Can LAPD anticipate crime with 'predictive policing'?
In an age of big data, California police departments are getting in on the action. In Los Angeles, the LAPD now uses software for what it calls "predictive policing."
In an age of big data, California police departments are getting in on the action. In Los Angeles, the LAPD now uses software for what it calls "predictive policing," which aims to anticipate where crimes are likely to happen before they happen. From the California Report, Aaron Mendelson has the story.
Outside the LAPD’s Foothill Division station, at the northern edge of Los Angeles, police Sgt. Tom Gahry sits in his black-and-white patrol car. He’s shuffling through a stack of maps with bright red squares drawn on them
“These squares,” Gahry explains, pointing to one of the printouts, “according to the computer system, there’s a high probability that a crime will occur within one of those squares.”
Gahry is one of the officers using predictive policing software called PredPol. It takes crime data, runs it through an algorithm, and then generates these maps. The maps tell police where crimes might happen –- before they take place. Police can spend extra time in the areas at risk for crime, the thinking goes, and prevent those crimes from ever occurring. The LAPD says predictive policing has helped reduce crime here in the Foothill Division.
This predictive policing software’s journey started seven years ago, at UCLA. There, academics and police officers began using math to study crime. One of the postdocs on the project, mathematician George Mohler, discovered an equation that transformed the work.
Mohler, who now teaches at Santa Clara University, realized that, mathematically, earthquakes and crime work in a similar way. Mathematical models for predicting earthquake aftershocks could be applied to predict the “after-crimes” of an initial incident.
According to Mohler’s model, one crime sets off a wave of crimes in an area. The equation draws in details from police reports, such as times, locations and types of crimes that already have happened.
Mohler explains, “The idea would be, after that initial report gets filed, then the model says ‘Hey! There’s a risk of after-crimes, or aftershocks.’ And then the police go into that area and they prevent those second and third crimes from occurring.”
The model seeks to predict burglaries, car break-ins and stolen cars. Mohler carefully points out that his algorithm predicts property crimes –- but it doesn’t predict who will commit them. Not like in the movie “Minority Report,” he says, where people are arrested before they ever commit a crime.
After the UCLA team (formally known as the University of California Mathematical and Statistical Modeling of Crime Project) developed this formula, they went to the LAPD. And in November 2011, they started testing the algorithm, out in the real world.
The team worked closely with Sean Malinowski, an LAPD captain. Malinowski, who has a doctorate in public administration, called Mohler’s discovery “a huge moment for us.”
His Foothill Division implemented the algorithm during a six-month trial period, and Malinowski says that it was effective: “In burglary, we had a 25 percent decrease in crime, versus the six months prior. When you think about it, that’s over 100 people that were not burglarized.”
He says he’s excited about the software, and his division continues to use it on every shift. But predictive policing raises potential issues.
Andrew Ferguson teaches law at the University of the District of Columbia, and says that “there’s an accountability problem, there’s a transparency problem and there is a recognition that because all the information is controlled by the police and the police department, it’s very difficult and very unlikely that they will give up that information.”
So, if police are collecting bad data, or if predictive policing isn’t effective, the public may not discover that until much later, if at all. It’s also unclear if spotting a person in a predictive box can contribute to reasonable suspicion, as required by the Fourth Amendment. Ferguson says the courts will have to decide that.
Back at the LAPD’s Foothill Division, Gahry drives around to some of the predictive boxes. His cruiser passes auto body shops and Mexican grocery stores. This police division has a fraught history: The officers who beat Rodney King in 1991 were from this division.
Today, the LAPD says there’s lots of police work to do here, including keeping tabs on 18 separate gangs. Gahry says predictive policing is part of that. It’s not perfect, and at one of the boxes he and I end up looking at an empty lot, full of horses.
Gahry explains that predictive policing changes how police work. “It’s very difficult when you tell the officer, ‘You need to go in this box right here and spend some time in this box.’ And you’re there and you look around and there’s just, there’s nothing.”
Gahry says that it’s his job to tell officers, “‘You’re there to prevent crime. You’re not necessarily there to arrest somebody. You’re there to prevent the crime.’”
And the academic team says its algorithm is doing just that. Its company, PredPol, has raised more than $1 million in funding, and a handful of police departments around the state, including Santa Cruz, Alhambra and Richmond, currently use PredPol. It is also in use in Seattle and in Kent, England.
In San Francisco, police plan to implement PredPol by the end of 2014. And they’re looking at a new feature -- one that predicts gun violence, in addition to property crime.
Note: This production is part of the STEM Story Project, with support from the Alfred P. Sloan Foundation. Presented by PRX, the Public Radio Exchange.