In this episode of the 10 Blocks podcast, editor Brian C. Anderson and contributor David Black discuss “predictive policing.” Read “Big Data on the Beat.”
Brian: A new policing approach, predictive policing, is being tested in several big cities across the country. The method uses advance algorithms and powerful computers to predict where crime is most likely to occur so police can then deploy officers to prevent it. It sounds like science fiction, but is it for real? Joining me today to discuss the emergence of predictive policing is David Black a scholar in residence at Kirkland House at Harvard University and a prize winning, novelist, journalist, screenwriter, and television producer. David’s fascinating new City Journal article, “Big Data on the Beat,” in our winter issue and available on our website, will be the starting point of this discussion. Welcome, David.
David: Glad to be here.
Brian: Let’s start with three very basic questions to get oriented. What exactly is predictive policing beyond what I just said? Who invented it? And can crime really be predicted in this way?
David: Uh, the last question first, yes. It seems to be able to be predicted and in 2014 in the Foothills precinct in Los Angeles, California there was a drop of 34 percent, over the previous year’s two digit drop in crime compared to other precincts and the same precinct in the past. So it seems to work.
What predictive policing is: The kind that Bill Bratton, and Sean Malinowski, and Jeff Brantingham developed in Los Angeles starting in the Foothills precinct is basically a development from the CompStat system that Bill Bratton and Jack Maple developed in New York City, which is the equivalent of putting colored tacks on a map showing where crimes had occurred, what neighborhood, what streets, what blocks and sending police into those neighborhoods because there seems to be a need for it. Predictive policing is a level more complicated than that.
They use the same kinds of history of crime in a neighborhood or in one of those boxes, block size boxes, and they also use weather. They use elements that may predict the possibility of a crime. Is the block near a strip club or a mall or a school where kids might come out and hang around? And all these elements including real time feedback from police cars, every police car has a computer in it, allows them to find out when one of the boxes becomes a hotbox: When there is a possibility of a crime appearing there. At that point along with the regular patrols the command center will send a patrol car into that hotbox to drive around maybe once an hour, maybe once every 40 minutes, in a way that prevents crime. And this causes, at the beginning caused, a bit of pushback from the cops because normally a police officer feels he has accomplished something if he makes arrests. This is about preventing crimes so the police officers aren’t making arrests, and they feel they can’t judge how well they are doing.
When I did a drive around in the Foothills district, when I came back, Sean Malinowski who worked with Bill Bratton on developing PredPol in Los Angeles and who is an unusual Cop. He’s a PhD and a former Fulbright scholar. He said how did it go? And I said, well it was quiet, which for the sake of the article I thought was terrible because there’s no action. He said, that proves it’s working.
Brian: LA is one of the forces that has tested this program, and is now expanding it I believe.
Brian: What other departments did you visit that are testing it out?
David: Tacoma, Washington is using it. Atlanta. There are about six different cities in the past two years that have deployed the predictive policing in this method. There are other forms of predictive policing. This form is more the gentler version. I mean, big data is here. It’s going to stay. It’s not going anywhere. You can’t put the genie back in the bottle, and there are various versions, various ways of using big data. One is the Darth Vader version, and as Commissioner Bratton once said, I could make New York the safest city in the world; you might not want to live there. And the other is the Obi Wan Kenobi version which is the kind that was rolled out in Los Angeles and Tacoma, and in an odd way it’s both very forward looking, and looking also to the past to Sir Robert Peele who invented the first professional police force in London. Based on his name the cops are still called Bobbies, and he developed a police force that was. . . They were citizens. They weren’t an army. It wasn’t a military force that was taking over a neighborhood or a city. The police were us and we were the police, and that’s why policemen in London were not given guns. And in the 50s and 40s and earlier decades of the last century patrolmen, foot patrol officers, knew their neighborhood. They knew the kids growing up. They knew the shop owners. They knew the parents. They knew everybody in a very intimate way. They could tell if a kid was in trouble, and they could intervene by talking to the kid. This meant that you didn’t have to have any kind of screening or stop and frisking. It was a neighborhood orientation. Predictive policing at its best is replacing the foot patrol with cars, but it’s also an attempt to give the police a chance to be visible in the neighborhood and prevent crimes.
Brian: So it has this preventative emphasis.
David: It’s absolutely preventative, and everyone who talks about predictive policing, the PredPol model, the more beneficent, I think, model, says it doesn’t work without neighborhood and community involvement. When I drove around with some officers in Los Angeles there were a number of parks where there used to be a lot of drugs and gangs, and there are now nighttime basketball games. So I think the community involvement is important. Also, the police who are most successful doing predictive policing tend to try to dial down confrontation and that’s also very important. One of. . . A number of policemen who are using predictive policing have experience in conflict resolution, and when I was walking around with one of the Officers in Los Angeles we were confronted by a bunch of kids at the end of a tunnel, and I could tell from the body language of the officer I was with that he was on the alert, but he engaged them and said, how are you doing? What’s going on? And he defused the situation. They disbanded and wandered away. Many many crimes are committed on the spur of the moment, or at least the crimes that are susceptible to predictive policing which tend to be burglary, auto theft, and theft from autos. Some of the more serious crimes, rape, murder, are less predictable, but the crimes that are more predictable tend to be a spur of the moment and they tend to be kids from within the. . . or people from within the community, very often kids, not kids from outside the community. So if a burglary happens in a high income neighborhood chances are it’s a kid from that neighborhood. If a burglary happens in a low income neighborhood chances are it’s a kid from that neighborhood.
Brian: To speak a bit about the other version of predictive policing. There’s a famous Philip K. Dick short story first published back in 1956 called minority report, later made into a pretty famous movie with Tom Cruise. In it Precogs, as I think they’re called, can see visions of future crimes and the police can then arrest people before they’ve committed the crimes. Now some might look at predictive policing, the application of big data to potential crimes and worry about similar patterns or abuses occurring. Couldn’t the use of big data to predict criminal patterns lead to disturbing violations of innocent people’s civil rights.
David: It could, and all the people who I’ve talked to who used the PredPol that’s in use in the cities I mentioned. . .
Brian: And so PredPol is a particular a model.
David: Is a particular model. There are other. . . and Commissioner Bratton is bringing it to New York. There are other models which get scarier because in Predpol they are really looking at a geographical spot and saying, in this spot for various reasons, history of the spot, crimes that have occurred there in the past. And crimes will very often occur at the same place. Someone got away with it and they’re going to come back to the same place, but its attempt to use geography rather than any kind of profiling.
There are other more troublesome models which combined what Predpol does with what big box stores do. For example, where they can tell from your iPhone what aisle you’re in, where you stopped in that aisle, which direction you’re facing, and if you’re looking at a bunch of cereal boxes suddenly that information goes into the computer. That can be dangerous and one of the things Commissioner Bratton said is, in a democracy where we vote people to take care of us, to basically, to enforce our rights, it’s important for citizens to vote smart, to vote for somebody who is going to be aware of these problems because any system, whatever the system is, is only as good or only as bad as the people who are running it.
Brian: Police officers would have to be comfortable using this kind of information. Is special training needed?
David: I think, not just special training but a kind of new model of seeing themselves in the Robert Peele sense as members of the community. There are some cities where policemen are encouraged or forbidden to live in the neighborhoods in which they police. There are arguments on both sides, but I think if the model that you are trying to get back to with the Predpol that Commissioner Bratton and Sean Malinowski, and Brantingham have developed is an attempt to. . .
Brian: Oh oh, I’m sorry to interrupt you David but who is the latter figure?
David: Jeff Brantingham is a cultural anthropologist who helped develop the algorithms, and of course that’s also a problem which speaks to your last question. Most of us don’t know what a good algorithm or a bad algorithm is and so the model could be flawed on that basis too, depending on the information put in. So there is. . . The same way we have to trust the people we elect to be honest brokers we have to trust the people who are putting in the algorithms to not weight the model in one way or another. But I think it’s, it does seem to be working in the places I’ve mentioned. There does seem to be a double digit drop in crimes. There doesn’t seem to be any increase in violations, in fact the opposite. So I think, it’s not so much of a special training although that’s important; for an officer to be driving around and get used to having a computer in his car is a big thing. It also means you’re watching sometimes the computer, not necessarily through the windshield, and I did talk to a number of policemen in different cities and asked, what are the tells? What would make you focus on one person or another to suspect something may be up? And he said, well somebody at night who avoids street lights, somebody who is walking close to the curb as though they’re trolling, whether it’s a street walker or a thief, somebody who has their hands oddly, as though they have a weapon and they’re about to grab it. So there are some body language tells.
Brian: This sounds like just traditional. . .
David: Just traditional policing, but it’s..
David: Intuition, and that’s important. I used to hang out with the 34th precinct in New York City a lot. Detective Jerry Giorgio, who’s by some considered the dean of New York City homicide cops, and a lot of what he went by was his experience over the years. He listened to his gut and I think we all do. I mean, when you’re walking down the street sometimes you cross the street, sometimes you don’t, sometimes you speed up your pace, sometimes you don’t when you see someone coming. And it’s not a matter of race or ethnicity. It’s something, usually body language. You can tell from body language.
One of the things I was talking about as I came in here was, when you have two different cultures colliding, whether it’s in Miami or in New York or wherever... For example if you have an Anglo culture and a Latino culture colliding, the distance between people in one culture which feels comfortable, not a violation of my personal space, may be different from the other culture, and that could lead to confusion, that if you get too close to me I may feel you’re being aggressive and you may feel you’re being intimate and just confidential. But whoever feels they may be being, their personal space is being invaded, could react badly and could lead to horrible things. And I think part of the training you are talking about has to do with cultural training as well. What do I think is happening? What does the other person think is happening? And a lot of what some people perceive of as violation of civil rights by the cops is really two cultures, to worlds that have very very different assumptions about personal space, about cultural cues, colliding, and there is no right and wrong, but it can be explosive.
Brian: There is an analogy here between what we’re seeing in other areas of the culture, certainly sports, where advanced analytics is being wed to old-fashioned scouting and intuition, to come up with better possibilities for drafts and better players.
David: Well a lot of the Predpol, or original productive policing models came from Moneyball, came from using that as a model, and it’s far beyond what I understand, I flunked science in high school, but that enables me if I can understand something I can describe it in a way that other people can because it takes me longer to understand it. But Moneyball was absolutely a model for predictive policing.
Brian: Final question. Could predictive policing methods help in predicting where and when likely terror attacks might take place?
David: Absolutely. I think it’s a more complicated model. The best model for predictive policing is the predictability of the crime. For example, in Tacoma one the people who brought in predictive policing did a PhD thesis where she found that a house that was burgled had a much higher chance of being burgled again, and if not that house the house next to it. I mean there’s a certain amount of predictability partly because of operant conditioning. We, y’know, if something works for us we do it again. As I said, crimes of violence or passion like murder or rape and terrorism, they’re harder. There’s so many variables. It’s harder to predict, but I think there are models that could allow for that, at least that would get us closer to learning how to predict. And again the balance is learning how to predict without violating our basic civil rights, and what typically happens whether it’s 9/11 or a more local crime in a neighborhood. I live in the country and if there is a home invasion in the country suddenly everyone is locking their doors for a week or two and people don’t usually do that. We get stampeded into responding out of fear rather than out of logic or reason, and I think one of the admirable traits of Commissioner Bratton and Malinowski and the people who are using predictive policing is that they’re reasonable people, and that’s more rare than we like to think in any culture.
Brian: Thanks again David Black for joining us. David has a new City Journal article entitled “Big Data on the Beat,” in our winter issue and it’s available on our website. You can tweet your comments and questions about today’s discussion to@CityJournal with the hashtag #10blocks. Thanks again, David.
David: I’m glad to be here.