SERVER AND PROTECT
Predpol turned an earthquake prediction model and years of policing data into a map of crime that’s about to take place.
Predpol turned an earthquake-prediction model and years of policing data into a map of crime that’s about to take place.
Two or three times a day in almost 60 cities across America, thousands of police officers line up for roll call at the beginning of their shifts. They’re handed a marked-up map of their beat and told: Between calls go to the little red boxes, each about half the size of a city block. The department’s crime analysts didn’t make these maps. They’re produced by Predpol, a “predictive policing” software program that shovels historical crime data through a proprietary algorithm and spits out the 10 to 20 spots most likely to see crime over the next shift. If patrol officers spend only 5% to 15% of their shift in those boxes, Predpol says, they’ll stop more crime than they would using their own knowledge.
Police departments pay around $10,000 to $150,000 a year to gain access to these red boxes, having heard that other departments that do so have seen double- digit drops in crime. It’s impossible to know if Predpol prevents crime, since crime rates fluctuate, or to know the details of the software’s black-box algorithm, but budget-strapped police chiefs don’t care. Santa Cruz saw burglaries drop by 11% and robberies by 27% in the frst year of using the software. “I’m not really concerned about the formulas,” said Atlanta Police Chief George Turner, who implemented the software in July 2013. “That’s not my business. My business is to fght crime in my city.”
Predictive policing is hot stuf: In a 2012 survey of almost 200 police agencies 70% said they planned to implement or increase use of predictive policing technology in the next two to fve years. IBM, Palantir and Motorola all dabble in it, but Predpol, a three-year-old startup in Santa Cruz, Calif., is among the frst frms to specialize in predictive policing. It has raised $3.7 million in venture funding and a year ago hired a CEO, Larry Samuels, who had stints at Atari and Creative Labs. He expects revenue of $5 million to $6 million in 2015, which he says will likely be Predpol’s “breakout year,” but hints that if the company hooks some big fsh, he hopes to triple that.
Predpol is being used in almost 60 departments, the biggest of which are Los Angeles and Atlanta, but Samuels is eyeing more. “My
goal by the end of 2015 is to have the majority of large North American metro areas using this,” Samuels says. “The market is ready.”
The public may be less so, given already edgy relations with police. Predpol has had to educate people on what the software doesn’t do. It’s not sci-f’s Minority Report, in which police target specifc criminals based on their intent. Instead, it focuses on location, using time, place and type of crime to unearth patterns. For young cops it’s a faster education on the streets they’re supposed to protect. “We have a very young department, and we have to be strategic about where we spend our time,” says Modesto Police Captain Rick Armendariz. Law enforcement agencies are exploiting the analysis in creative ways. For a while a Los Angeles division tweeted out daily hot spots so citizens could keep an eye on them. Modesto parks its “armadillo”—an armored truck with four live-feed cameras—in one of its Predpol red boxes each day.
Predpol’s algorithm was born when Jef Brantingham and his cofounder, George Mohler, were working at UCLA poring over large data sets in the late 2000s. They saw that criminal activity and seismic activity follow surprisingly similar patterns. Each new event—an earthquake or a crime—can be traced back to one of two causes: a fxed factor (like an earthquake fault or a rowdy bar) or a variable factor (like another earthquake, which causes aftershocks nearby, or a gang shooting, which triggers retaliatory shootings in the same neighborhood). Each factor can be boiled down to the usual rate it triggers other crimes. For example, in Long Beach a home burglary instantly puts any home within a mile at heightened risk, with the house next door at the highest risk. “A lot of human behavior can be explained with very simple mathematical models,” says Brantingham.
Predpol is sexy but not quite proven. Atlanta deployed it in two of six precincts and saw drops, compared with flat or higher crime rates in the others. But the cops in those precincts may have acted diferently because they knew they were being given a fancy new tool. Brantingham hopes to get into a peer-reviewed journal Predpol’s most rigorous study, in Los Angeles, which showed that it accurately predicted twice as many crimes as LAPD’S analysts did. An independent RAND Corp. study of a non-predpol predictive policing efort in Shreveport, La. found it had no efect on crime reduction, in part because cops stopped following the program after the frst few months.
Even if Predpol reduces crime, it raises qualms about how it’s applied. Louisiana State criminologist Peter Scharf worries that the red-box designation might cause young cops to exaggerate a neighborhood’s dangers. “I go in this box and everybody’s Michael Brown,” he says. Joel Caplan, a Rutgers criminologist, says predictive software would be better if it helped fx a crime spot’s underlying problems. Others worry that police chiefs and city governments will rush into the embrace of Big Data without understanding how it works. “It’s such a seductive idea that you could have a computer predict crime,” says University of the District of Columbia law professor Andrew Ferguson.
Brantingham acknowledges the concerns but ultimately trusts a cop’s intuition. “We tell officers, ‘It’s up to you to use your knowledge, skills, experience and training in the most appropriate way to stop crime.’ ”
FINAL THOUGHT
“The idea that the future is unpredictable is undermined every day by the ease with which the past is explained.”
— DANIEL KAHNEMAN