The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement

How much is the Cato Daily Podcast worth to you? We certainly enjoy putting it together for you and we know from all the positive feedback that it's an important part of many of our listeners' days. If you value our distinctly libertarian perspective, I hope you'll consider joining our new podcast sponsor program. If you visit kato.org/podcastsponsor, you can learn about various levels of support and the benefits you'll enjoy as a Cato sponsor. For example, if you become a patron sponsor by giving $1,000 or more, I'll personally thank you on the podcast and you'll get the regular benefits of being a Cato patron as well. If you prefer, you can donate in a friend or family member's name as well. It's the perfect gift for someone who values liberty but has everything else. Learn more about the benefits of becoming a podcast sponsor at kato.org/podcastsponsor and as always, thank you for listening. (upbeat music) - This is the Cato Daily Podcast for Thursday, December 14th, 2017. I'm Caleb Brown. Some police departments have begun using algorithms to tell them more about the areas they patrol and the people they are charged with protecting and more departments are considering it. Andrew Ferguson is author of the rise of big data policing, surveillance, race, and the future of law enforcement. We spoke at the Cato Institute's Spicon yesterday. - How has technology changed how police do their jobs? - So technology right now is changing where police go in patrol, who they're targeting, how they're investigating, and how they start dealing with new patterns and problems of crime. So it's changing every way police do their jobs, it's changing the type of police officers we might need and it's certainly changing the citizens' relationship with the police. - Give me an example of how that actually functions in a day-to-day setting for police departments. - Right, so right now, if you're a patrol officer in a city like Chicago or Philadelphia or LA, you are given a predictive forecast of crime in a particular area. It might be a printout on a piece of paper, it might literally be part of the computer model in your squad car, so when you go from location to location, it will change color depending on the forecast for crimes. You might be driving through and see it change from a high burglar area to a high aggravated theft area. And that obviously will distort where police go, it literally changes where you might be at a physical time. It also changes how police think about the environment. You just gone into a particular place told to be on alert for burglary. So suddenly everyone in your mind's eye is going to be a potential burglar. Why? Because the computer told you. In Chicago right now, they have a person-focused targeting, and algorithm determines the people most at risk for violence. He is a perpetrator or a victim. And that algorithmic list creates a number, a raw number. So anyone who's been arrested for the last four years in Chicago has a threat score, one to 500. And that threat score pops up on the computer of a police officer when they stop you on a car and they can see your name 500 plus. They can change how you would deal. If you're that officer and you know the individual in front of you is at risk, or more at risk of violence, might be more violent. It's going to change how you deal with that person as an individual. And so police are starting to get new information about the areas they patrol, about the people they are patrolling and interacting with. And it's just changing how they relate to them. And this is being done by data collection, finding the crime statistics there and crunching them in models to be able to particular areas. It's trying to figure out who among all the people in a city might be more at risk for violence and using an algorithm to try to determine a rank-order list of who is the most risky and all the time maybe because of technology. The way you describe it, it sounds like that's a problem. But to the extent that police are going to areas that we have solid information, are areas where crime can be expected. I mean, are these heat maps to the extent that they're generated? Are they inaccurate? Are they wrong? Do they perform well in terms of telling police where to go? We don't really know the answer to that. Obviously, the theory behind predictive policing be it either place-based, predictive policing or person-based is based on fairly good ideas, right? Not all places have the same level of crime. In fact, some crimes, if you go back to the original predictive policing models, they're based on property crimes, things like Burglars Carthes' Thestromoto. And why? Because those types of crimes are kind of viral. They're kind of contagious. If your house is burglarized, it is more likely that your neighbor's houses are going to be burglarized. Why? Because the Burglars understand there's some environmental vulnerability there. Maybe people aren't around. Maybe there's no police presence. And so they've shown through social science that those types of crimes are actually contagious, like that there's actually a viral sense. And sometimes it's just like the same guys coming back to rob the next house, because I realized I got away with it before. Sometimes it's their word gets out in the street. So there's a theory behind it. There's a theory behind the idea of particular places have environmental vulnerabilities that it makes sense to send the cop car there to deter it. That's the theory of place base. Same with person base, right? Not everyone is at risk of violence. They're in Chicago. The people who are more likely to get shot are involved in either activities in primarily poor areas. Some are gang related. Some are retaliatory. A lot of times in Chicago, one of the sad parts is because certain communities don't trust the police. If their friend gets shot, they take it on themselves to do the responsive shooting. So going to the police to get someone arrested, they do it. And so there are these cycles of violence. If you can figure out, OK, who are the people involved? And can we intervene? Can we stop? Can we get into those people's lives before the violence? It makes sense to target them. So this isn't some mystery black box that's just picking things out. There are theories behind it. The hard part is we just don't know if it works. And every time we've studied it, it has been more complicated than we'd like. We haven't had a study that said this perfectly works. There have been some studies that have said, it might work better than what we currently have. And some that said it doesn't work at all, but maybe we'll change the model to make it work. Based on what you've said so far, it seems that there are a couple of red flags that jump out at me. One is, how does that change the relationship that police have with people that they meet on the street in areas where they expect there to be something going wrong or an area where they don't expect something to be going wrong? And the other is, when you assign somebody a threat score, you've created a suspicion where there might otherwise be none. Right. So there's a practical effect. If you're the officer and you're patrolling this area that is particularly crime, you're going to be on the lookout for it. It's going to change the suspicion of the area. You're really going to be looking for burglars. But normally, if you see a high burglar area, like a guy standing on the corner with a bag, it's not enough to stop him. Right. It's a guy with a bag. Everyone has bags. But what if you know that this is a high suspected forecast burglar area? And here's a guy standing with a bag. Burglar's need bags. Do you now have a reason to stop him? Has this changed what you see and how you see it? And one of the people can run his records and say, hey, he has a high threat score. In fact, he's one of the people that computer predicted to be the most out of risk. And I have two facts that come from this background information, this big data information that might change how you view this individual. And notice, the guy has a slightly different. He's still standing with a bag. And so this background information is going to change how police see it. It may distort the Fourth Amendment and the protections we have about what police have to show in order to stop individuals on the street where they need reasonable suspicion or probable causes that distort this standard. And it obviously changes how a police officer might feel about this individual. I really think that one of the dangers of this threat score is that it's going to encourage officers to use more force than they need to because they are going to be more frightened because of what the algorithm says. Now, maybe that is smart for the officer, right? The officer suddenly has more information. They're better equipped to deal with this person and what they know about it. But that, if that's true, you better be sure that your algorithm is correct. And currently, the algorithm in Chicago, what it's doing, it's predicting both people who might be more at risk of violence, being the victim, as well as a perpetrator. So you might be dealing with this person, thinking they're more dangerous. But really, they're the likely victim of the crime, not the perpetrator. And that to me is distorting and inaccurate. When police departments and mayors of cities receive a pitch to do this, to use big data, to try to assess where the high crime areas are, make an assessment of who the people who are most at risk of being victims or perpetrators of violence. What do they hear? You've said some of it, but I'm assuming that they're hearing, you're going to have a much more efficient use of your police resources. You're going to go at crime, right at its core, where it's actually occurring. You're going to save money, the voters will love it, and you're going to clean up neighborhoods. Is that basically what they hear? You know, it depends on the technology. It depends on the caution of the people selling that technology. But yeah, I mean, that is what it is, right? Chiefs of police are tasked with an impossible job. They have to come up with the unanswerable question of what are you going to do to reduce crime? And the real answers to that are, hey, we need better schools. We need more economic development. We need more hope in our communities, right? They don't get funded for that. But instead, if they say, look, we have a new technology. We can do more with less. We can do it in a better way. We can be more efficient. That sounds great for everyone. Even if it doesn't work, it's an answer. And it's a good answer for of the chiefs who want to go forward, right? And that's part of what's going on here, is that these technologies are being pushed by companies, by vendors who have an interest in selling it, the chiefs themselves are not always equipped to evaluate whether it works or not, and sometimes the technology is too new to actually be able to evaluate it. And so we're adopting these technologies, sometimes through federal grants, sometimes through other kinds of extra money services to sort of pilot this new test, we're not thinking ahead of time about the consequences on individual liberty, we're not thinking about how it's impacting communities, and we're just not asking the hard questions. That's not to say that these technologies are bad, some are good, some are bad, some are just undecided, but we haven't had that conversation about how we can check these, audit these, make sure these new ideas, these new big, that ideas are really in everyone's interest, and whether we have the controls on them to make it worthwhile. And I also think that chiefs of police would welcome that conversation, because they are like everyone else, sort of behind the ball in not being data scientists, not being computer technologists, not necessarily even being lawyers, and yet are asked to make calls that affect all of those things. - What's driving it in terms of the crime problems or problems with laws on the books, what's driving police departments to say we need to move into this direction? - I think it's actually reversed. I'm not sure the police departments themselves are asking we need this technology, it's the technologies that are being created, that are pushed to the police, that you need this technology. You can do more with less. You can have an answer to the mayor when they complain about the fact that crime is rising. And many ways, because of that sort of pattern, there hasn't been the usual checks. We don't have community meetings about what technology should we use, where they work, and whether it makes sense that we're doing. So we haven't done that. We sort of done it in the back way. - It seems that to the extent that there a crime has occurred, and you are looking for suspects that you're sort of turning probable cause on its head, or you're turning a reasonable suspicion on its head, because people who are in neighborhoods are certain neighborhoods are more likely to be both victims and committers of crime, but proximity doesn't make you more likely to be guilty necessarily. - Right, so there are two things, right? So in the predictive world, and we've only really been talking about predictive policing and the different variations of it, when there is no crime, and police are sort of looking for their suspects, a lot of times these predictive lists become sort of the shortcut that's like, I call it the virtual most wanted list, right? You sort of have an already made list of, well, we know there's been a shooting, we have this list of people, let's go check on them to see if they know anything about it, right? And so the list has sort of created your suspect list in a way ahead of time. But there's a whole other aspect of this world of big data policing that's incredibly valuable for investigators, right? A crime has been committed, you wanna know who did it, right? What are the digital trails that exist, the cell phone records, the bank records, the video cameras that are out there? In a new world where we are building out new big data surveillance technologies to be able to watch whole areas of the city, being able to see who's coming and going, automated license plate readers, those little digital clues are incredibly valuable for law enforcement to investigate crimes that have already occurred. And that's also distorting what police do, right? It used to be that officers were sort of, there's just too much information, you couldn't necessarily trust the witnesses, we just didn't know, so much of criminal laws about trying to fill in the gaps of what people don't know. But with the digital trails that are literally tracking, everyone has a little spy on their pocket called a cell phone. Everyone has a smartphone, everyone has digital trails of where their license plates are being read by these automated license plate readers that are reading millions of license plates a day. Surveillance cameras, private and public are capturing what's going on in public, right? And all these little clues are also changing how police think about investigating, right? Investigating crimes that happen, and also who might be associated with a particular crime that happened. - There is a Supreme Court case and correct me if I'm getting the name wrong, Carpenter. - Yep, up this term, yes. - How might that case change? How police do their jobs? - So Carpenter is a case about whether the government needs a probable cause warrant to get third party records from, in that case, cell phone providers. And the government had requested 127 days of cell phone location records, meaning that if you're using your cell phone, if you're not using your cell phone, your phone is picking off different towers. It's basically able to determine where your location is. And Timothy Carpenter, who's an individual involved in the case, was allegedly robbing ironically enough, cell phone stores, with a bunch of compatriots, and the police in order to figure out was he at the place at the right time, asked for his records, right? They didn't use a Fourth Amendment, probable cause warrant. They used a storage communications act warrant. And the question was, did they need this higher level probable cause to get it? The reason why it's such a big deal is that the decision on Carpenter and those third party records probably will carry over to all sorts of third party records, which means the things that you are Googling, that Google has on you, the information from your Fitbit, and you're in the world of new medical devices, you'll have smart heart stents that are telling your doctor about how you're doing. All those are third party records, right? All of the, everything you do in your smart home, in your smart car, in your smart life, are third party records owned by and controlled by you and the company. And so if you don't need a warrant to get that information, it means police can have a much greater tool to figure out the things you're doing that third parties know about it. And since almost everything we do in a digital age is mediated by some third party, it means that it makes it much easier for police to go investigate crimes and develop patterns and figure out information about what each of us are doing. It seems like police agencies be they state and local or intelligence agencies at the federal level. It's almost this affectation that drives them to get more data. And we want more data, we'll be better at solving problems with more data. How often is that really true? - Well, you know, it's difficult, right? So it takes like automated license plate readers, right? So automated license plate readers in a state and like the state of Maryland are collecting literally millions upon millions upon millions of license plates. Individually, you may not care that your license plate at one point was captured, but the patterns can be developed and created so that people can track where you're normal patterns of life. They're only get used maybe in the hundreds, out of those millions and millions and millions of things, in the hundreds in any year, right? And so state and Maryland has to audit it. I think they used it maybe 250 odd times in court, right? So you've collected 250 million records and you use it 250 times. Is that worth it? - Well, you know, if the time they used it was defined like you're missing child, yeah, it feels like it's worth it, right? But in the larger sense of the balance between the liberty of everyone being sort of exposed and the fact we're not using it very often, how do we know? And that's part of the difficulty. The government say, look, we have to collect this 'cause we don't know when we might use it. We're not really using it all the time. We're not necessarily doing pattern matching at this moment, but what happens if we need to find the kid? We need that for that moment. That's sort of this theory that in order to find the needle in the haystack, you have to collect a whole lot of hay. Right? And that's sort of this idea of collect everything and use it later. But the problem is we and we as society, we as legislators, we as courts, haven't figured out what the rules should be about how you can use that information. In fact, we spent most of our time focusing on collection, right? Most of our laws about whether you can collect or not, not what you can do with it. And we don't have regulation now. We haven't even had the conversation about whether we should have regulation. And in that situation, it's dangerous that all this data exists and continues to live out there because with growing technology, we can do more with it. We can data-minor, we can do things with artificial intelligence. We never could do before. We're able to find these patterns and these practices and thus reveal a lot about ourselves. And it's only get worse because the data trails that we are creating are getting bigger and bigger and bigger and more attractive to law enforcement. What is a research set about this so far? You said that there wasn't any particular gold standard study on this kind of policing and what it's revealed. What do we know based on research so far? So it's important to take each technology separately. So for place-based predictive policing, there have been a handful of studies. One of the sort of most known, one was actually done by one of the companies, PredPol, that is advertising and selling this product. But they had a peer-reviewed study where they said, "Look, our system of predicting particular areas of crime is better than crime animals, which is sort of what we do now." So we actually don't have any information about whether crime analysts get it right. But we do know, at least by this study, that this one particular algorithm did better than the crime analysts. Is that good? It's bad. I don't know. Rand, the Rand Corporation, does these objective studies, created their own predictive policing technology in Treeport, Louisiana, and tried to run the study and came up with the conclusion that there's really nothing-- no statistical significance about why this works or not. We don't know if it's our fault. We don't know if there's something else going on. It didn't seem to work. Other products have tried other kinds of testing. But at the end of the day, we simply do not know if place-based predictive policing is a benefit if it really works. But it may actually be better than what we currently have. Take person-based predictive policing. One of the interesting things about the heat list in Chicago is that with those numbers of, let's say, 400,000 people who have been collected over the last few years. last four years. When you look at the people who might have an elevated list, they have, it's upwards of like 250,000 of those 400,000, a lot of people. So Chicago police gets on the radio and the news and says, look, the people being shot in our community, they're on our healers. We're right. We got them, right? Because we were right, they made our list. But the response that is like, well, you have a completely over-road list. You have 250,000 people who might make your list. That doesn't make sense. That can't be the answer, right? And so we don't know because A, we're dealing with an imperfect real world of real crime solving and real people. We're dealing with police who are not necessarily in the job of doing, you know, double blind objective studies, because they got to solve the last murder. We have a crisis situation of relatively underfunded research opportunities in these worlds where we really aren't doing enough to figure it out. And I'm not positive. This is my own cynical sense. I'm not positive that people really care that much about whether it works. I think that they want to have an answer. My sort of pitch in my book is that part of the lure of the rise of big data policing is a fact that just gives you an answer. It allows you to respond to the reporter or the mayor, whoever to say, we have a solution. Whether it works or not doesn't matter because honestly, it's going to keep changing. By the time you've done the audit, by the time you've figured out whether it works or not, you're going to change the model and we'll be on to something new. And it's a wonderful, seductive way to try to change the conversation and answer the unanswerable question, which we've had forever of, what are you going to do about crime? This holiday season considers supporting the Cato podcast in the broad mission of the Cato Institute by visiting kato.org/podcastsponsor. And learn more about the benefits of sponsorship. That's kato.org/podcastsponsor.