable cause, and there's no warning or notice since the owner at all, if it's a peanut for Amazon. And this obviously isn't even to talk about the people recorded by these devices, who really clearly as the rollerblading example shows. Never consented to their video being recorded in the first place, let alone shared on a public platform or with police. sometimes those ring videos are used for advertisements themselves, again, with no consent for the person being recorded. And it's worth noting here and stressing that this 60 day policy is just that, a policy. There's no regulation on these devices, no oversight into how they're used or where they are and no transparency into how police are using them in their everyday work. Amazon could just as easily switch their policy next week to have all this video streamed directly into police fusion centers, and we might not even know about it. So, in 2019, to understand more about ring as a phenomenon, and to increase transparency and visibility into where these cameras are and where we're being surveilled day to day I decided to try and download the app. And just as a first pass, try and find all the cameras near me. And it became really clear really quickly from, you know, you can just think back to that video of me scrolling through the ring app that I can't just poke around and find my neighbor's cameras, they have to post. And when they do post the location isn't really clear it's sort of this big circle around where the camera might be. There are three main things that basically stopped me from finding all the cameras in my city just by using the app. One, there's space limits that ring puts on your neighborhood or area of interest when you sign up in the app it's limited to about five square miles. And that doesn't really help Boston is small, it doesn't really cover all of the area that I wanted to see into, you don't really see where all the cameras are when you use the app. Unlike cops who get an active camera map when they sign up to partner with ring like this one with all these dots normal users just see an approximate location of where alerts were posted, and it's usually in the form of like a circle it's maybe 100 meters in diameter. And three, you only see cameras that have posted on the app, not all the cameras that are recording video out in the wild, as I'll show a little bit later. There are plenty of people who use ring cameras that never posted the app at all. Now, I'm a computational social scientist by training not a security researcher I don't have a lot of experience breaking down apps or reverse engineering, but I do know how apps work, so I decided to run a man in the middle attack to see how rings app was communicating with the servers, thinking, maybe if I'm lucky. I'll be able to see all the alerts posted in my area and map them out a little bit. And right away I noticed that when you interact with rings API, most of your requests are associated with some assigned area and this area is just like a number and it seems random at first. And so all the requests are doing things like pulling all the recent posts for a specific area ID the API seems really simple. And at first I thought this area defined some kind of geo hash for different areas around the United States so one area here would be one number and then you increment want to go like over maybe a mile or be right next to each other, but the IDS didn't seem to correlate spatially like at all. Here's a map of a random selection of consecutive IDs, and you can see that they're really all over the place. I did quick statistical analysis and what I found is that the area IDs seemed truly totally random, even if two IDs, were right next to each other in space or clustered around an area. they could be really anywhere in that number space. So my first attempt at scraping was literally to iterate through every identifier starting from one that I knew and just adding one every time. And do API calls on that ID. And basically, to pull down data from the API I just pretended that I was a user scrolling through that infinite feed and the neighbors app features that infinite scroll and apparently there's no rate limiting because I did this on 16 cores on a server repeatedly for every area that I could find. And in the end, I had data going back to 2016 for 236,000 areas. And that only took like maybe seven hours, or something. So, and this time ring really didn't have any security in their app at all. And in the end, you know I didn't understand how their authorization worked for accessing different areas. So this was a problem is that some area IDs were accessible to me just by changing the ID and using my existing authorization it some weren't I would get errors, and it seemed again really random and I didn't understand why. Plus, after collecting all the data and putting it on a map and looking through it. It became pretty clear that this wasn't going to be a complete scrape like I had hoped, and that it was really really inefficient. I had no way of getting posts from spaces between areas, like here, this is San Francisco and around Berkeley, and the area ideas right are completely different. And if I wanted to get Oakland, for example, I have no idea what Id to use in my API call. So all I can do is just keep going through them and pray that I'm going to get something that looks like a complete map for an area. Plus, sometimes areas would be really close to each other.
And that would create a lot of duplicate data, and just really fill up my hard drive super fast. So then I remembered that as a user, I defined that home area of interest. And what if I could just change my home area. Now, ring says that on signup they validate home addresses and limit how often you can change it. But it turns out that either this is only enforced on the client side are not really enforced at all. I've changed my home address a ton using the app and have never really seen an issue. And, in fact, a lot of things including that are only enforced on the client side, if at all. Not only is the home location limitation. Only enforced there, but the size of your home area is too. If you remember from just sort of peeking at the app earlier the home area that you can define is limited to like five square miles. So here's the limit of an area that you can set as your home in Florida in the app. And here's the limit via the API, the API, gives you a maximum limit of 800 square miles versus the five ish through the app. So I just started making grids, with a buffer around the borders of the continental US to make sure I didn't miss anything. After testing that I could programmatically change my home area and this square repeatedly and quickly. I just started going through every grid that I defined and pulling the alerts in that same infinite scroll away. So now I had 16 fingers scrolling through the infinite feed for every one of these grids across the US. There's about 8000 of these boxes, and this gave me what I think is a pretty complete view of all of the alerts that have been posted to the ring app since 2016. So, what does that even look like. Because we have every alert posted since the dawn of ring time, I can actually just show you. So here's a time lapse video starting in November 2016 and ending in February 2020. Every pink dot represents one alert posted to the app by a user, which then just fades to black sees these black clouds sort of spreading over urban and suburban areas around the US. It's a lot. Um, but, you know, it's not everything. So it's really important to note here before I go into other work and projects that we've done with this stuff that the data set, we're using, it doesn't cover every camera taking live video, if you remember you know the way we scraped it is just looking at alerts. So really it only covers cameras that have posted to the ring neighbors app. And there's, I think a big difference between these two. Now it's hard to know like how many people have just purchased a ring camera or using them outright.
But our data set shows a peak of about 1800 alerts on a single busy day in February 2020, but ring shared their usage data in a blog post during Halloween last year they have this whole PR campaign around Halloween because, well, it's basically the doorbells favorite holiday. Anyway, they reported almost 15 million doorbell rings on ring devices in that day alone. 15 million, with over 5 million rings, but just between seven and 8pm. Now in all this scraping if we go through and find unique locations and user IDs, we found about 530,000 cameras across the United States. If that was really the total extent of rain cameras, then each of those cameras would have been wrong three times on Halloween. A typical day ring says that maybe there's 2.5 million rings, which would mean about five rings on average for each household on a typical day. Now I couldn't find great data about the average number of rings in normal homes, but unless ring users get, like, five Amazon packages a day, which maybe isn't such a stretch, we're capturing probably like under 20% of the total number of ring cameras out in the wild. And that's if the average is about one ring per day or below, which seems about right. In my experience, and that means that what we're measuring the data set is usage of the ring neighbors platform. But, you know, still, basically the most available subset of ring cameras that exist out there in the wild. Now, there's some basic questions to be asked about ring as a platform, and its relationship to crime and demography. And I've taken a kind of first stab at answering some of them here using a spatial analysis of ring cameras, using this data set that I've created. So first, I just wanted to share some descriptive statistics to give you a feel for what's out there. Like I mentioned, we've detected about 530,000 cameras across the United States, but some states were earlier adopters than others. California was a super early adopter hitting 1000 alerts only a couple of months after its first post in November 2016 and pretty quickly after Florida and Nevada saw critical massive use other big cities, states like New York, Washington and Illinois followed, and then came a steady spread across the United States through 2018 to 2019. And this adoption sort of shows the counties with the highest rates of alert postings. California is the bigger biggest offender here by far with LA and surrounding areas, basically leading the races and the number of alerts, per 1000 people. And the number of alerts that gets posted every day tracks the adoption plot that you saw before, with the kind of steady state, getting reached by mid 2019. And this is somewhere comfortably around 1500 alerts from users every day. One of my main motivating questions after scraping all of this data was to start measuring whether there are racial patterns or discrepancies in how ring is used policing platforms like ring are often discussed as mechanisms that enforce racial and economic norms and neighborhoods and rings serve invites users to label people they see for their cameras as suspicious and share their images, and another lingering question is about rings relationship to package theft, as a commercial product sold by Amazon, the world's largest online retailer ring is marketed essentially as a package of return. So the question is What relationship does ring use have to package the reporting and rates across the US. So to answer these questions. I aggregated all of this ring spatial data to the county level, and I made variables for each county that represent the number of brain cameras per capita, most counties have a pretty low rate of reading usage according to our data set, but in some densely populated areas, a rate of five or 10 per thousand people on this graph can mean one camera surveilling every street in a census block, and to measure demography I took census level population estimates from the 2012 to 2015 us American Community Survey, and to measure crime, which is always sort of difficult I used FBI crime reporting statistics at the county level. Now, just a reminder about crime data, generally, whenever you're using crime data, what you're measuring is reporting rates, and really rates of policing certain kinds of crime, not actual behavior of people. So that's just a caveat for really all of this. And I don't think that I'm going to bore this audience with the technical details of using statistics to model what's actually a pretty complicated spatial process, because everything is sort of all muddled together because neighboring counties affect each other. So I'll just cut to the chase and show you the results and talk about the model, the different variables I made are on the left on this plot and the x axis shows the impact of those variables, which basically just measures how important they are to predicting the number of alerts per capita in every county. Surprisingly, to me at least, no racial variable showed a really significant impact. And that sort of suggests that at the county level racial demography isn't really driving ring usage wider counties, for example, aren't more likely to post unring than minority dense counties. And we also see that theft, which includes package theft doesn't show a significant impact, either. So places with high package left aren't really using ring, as a platform. Instead, some of the most predictive variables for ring usage seem to be the income of a neighborhood and its rate of homeownership controlling for everything else. This makes a lot of sense to me when you consider who can even install these cameras in the first place. Mostly, you know, people with disposable income who can buy them, and people who own their own homes who can install them and where it makes sense because although this abomination exists near my house. Most apartment building owners or renters aren't using ring as their surveillance or smart home tool of choice. And the strongest relationship here going back to the graph is median home value, which has an enormous negative impact on the number of cameras used in a county. I've been taking this to mean mainly that pretty property rich areas have other forms of Home Security places with really wealthy homes, don't only need to buy a ring product, which is marketed as an affordable version of other more expensive systems, and it does turn out looking at some of the other variables here that violent crime rates, not including robbery, have an impact on the number of ring cameras and accounting, but a huge increase in violent crime corresponds to a 12% decrease in the number of cameras present places that experienced the most violent crime, don't really use the ring neighbors app very much. This, combined with the strong impact of vehicle theft and robbery rates suggest to me that highly visible property centered crimes are part of the equation here. Not package theft or even necessarily personal safety, and we can look more closely at the alerts themselves to get a better picture. So, these are the different types of posts that are on the ring platform I've sort of collapsed them into four main categories crime safety strangers and suspicious activity. And you can see basically that the stranger in suspicious categories together are almost twice as common as crime post specifically, and looking at the language, this sort of reveals what these different categories are actually about. So if we analyze the text, we can get a sense for what they're used for. Here are some different bi grams, which are basically just unique combinations of two words, use on the platform. And this is ordered by their log odds ratios, which basically just means that words that are more unique to a category appear higher up on the list. So try to is really unique to the suspicious category and broken into is really unique to the crime category. The crime and suspicious categories seem to be mainly about people stealing from cars on the street with crime appearing to talk, basically explicitly about like a theft that gets seen and suspicious sort of speculating about people who might break into a car, trying to versus actually stole my for example, and the stranger category is really about what it kind of sounds like on first blush unknown people being recorded and safety seems mainly about hearing gunshots in the neighborhood. And so all of this together with the results from earlier, kind of suggests to me that ring is used overwhelmingly to talk about and record people's cars getting broken into or people thinking that their cars might get broken into the video that I showed at the beginning of those two kids rollerblading around the street is a pretty representative example of the kinds of posts that I've seen on the platform. Now, I also suspect, though, like lots of people that ring is a filing and gatekeeping and this seems pretty apparent when you see the kinds of people who are recorded on these videos. And that people usually use this sort of ambiguous crime like oh they looks like they were casing cars as a rationale for racial profiling, but so far I haven't found a great way of finding whether or not that's the case. Now, I do have a lot of the videos downloaded. And I started to analyze those but those results aren't really ready for primetime yet so I'm not going to talk about them here. But I want to take a step back and bring us back to the big picture of this platform as a growing unregulated and really opaque surveillance system that's appearing in our cities. So, for people who find value in buying these cameras. The big question to me is whether monitoring and talking about like your car is worth, creating this enormous surveillance network that is almost impossible for everyday people to avoid. One of my original questions if you remember when I started collecting the data was just where I would be surveilled by these cameras in my own city or my own neighborhood. So in a pretty rough new project that I want to share, we've started to sort of answer this by just mapping out streets and routes and cities, showing which streets are more surveilled than others. Here's an example this is Minneapolis, you can see that certain streets are pretty strong hotspots of camera usage, the more red a street is the more cameras that we've seen on that street. Now, I just want you to imagine going from the north side of the city to the south side or the east side to the west side without ever using one of those red streets. It turns out that that seems pretty hard.
And it's not so hard to actually measure using modern routing algorithms. Here's a route from Boston, my home neighborhood, the original sort of motivation for this whole project, looking at walking from the south side of the city, up to City Hall. Now this first image is a normal route that basically just accounts for distance is trying to find the shortest route possible in the city and it's pretty linear. But if I plug in to the routing algorithm and attempt to try to avoid all the cameras that we've seen the route gets pretty complex pretty fast. And this pattern, basically exists, you know, in every city that I've thought of to measure. And so part of this project, ongoing is to try and quantify how much you'd have to change your daily routine, or change your routes in order to avoid surveillance in cities across the US.
So those are some of the projects that are
ongoing that I've been working on either by myself, or with a few other really talented people, the data itself is not publicly available unfortunately because there's a lot of personally identifiable information in there, you know, ranging from the videos that gets shared of people who are totally unaware as to other PII that's sort of sitting inside of the data, but I have been sharing it sort of selectively with people who I think are doing important work or whose work is directly relevant so if you think you're one of those people or you think that you have a project idea, please just reach out to me, and we can talk about it, happy to do that. And before we end, and I just want to leave on a few notes of stuff I've been thinking about for the past year from talking to journalists activists community leaders about ring the platform and its spread. The first thing that's in my head all the time, is that it really shouldn't be this hard to get transparent data about where people are being surveilled in their own cities, especially when that data is used all the time by public officials, even if it's only shared in the most aggregate level like the number of cameras in a census block that should be something that is shared and available to the public. Daily. The next thing is that, you know, I think there's really a worthwhile distinction here between people who are recording their front porches with cameras that just you know record locally or to their local home network or whatever it is, and this system of millions of cameras pouring data and video streams into this big private wealth a powerful company. The biggest defense that I see from people on the internet, defending their use of ring cameras in response to critical articles about the whole system is that they should really have the right to protect their property or recorded or whatever and then it's useful for them. And like yeah sure I'm sure it's useful for you. But the problem is that ring isn't for you it's not a product made for you it's a product made for Amazon and for police to collect this huge trove of data. There's a difference between you buying a product to record yourself in your home and you participating and being complicit in this enormous surveillance network. The next thing that I want to say is just, you know, the neighbors. A lot of people have made this point, the neighbors app, only really works if residents view strangers and everything outside their home as a threat. And people are worried that this is kind of self reinforcing the more we see these cameras and platforms, being used, the more maybe the social fabric and trust in our neighbors and place sort of disintegrates, and that's pretty scary. And finally, I'd be remiss if I didn't mention the current incredible movement against police brutality in general modern policing in the US and in a time when we're questioning is becoming more popular to question the purpose of policing abolition is becoming more popular. And in places like Minneapolis, where we're seeing total restructurings of police in general. It feels super ominous to me to see private entities like Amazon internalize or operationalize, our current ideas of policing through things like rent. You know my biggest worry is that long after we see restructuring of policing which I hope we do. People will still be policing each other using the same kinds of patterns and thoughts, because of platforms like rain. And that is not really a future I want to live in. I hope that this project can serve as like a jumping off point for people who are interested in measuring community self surveillance. I hope that it's helpful in thinking about ring as a platform, and other forms of surveillance that look like this, of people participating in this big big network. And I'm hoping that this sort of starts off some really interesting collaborations and projects with folks. Thanks.
Rely okay we're lucky. Thank you. So Dan. Dan Colucci is with us now to answer your questions. Say Hi Dan.
Hey everyone. Yeah. All right,
going to go right into some questions. It was very busy in the matrix chat. So the first question we're going to do. Did you ever look into hacking the ring, so that it could not send data to Amazon and maybe send it to your own domain instead.
So, no not really is the answer to that, I my analysis in the project is sort of focused on the data that's being collected by Amazon and the scope and breadth of this stuff rather than hacking the device itself, although that sounds like a pretty useful project if it's made easy enough for existing owners.
Sure. All right, very good. We'll go on to the next question then given Amazon's work with recognition. Any word on whether they plan on integrating this with ring at all. Outside of Illinois, or other or other jurisdiction, with legal facial recognition and restrictions.
Yeah, I can't say off the top of my head.
I can say with certainty that Yeah, they've discussed it for sure. In patents, and in certain releases I think they've mentioned using facial recognition, and they already do some right so like a lot of these cameras have person detection already so they already do some object recognition on the feeds when they're uploaded to the cloud. But, you know, the big risk is sort of like the integration of Clearview AI with ring. And, and that kind of that kind of system, I think is somewhere on the horizon, hopefully with movements like we're seeing now against facial recognition tech and at least like the year long hiatus I recognition being used with cops, we won't see it, but yeah, it's there's news about it. For sure.
That's it's really a dynamic fluid situation right now. All right, the next question that we have is what statistical model did you use for your regression analysis.
Good question. Um, I use something called a spatial autoregressive lag model. It's basically like a linear regression, but it takes into account the fact that all these counties that are next to one another. They'll influence one another. So, you know, if I, all the the median income and county a sort of correlates with a median income and counties all around it right and this is true for lots of demography lots of different demographic variables, and also actually appears to be true for ring camera uses counties that are closer to other counties with ring, are more likely to have more people using ring so normal models, can't capture that interaction very well, so you need to use some tricky statistical methods to suss out those estimates that I showed in the kind of unbiased way.
Right. The. is ring only working in the US, or did you only focus on the US usage.
Yeah, so I you know I just saw an article about this the other day I'm trying to remember what it was, I can literally say from looking at the data that there are cameras, outside of the United States, when I did the first attempt I did it scraping where I had these random area IDs. Some of them were in the UK. So there are people in Europe, that are the UK that have these cameras, but they're definitely way more in depth the United States and ring is really focused on the United States, as far as I know, although their development team is in the Ukraine. Most of their product bases, this year. And, and it is where I just chose to focus because that's where the highest density is.
All right, The next question that we have is, are you thinking of constructing any summary measures for major metropolitan areas on how much ring usage, there is for example, major cities.
Yeah, and I haven't framed it that way, but I think that what we're doing with the routing is part of that. I think the challenge you know as for advocacy is, you know, if you just say the number of cameras or the density. It doesn't always mean a lot to people, or they don't really see how it might affect them. For me, the question of like making a summary measure for a city is really about how do you make a visualization, or some measure that gets people to imagine futures where this actually affects them. And so the routing project is part of that right where you can say there's enough ring cameras in the city where you actually have to change your behavior pretty significantly, just to not be recorded. And that I think combined with some summary statistics for city by city and trying to share that out, is part of the goal of the project.
Right. Very, very good work. All right, next question is,
do you think running the videos, through some sort of machine language tool or machine learning tool, excuse me, would get useful data, and there, or would there be a relationship between that the categories and big rooms.
Yeah. This is a great question, and
the answer is that is yes, and so I do have the full like video data set, sitting on a secure server and and have done a bunch of stuff to like extract faces from it to try and see the average face, not to expose people but to try and measure, you know, who are the types of people that are seen as suspicious Andre by using machine learning methods. Another thing that I just started thinking about but haven't done yet is, you know, if I trained a generative algorithm to create the text descriptions of the videos right so you remember that rollerblading video had that description of these two suspicious kids using this thing as a cover for stealing cars. That's what that scene looks like through the eye of ring as a system. Right. And so the, I think one of the questions here is would it be interesting to make a machine learning algorithm that essentially did the same thing, I could feed it a video and it would produce a description, as seen through the eyes of Ray. And I think stuff like that is super interesting. And I'm hoping to do more of it. Depending how much time I have. If you're interested in doing that whoever wrote that question, reach out. Yeah.
Very good. We've got about another eight minutes worth of questions. I'm going to find another one for you, the follow up that last question goes with, or not even machine learning, but just the situations themselves which you spoke about. Is it more likely that kids playing will get flagged as barbaric raking then people walking, you know is that the natural extension, as they say with this type of description to the video that you're just describing.
If I understand the question right i think i think generally Yes, and the question if you could make something that does that or is the question like just, you know, what's more likely,
you know like what's more likely because you know the example that you used was really good with the kids rollerblading. So, and with machine learning, you know, is there a chance that you know in the future. Kids rollerblading will be understood by machine learning to be, you know, suspicious of say, I mean
yeah if Amazon's doing sort of what I expect they might be doing with these, like if you're Amazon and you're building this product and you have, you know, terabytes and terabytes of this cloud video with these text descriptions, like I mentioned before for us it's interesting, the idea of like generating that text from an image is interesting just to like understand how, you know, this whole system views these different videos, but for Amazon it's sort of it's interesting in this whole other mean sinister kind of way where it's trying to generate automatic descriptions of surveillance footage. And I think that if you're using this as training data then yeah, absolutely you'd see, especially like kids hanging out in the street and public, you know, going through a random sampling of these videos, um, the majority of sort of the suspicious or like even criminal posts are primarily young people. And a lot of times young people of color who you know viewing it don't necessarily seem to be doing anything suspicious or wrong besides being captured on camera, you know, and the person happened to like see it when they were scrolling through their personal reading feed or whatever. So I think that, you know that they'd have trained on this whole system, and an algorithm chancel system would absolutely
produce that kind of an output where it sees kids rollerblading is associated with stealing cars.
Yeah, it does make sense like that. Alright let's find our next question.
Are you considering doing, uh, are you considering doing an academic paper on this.
Yeah, totally. And, yeah, just the statistical analysis was sort of just presented at A. International Conference on computational social science which was last week. And, and I'm poking around at like journals and other conferences to submit the more like academic side of this work. Yeah,
absolutely. Very nice. The second part of the question is, how much money does Amazon make from police departments, subscribing to the ring and network.
You know, just like everything else related to policing. data is extremely scarce, almost impossible to come by. I doubt, even if you did a FOIA request on the budget around this kind of technology for police that you would even get a response back, although FOIA requests have been useful in getting materials for training and like the emails forging relationships.
I mean, maybe we could find out, but I don't know.
Okay, our next question is, are you aware if ring data is being correlated with Alexa, Alexa data.
I'm not. But this is something that I don't really mention in the video, which is that, you know, think about all the people who have these cameras. Most of them are probably also people who have a Lexus or somehow otherwise in the Amazon ecosystem. You probably don't have a ring camera but don't have amazon prime. And so, inside of Amazon, like I can't imagine that it's not right, like these are two very tempting data sources of behavior, about people that Amazon has inside of the home. And they're all collected by the same Corporation, and they're trying to predict how people will be ordering packages what they're interested in what they'll be buying etc. I can't imagine that they're not doing that to build some kind of behavioral profiles.
So yeah, probably, it's my answer.
Yeah, I would, I would believe I would agree with you on that one. All right. Our next question is, if the technology was near sighted, would that change things, like for example, can you change the optics on the camera. And I know some security systems allow you to block out areas that you don't want to be monitored. What's riding on that.
I mean I think it would change some things right, obviously one of the big critiques, and one of the big critiques I sort of levy against the platform is the fact that it in effect monitors public space, all over these cities instead of just porches. But in addition, there are some other bits and pieces to how ring operates that I didn't touch on in the video that I don't think that would fix, really, like, one is worker surveillance, and I think one way that Amazon probably uses these devices and also to surveil their workers who are doing deliveries, and whether or not that's something that we're okay with is a separate question then you know Is it fine if it's just recording my porch. So I think there's a lot of questions about still what is just reporting your porch, giving to a corporation when it's centralized, like that. It might even be you know all the people who ever run your doorbell. Right, closer together into a behavioral profile for you, that okay with you, etc. So,
Exactly. All right, we're about out of time. I want to thank you very much on behalf of all the attendees here and the staff and all the volunteers, really appreciate your, your great work. And it was very interesting. And, you know, we hope that we'll see you again at home.
Thanks. Yeah, this has been great, thanks for the opportunity to share my work.
Okay. All right. And now, please stick around, we're gonna have our next video coming up in just about 10 minutes. See you soon.