Policing the Predictive Policing: The Promises and Perils of AI Technologies

Welcome to the Harvard Data Science Review Podcast. I'm Liberty Vitter, the feature editor of the Harvard Data. Review and joining me is my co-host and editor in chief. In recent years, police departments nationwide have increasingly embraced data and AI tools to enhance their crime prevention, investigation, and conviction efforts. These technologies range from image analysis on body cams to license plate trackers predicting potential involvement in drought track.

However, a crucial question arises. Are these technologies both accurate and fair? Are law enforcement officers adequately trained to utilize them effectively? Is legislation adapting swiftly enough to keep pace with these transformative changes? Join us as we engage in a conversation with Andrew Guthrie Ferguson, a professor of law at American University, Washington College of Law, and Chief Virgil Green, a Minnesota police chief and co-host of the You and the Law Podcast.

Together, they will provide us insights into the evolving landscape of policing technologies. Stay tuned for a deeper understanding of these advancements and more on the Harvard Data Science Review Podcast. Professor Ferguson, I think I wanna start just sort of with the big picture here. You know, we we hear these words big data, AI, ML. What is big data policing, or if if that's what we call it, and how you know is sort of data being used to really assist law enforcement right now?

Sure. So big data policing is kind of an umbrella term to use uh to cover new Uh digital technologies, surveillance technologies and new predictive algorithms, predictive analytics that let police uh investigate in different ways, do different resource allocation for where police officers go in the city. or county, uh, allows police to investigate in different ways in different places and also changes the relationship between The citizens and police. You have

sort of first generation uh big data policing technologies like predictive policing or automated license plate readers. You have second generation uh big data policing technologies like facial recognition or video analytics. And you can see the the growth of the technology that begins first in bigger cities with bigger resources and more money.

Uh but we have things like real time crime centers where there are, you know, police officers watching the the city as uh it goes about its business. Cars are captured by automated license plate readers and Honestly, almost everything we touch now is digital, uh, and so it offers a new potential for digital surveillance.

Well Chief Green, uh first thank you for taking time to talk to us, talk to our audience that uh Uh my question for you is in your capacity as a chief of police in Minnesota, could you share insights into how your organization leveraged data and artificial intelligence To support law enforcement efforts. Additionally, how have you observed your neighboring agency and departments responding to these advancements?

Thank you all for inviting me to be a guest on your show. Uh I've been at the agency up here in Minnesota for about a year. So coming from Oklahoma City. since I've been up here, I haven't really noticed some of the neighboring agencies, how they utilize uh AI. As the professor mentioned, there's a lot of LPR systems that are being used. There's body cameras.

that are that are being used. And here at the agency that I'm with in Golden Valley, uh Minnesota, uh which is a suburb to Minneapolis, one of the things that I've implemented since I've been here. is Comstat and Comstat is looking at um what is the the crime statistic telling us about where certain crimes are taking place at and how we could u utilize that data

uh to do some type of predicted policing. We're able to see uh where certain crimes, such as property crimes, are in a city where it's probably about twenty three thousand people the majority of our crimes are property crimes. So we don't have a lot of shootings or other uh type of high violence crimes. So that type of data is really able to help us um pass that information on to our sergeants and patrol officers.

So just so I'm clear, this is basically saying there's a lot of crimes being committed in this particular area, so we're gonna send more police to be in this particular area because we know that that's where crimes are being committed, and we think that's where crimes will be committed again. Yeah. Uh again, when we talk about property crimes such as, you know, auto burglary.

or, you know, somebody leaving their car door open. We can always tell our citizens, hey, you know, uh lock your cars, lock your garage doors, all those kind of things. uh it always doesn't happen. So if there's a a a certain part of the city where we're seeing an increase in those type of property crimes and at a certain time of the day now we can deploy uh officers in those areas to where Potentially, hopefully, you know, they can catch some suspects who are committing those type of crimes.

It's such an interesting way, um, to really use the data in a way that can be used sort of across police departments across the whole country. And I think Professor Ferguson, that brings me to your book, you know, on predictive policing. And it really it sheds a lot of light on sort of the widespread adoption of predictive policing technologies across the nation.

What do the sort of status quo predictive policing technologies look like versus what the future of predictive policing policies might look like?

So it's a great question. And building off what the chief said, so you know, many um police departments have used data since. You know, there were police departments. We've we've been collecting information about our communities and recording What has changed initially with Comstat, which comes from Chief Braddon in uh New York City, uh sort of using data-driven policing to guide police patrols.

was a recognition that maybe if you took enough of this information, you could run it through a predictive algorithm and then be able to predict where the next crime would be, or more uh truthfully, that there would be a higher risk uh of crime in that particular area. And so beginning almost uh twelve years ago now, thirteen years ago now.

uh companies started selling predictive algorithms to go sort of on top of that idea of ComSat. There were companies called Predpole and Hunch Lab, which were startup companies that said, look, we can take this crime data that identifies particular high crime areas, uh run it through our algorithm, be able to tell you where you are more likely to see crime and thus that's where you should uh put your patrol officers. That's where you should put your patrol car and deter

uh the next crime. They've tried it in LA, they've tried it in Chicago, they've tried it in New York. And generally speaking, uh it has shown uh not to work, at least in terms of the the software add ons that you can buy in terms of predictive policing. Uh and that doesn't mean that data isn't important to police. It's not obviously police chiefs have to keep track of where the crimes are and what's going on.

about the idea of simplifying this with a particular algorithm or proprietary algorithm just hasn't actually worked out in practice and LA, you know, backed away from uh using Predpill. In fact Predpool stopped being a company or is about to stop being a company. Uh HunchLob has you know been bought by a company called Sound Thinking, also known as ShotSpotter.

Uh and generally speaking, this idea of predictive policing as just using an algorithm to take that past crime and predict future crime hasn't really been able to hasn't really shown to work as promptly. To follow up, Professor Ferguson, I know you in one of your articles you talk about how these uh particular precinct technologies have outpaced the legislations and the academic in a scudny. So I want to ask uh Chief Green

Um could you elaborate on like what is your department approach to on evaluating these technologies, right? And also to both of you, have you either of you observed any progress within the academia or other institutions on catching up with these advancements because obviously somebody needs to evaluate them.

Well and and as you know, policing is always to some degree behind the time. even though there's a lot of technology out there, uh you find some police departments who are a little bit behind technology. And so then when that technology, when they do get that technology, now they really don't know how to utilize that technology to benefit what they're trying to accomplish. And so, um that's where you got to have, you know, a really good, you know, crime analyst. You got to have

really good uh command staff who is able to understand what that data is telling you and how you can utilize that. As the professor mentioned, you know, predictive policing, you can predict and say, Hey, you know, there's been ten homeburgeries over here that happened on a Tuesday night and so next Tuesday we're gonna predict that that's gonna happen again. Well you show up on that Tuesday night, that doesn't happen.

So you wasted a lot of resources trying to predict that. And so one of the other things that I'm looking at at this agency here is how to use a data dashboard to help our patrol officers uh with their patrol pattern. one of the things that it does, it shows officers where they're at, where they're going. And now we're able to look and say, okay, well, you know, we had a crime over that occurred over here and you may never even patrol that part of that neighborhood.

So now we're able to show officers where they're driving at, how their patrol patterns are, and coming back, letting the supervisors know that and how that can really help. put them in uh better positions to patrol uh much of the city. Professor Ferguson, what's your your thought on that?

Well, I think that the lack of any sort of legislation or or regulation around uh policing technologies is a problem. Um, in part because it's a problem of just where we are. You can imagine Chief is no different than many other chiefs in America who has a community that would like crime to go down, uh isn't given a lot of resources, and is offered new technologies that might uh help him or the other chiefs address these crime problems.

The problem of course is there isn't anyone necessarily in the policing staff who has sort of done the the thought process well what will happen if we start deferring to the algorithm that tells us to patrol this area or that area? Or what about, you know, racial justice issues? What if the algorithm keeps pushing us to the same area so even though we're we seemingly are following the data, we're really following the data that looks like we are now over policing certain communities.

And there isn't really an institution or entity or regulation that has done the due diligence before these technologies get rolled out. Um, there are lots of vendors out there selling something that sounds really good. There's actually a lot of federal money that goes in through the NIJ and other uh National Institute of Justice and other uh a grant programs that allow sort of pilot programs for these new technologies.

But there isn't a lot of money to study the impact on either the communities or even honestly the police officers uh who have to follow the technologies or in some ways doing some of the data collection themselves. Uh and so we have this gap where the technologies are outpacing the sort of policing community and the regulatory space.

The vendors are sort of taking advantage of that. Most chiefs of police in America today are not data scientists. They don't have a data background. They couldn't explain the algorithm underneath it. That's not their fault. They're a chief of police. But there's a an information imbalance. And that has resulted in a lot of technology being sold and used without really thinking through the actual risks in terms of how it's impacting communities, how it's diverting resources.

and whether it really makes sense to do this kind of predictive policing investment as opposed to investing in other parts of the community that might deal with the underlying issues that create crime in the first place. You know, Professor, you bring up a good point. Uh, when you talk about the technology outpacing you know, you can go to an IECP conference. IECP stands for the International Association of Chiefs of Police.

And you got all these vendors there who are trying to sell you the latest product. And if you are not up on researching that product that they're selling, you know, you can come back to your state and and say, Hey, I like this and we'll we'll try that out.

and then, you know, a year or two later you spent fifty or a hundred thousand dollars on something that you you're finding it's not useful for your agency. So one of the things that I would encourage police chiefs to do is include whether it's your crime analyst or talk to other people to see if that piece of equipment or that software

uh is gonna benefit you. Uh being new to the Minneapolis area, I think the city of Minneapolis passed some type of legislation where facial recognition is not utilized by that agency anymore.

And so, uh I'm not trying to speak for that agency, but it's something that I just recently ran across where uh that technology is not used. Well, obviously the the former police chief felt it was a benefit to the agency, but I think a lot of people look at the uh racial disparities utilizing uh facial recognition because

Some people think that it's targeting the minority communities more than it's targeting uh any other community. So there's a lot of uh unknown uh answers to to this type of technology. You know, your both answers that reminds me to ask uh particularly to uh Chief Green that what has been done now to uh provide, you know, training so basic Uh data you know, f preparations for the police officers.

'Cause most of us as y we know that, you know, t uh communicating about statistical data is not that easy to the broad audience. I wonder what what's being done here. Well, I I have to say that there's really a lack of training. Uh and I think that's again, that's where we're behind the curve on understanding the technology, how we can use the technology. And then once we have all that information, what does that information tell us?

So, uh right now there's just not a lot of uh training for police officers or really for, you know, crime analysts within agencies to fully understand what that data is saying and how to utilize that data and disseminate that data to to help you solve some crime. And almost every analyst or you know, police officer that gets up to speed in being able to be sophisticated with the data then gets hired by those vendor companies that are selling at the ISCP. And there's actually a a real loss of

uh internal learning. Like many Departments have recognized that there's a sort of a data literacy gap. And so they spend the resources to train internally uh to build people up. But then that deputy chief that's sort of been in charge of technology is immediately snapped up by a private company that will pay a lot of money. And they're the ones who are actually at IACP selling the chiefs later on because they can talk the talk.

uh and they know from the inside. And so there's a real lack of sort of a adept in policing itself that can sort of ask these critical questions because there's a lot more money to be made on the vendor side and that's what happens. Yeah. And and you're getting stuck you're getting stuck with some equipment that you really don't know anything about.

I think this brings up a really good question. You know, historically we look at these algorithms, everything from, you know, Chicago's heat list or hot list or I I think it was also called Chicago's strategic subject list that would sort of try to predict who is gonna commit crime.

um to facial recognition systems where, you know, we saw it in Detroit where, you know, a a man got arrested and he was, you know, wasn't thirty miles within where the crime happened or I I think it was in um Atlanta where someone was jailed for a week because of facial recognition system being wrong. I think it really puts the police officers also in a bad position. You know, if you tell a police officer, if you tell anybody,

That there's a 69% chance that this person's the bad guy. I mean that's really hard to step back and understand what 69% really means. I'm gonna be scared if they tell me there's 69% chance that's the bad guy, even though there's a you know 31% chance he's not. And we saw instances with Robert McDaniels from the Chicago's Heat List, who in twenty thirteen was told he was ninety-nine point nine percent more likely to be part of gun violence, but

had no idea how he was gonna be part of gun violence. Was he gonna be the victim or the perpetrator? So how can we ensuring public caution and using the fact that technology can do great things and can be really important for policing, but also not induce panic or have really bad things happen from this new technology. So

Two points on each of those different problems. So predictive policing, which began as place-based predictive policing, eventually evolved into person-based predictive policing, like the heat list. uh that you described in Chicago where the thinking went, well, if we can find risk factors that might uh be present in people's lives, maybe we can predict

who are the likely people who will be involved in violent crime. And so they came up with an algorithm of sorts where basically you got extra points. They sort of added up the points. of whether you had been arrested, whether you'd been involved in a violent crime, whether you'd been shot. So sometimes victims were put into the same list.

And based on essentially arithmetic, it wasn't even much of a an algorithm, they added up these lists to come up with the people that they thought were more likely involved in violent crime. Now there were a couple problems with the system. The first was many of the inputs were sort of police related. So for example, if an arrest uh gave you a heightened risk for further uh violence, that meant that police could sort of influence the inputs because

police officers were contacting certain communities, poor communities, communities of color in Chicago more than other communities. But also the fact that you put sort of violence in and that y a victim of a violence could be uh seen as a higher risk. uh led to a system where at one moment in time everyone arrested in Chicago was given a risk score from zero to five hundred plus. Five hundred plus being the most risky people.

And that information was given to police officers, so when you pulled someone over, you could pull them up on the screen and see the five hundred plus, which of course might cause you to tr treat them uh differently. The problem was there was no uh basis behind why someone was more riskier than others. It sounded like there was an objective number, but that number wasn't necessarily based on anything except for this, you know, arithmetic of uh prior contacts with police.

And eventually, as the system expanded and more and more people got involved, the police themselves realized it wasn't all that helpful. The risk itself was sort of uh spread out over eighteen months, so it might be statistically more likely that this person could be involved in a violent crime the next eighteen months. But what well how does it help me today? I pulled him over. I don't know what he's doing, how he's doing it, how do I treat him now?

Um, it also didn't solve what was supposed to happen. The original theory of the pla person-based predictive pleasing was the idea of like, well, if we identify risk. Maybe we can address social services or help this person get out of this risky lifestyle where they might get shot or shoot someone. Uh and the recognition was that the underlying theory behind predictive policing didn't quite work. And it was targeting poor communities and primarily men of color in Chicago.

And eventually it fell apart. Though sh they they backed away from uh using it because they realized it was too fraught with the mistakes that were written up like Robert McDaniels and other people who were targeted because his friend was shot. Uh suddenly police came to his door and said, Hey, you're on this list. You're on the heat list. We're worried you're gonna get shot or shoot someone and he was like

All I am is like a a friend of someone who got shot. That seems unfair. And as you point out, the same problem of this reliance happens in things like facial recognition, where the system was largely normed on pale white faces, male faces. And so when confronted with anyone who is not in that norm, like an Afro-American or more particularly actually African women are the worst in terms of the matchings, uh, they do a poor job.

Uh and that means that someone who's a closer match might get targeted or someone who uh doesn't quite meet up is m is targeted. And the thing that connects them both is this reliance.

on a essentially untested or not fully tested theory of predictive validity and reliability that after we studied it didn't quite that it worked as promised, which has sort of been the history of all of these predictive analytics policing systems where we kind of trust it and then we test it out and we realize it doesn't work as promised.

You know, what you just said is really uh important because we know as a data scientist a lot of these algorithms based on you know, uh machine learning algorithm, trying to just recognize some patterns, you know, without much the humans thinking hard about, you know, do these patterns really make sense, you know, what what they're really telling us and this m these terrible mistakes.

Uh, definitely can happen. And so my question to both of you that considering those things can happen, and there's a rapid advancement in the technology. What other emergent technologies or things you can think of that will we should be particular v you know vigilant about them just because of You never know what's the next thing happens. It could do even worse, right? Is there anything it's again a question to both of you that we should really be very careful.

Well, I think for me, uh obviously there's a few. Uh obviously facial recognition and and obviously, you know, how do agencies are utilizing uh LPRs, license plate reads. especially in uh communities that are marginalized communities. And so again, you know, one of the things that we in the police of the industry do when we have all these tools.

we don't always utilize those tools and resources to the best of their ability. Uh and we're relying on men and women uh to make some good judgments decisions and oftentimes we've seen where that has not always been good and the public suffers from that. And so one of the things is that how do we make sure that the men and women who are utilizing these technologies, uh, with new technologies coming out and agencies are going to

Try them out, see if they work. Um, how is that going to benefit the community? I think the biggest question that a lot of us need to be asking ourselves is. What is this uh technology going to do to help our community? We have to have enforcement uh tools there, but there again, I think we just need to make sure that with some of this technology that's coming out is pretty advanced.

especially when you get into some of the advancements of AI, how is that going to really help that police department uh make a city much safer versus how is going to help them on the enforcement? Let me give you two examples that I think to me, even for someone who studies uh this stuff uh all the time still finds shocking. So this uh spring in in March twenty uh twenty three there was a a court case involving a gentleman who was driving in upstate New York in Scarsville, New York.

And he was pulled over because the automated license plate uh reading system in the state of New York had identified the pattern of his car as being suspicious of a drug trafficker. Now in upstate New York, there are about 480 automated license plate cameras that they read about 16 million plates a week, 1.6 billion license plates over the last two years. And the algorithm had picked up his pattern of driving back and forth from Massachusetts to upstate New York.

as being suspicious. So his license plate was put into the system, police pulled him over, and lo and behold, he had cracked cocaine,$34,000 in cash, and a handgun, and he was promptly arrested for uh drug trafficking. It is a situation where in order to find this one person they searched.

almost everyone driving in New York State's license plate to flag him for suspicious uh driving. And when you look beneath the surface, the suspicions aren't apparent. There are about six traveling patterns that he did. Didn't, you know, seem uh drug trafficker, but something worked for the algorithm, so they picked him up, they flagged him.

And they arrested him and he eventually pled guilty to the crime based on this true that's true AI uh automated license plate uh uh system that is a a game changer in terms of the way we normally think of automated license plate. Second story involving video analytics, which to me, like facial recognition, this idea of pattern matching is gonna radically transform policing. In real-time crime centers and major cities, which many major cities have.

There is an a software overlay on the video cameras that are constantly running called video analytics, which is basically a pattern matching. uh system where you can match out car, white car, white van, man, woman, man with hat, man with red shirt. You pick the object, it can it's trained to learn and find it in all of the camera feeds. So imagine camera feeds running all day, 24-7, throughout the city. You can program in, I would like to find the man with the white hat and red shirt.

in the city. And you can find all objects that are man, white shirt, red hat. superimpose them in time so you can see where they are in location and in time. And then if you find someone who sort of fits the pattern you're looking for, you can trace them back through the cameras over time. It's an incredibly powerful investigatory tool. It's also a surveillance tool if you just want to do what are called virtual patrols by just like sort of skim through

the camera systems that are there. And as we are building more and more cameras into this network, not just police cameras, but private business cameras, your ring camera on your doorbell, whatever it is you i if you wanna bring it into the system, that can then be searched through these video analytic systems. And it is incredibly powerful. It'll obviously be very valuable for law enforcement to do its investigative work.

Uh but it's also incredibly invasive to individuals who are going about their business, just like all those drivers in New York State, that didn't know that they were in a database of one point six billion license plates and some algorithm is seeing whether or not they're suspicious or not. You know, for this I it's great the drug traffickers off the street, I think everybody would agree.

Did they catch anybody else? How many people did they pull over, or how many people did the system flag before they caught the one guy? And Chief Green, this you know is to you as well. How many people had to have interactions with the police?

that they didn't necessarily want to have or weren't right for them to have. How do we balance these interactions with the police that may not necessarily be a good experience for people with actually catching the one drug trafficker that everybody agrees should be off the street? Well, y you know, when you look at the technology, you know, let's say, you know, license plate read is what we're talking about.

You can have a hundred thousand plates being read and you've got somebody going through there looking at that information and as the professor mentioned, how that um uh was being utilized in upstate New York. All of a sudden now you you got certain people who are being pulled over, you can say targeted. And all of a sudden now these people are untr trying to figure out why they're being pulled over.

uh they haven't really committed any type of traffic violations or there's some other very minor traffic violations that they may have committed, but Uh, that officer is going to try to get some kind of problem cause to get inside of that vehicle based on the information that they've received. There again, who's checking that information to make sure that that information has some credibility to that vehicle? And let's say

that same person isn't driving that that vehicle that day and all of a sudden now that person becomes a victim. That's something that we definitely need to be mindful of is how that technology is being being utilized.

Uh when you talk about license plate readers, it is one that has benefited law enforcement, especially, you know, when you're looking for a child molester or a murderer, but in other ways it it is definitely been a a a tool that has targeted uh especially uh people in black and brown communities. And just to the drug trafficker case, the only reason we know that is the intrepid uh defenseler, the federal defender, uh did a FOIA request or a FOIL request in New York.

uh state to determine what the original suspicion hit was because it wasn't obvious. It wasn't in the police paperwork. It wasn't mentioned. They weren't sort of bragging about the fact they use this technology. It was hidden. Uh and the danger is that not only is it obviously warrantless, but it's kind of a warrantless search against everyone in New York State who's

license plates uh was used. But you can imagine how it could be used. Let's say you wanted to target New York's not the greatest example, but in states where uh abortion is now criminalized, you could target the cars around an abortion clinic. or that cars around a, you know, um clinic for undocumented individuals. Or uh really any place you can be able to use these technologies to surveil people and reveal

By inference, what they are doing outside that place. And that can obviously impact real people and real liberties. uh as they go about their business. And again, without regulation, without rules, without even a again a probable cause warrant requirement, uh which isn't required at this moment. It's all fair game. And that's just one technology. And in fact, one of those technologies that actually most people would think is

now commonplace and pretty benign because generally ALPR is used to identify like stolen cars. And so if the camera passes a stolen car with a stolen license plate, it alerts and that's Pretty good. uh policing to stop the the car. What we haven't seen is how when AI algorithms are taking that same data set and doing something new, it changes the power and adds a whole l new level of questions and um concerns. about how police are using uh this information.

I think that's really interesting and it's it's so pertinent. As you said, it's just one technology. You know, we've already seen um uh prosecutors use people's cell phone data, location data against them in cases like this. So it's it's even sort of more serious when it comes down to license plates and all these other ways that we can track people.

And you know it it it's obviously, you know, just as a citizen, these are concerning questions, but how are these technologies viewed by the police officers themselves? You know, they're reading these stories, they see these cases just as any American does.

And they're the ones that need to buy into this technology and learn about it and be trained on it. So how are police officers seeing this technology? Are they are they buying into it? Do they want it? Do they think it can help? You know, where where do they sort of stand on this?

Well I think you see a lot of police officers who do buy into the technology. Again it goes back to how that technology is being presented to that officer. You know, we go back to L PR. this all really kinda started out as being how do you uh track a stolen vehicle. That was the biggest thing with LPRs when it really first came out.

Then you have some states who are using LPRs to determine if a vehicle uh has insurance on it. And so anytime you uh present this type of technology to police officers, That's just a a way of saying, Hey, we can go out here and we can catch the bad guys but what you're not really understanding is once that technology is there and that data is there, what somebody else is doing with that data because now

Uh you may have a supervisor come back and say, Hey, we've got this going on in this part of the city. We need uh d to deploy more officers over there. They're not r fully understanding what that data is is telling them to do. So it could be abused. And so again, we need to make sure that if we are utilizing this technology that we're utilizing it to benefit the community, not not to benefit the department, but that it has some way of helping reduce crime in our community.

Two quick points. One is just on the ALPR, so you could actually use the benefits of the technology to be able to identify a stolen car or you know, an Amber alert or whatever it would be from the vehicle without necessarily creating the data set of 1.6 billion license plates and then using that for a search. You could literally just get rid of the collected information after an hour, a week, twenty four hours, whatever you thought was the appropriate time.

and avoid the problem of predictive analytics into an entire data set that allows you to go back in time and see where people were and what they were doing. Uh in terms of the police officers uh reacting to it, of course, you know, in that case All the officer knew was that there was an alert for this license plate at this time and if you see it, pull it over. So they saw it, they pulled it over, and it there's no way for that. You know, line officer, that state trooper.

to do the work to figure out, is this a good algorithm? Is it racially biased? Was it vetted? You know, that's in some way an unfair ask for that officer to do that. We would hope that before it gets rolled out, we would have asked these hard questions. had data scientists and other people interested in these ideas.

sitting down to uh really interrogate whether there could be a problem and we've almost never done it. We've almost never done it at before we've rolled out these technologies'cause the sort of commercial pressures to sell and get it out first have overwhelmed the caution and the risk assessments and the lawyers worried about, well, how will this be uh it how will this impact communities? Because if you take time to do all that, you kinda lost the contract and you won't be able to sell it.

uh and you won't be first uh to move in the market, which has been sort of the game. It just to touch on if I can real quick, uh license plate readers. You have a lot of cities who are utilizing these license plate readers. Uh they're, you know, position'em in as soon as you drive into uh a city. and uh that information is being collected. Now you have agencies who are saying, Well, hey

This person is from from another city. They're coming into our city. Obviously they're coming into our city to commit crimes. So

Again, you have that technology, which is a good piece of technology, but it's not being utilized the way uh that it's supposed to be used. So now you got officers who are now sitting somewhere They've identified maybe a handful of cars and now they're pulling those vehicles over based on the fact of the information that they've received from that data that's been collected. So far we have been talking about everything is really what it what's has happening in the United States.

uh this policing. So but I wonder, you know, technology like this obviously is not restricted to US and is there any country doing particularly better and anything we can learn from and uh lessons and you know anything about There's a political precinct going beyond you.

I think many countries have uh are doing worse than us in the sense of they are uh using the technologies without with even fewer restraints. Uh China is a great example of a surveillance uh world where uh we have seen, you know, cameras everywhere, video analytics everywhere, social media and social uh uh credit scores that are basically the heat list on steroids.

being able to identify uh people who are at risk. And so we've seen that. Europe has borrow you know, is maybe about five years behind. It keeps borrowing some of our bad ideas about predictive policing and technologies.

Um we've seen a little bit of pushback about facial recognition concerns in Europe about that. So you do see a little bit of fear there. But you know, in the US we're we're pretty unregulated with facial recognition, so maybe we're equally bad. Uh but the technology is uh you know, it's it's a global world where people can buy it and are using it and uh we're seeing AI deploy in different places and different

uh environments and a lot of times w they're piloting it there to come back here after uh they tested it on some border. If you fly through Europe. It's all facial recognition now, right? You literally it's all passport control, facial recognition. Singapore, all facial recognition, like it's everywhere and there's no way it's gonna roll back. And we'll see that technology uh, you know, trickle down through the US.

even though people are raising concerns, even though people are worried about it. Um but we'll see, you know, the technology will evolve and and then be sold back to us. uh in different ways and eventually probably be, you know, used by law enforcement and cities to surveil communities and uh investigate crime.

So I think this brings us to our magic wand question, which is always my favorite. So if you could wave your magic wand, what technology would you want for policing? If you could pick one technology that you would give all police departments, what would it be? I would have to say body cameras. It gives two sides of what's going on and that's the officers uh what the officer is seeing and obviously

having that communication with the public. It's a piece of equipment that's been around for for quite some time, but there again it's it's also been controversy as to when the officers uh or using them and when they're not using them. So for me I would have to say how body cameras I think the the question, you know, assumes that there would be a technology that would necessarily deal with some of the problems we know that exist in policing in America today.

And I think in some ways it's a a backwards way of looking at it. I always joke that the problem with predictive policing is the policing part. The idea of doing like risk analysis and data-driven uh sort of studies about why certain environments sort of create crime, certain places are higher crime areas, or why individuals might be more at risk.

is probably a worthy endeavor, but that the remedy of policing is not obvious at all. Like why would you if you knew a young man was more at risk for being shot, why would you send a police officer to his door rather than someone who could hire him and give him an education? If you knew there's a certain area that had a higher risk of car theft

Why not do something to improve the lighting in that area, build a park, you know, give kids something to do besides steal cars instead of policing your way out of it? The idea that the solution would be an algorithm tells you to put a police car

at the right place at the right time is such a a a sad way of thinking about these problems, right? There isn't a technological solution. And there won't ever be a technological solution to a crime. The crime is human and so the solution should be human and we should be looking to reinvest uh the money that we might spend on some algorithm in those communities to help those kids so they're not doing the dumb things that they're doing to get in trouble uh with the law. And so I think that

the technology solution for crime is a problem. I think if you wanted to use technology to improve policing, you could turn all of the predictive analytics and surveillance on police. to see how they're doing their work, forget body cams on the c community, put it just directly on the police and cameras and everywhere else.

And the backlash you would get from police officers saying, Why are you surveilling me? Why are you doing a risk analysis of me and whether I might, you know, be more violent will give you insight about what the problem is. Like it's unfair to target someone who, you know, based on an algorithm and judge them for that. And I think that, you know, if you turn the architecture of surveillance on police

um you would find resistance of police themselves. It's like that doesn't seem fair or right. I don't want that. And so if you had to pick a technology, I would pick a technology that surveilled police as opposed to community. Well thank you. I think that's very inspiring answer. I think what you're saying is when we think about technology we s should think about a lot more holistic way. Uh by the time you

Focusing on policing is is kind of too late. You want, you know, be be preventative. I think that's a really interesting point. And thank you to both of you again for coming on to this podcast and thank you for your great work. Thank you for listening to this week's episode of the Harvard Data Science Review Podcast. To stay updated with all things HDSR, you can visit our website at hdsr.mitpress.mit.edu or follow us on Twitter and Instagram at the HDSR.

A special thanks to our executive producer, Rebecca McCloud, and producers Tina Toby Mack and Ariane Winfrank. If you liked this episode, please leave us a review on Spotify, Apple, or wherever you get your podcast. This has been the Harvard Data Science Review, everything data science and data science for everyone.