In Hot Pursuit of Numbers to Ward Off Crime

At the Seattle Police Department, the morning roll call begins with Google Maps and computer algorithms.

A supervising sergeant pulls up a map of a precinct that is dotted with red boxes marking the beats that the officers should focus on. The map is the product of software that crunches crime data and tries to predict where crimes are most likely to occur over the next few hours. Patrol this parking lot today after 8 p.m., the algorithm might suggest, or keep an eye on that stretch of road for car break-ins.

“It gives the officers more direction,” Sgt. Bryan Clenna, of the city’s West Precinct, said, “instead of driving around like drone bees.”

Predictive policing programs like this one, used for now only for property crimes, are a harbinger of how, for better or worse, police officers across the country are relying on the data analytics business. Seattle is at the center of this experiment.

The city’s license-plate readers record the movement of vehicles, information that is then stored in a vast database to be tapped later in criminal investigations. Its red-light cameras record traffic violations and issue tickets, though unlike many other municipalities, Seattle requires a police officer to vet each citation. The police department uses Twitter, that rushing stream of data, as its crime blotter, letting residents report possible crimes in their neighborhoods.

And, starting in July, the Seattle police will use a similar software program to predict where gun violence is likely to occur.

The department even procured an unmanned drone equipped with a video camera to take aerial shots, but the plan was halted after residents raised concerns about privacy.

“Where technology has caught up is to allow police to ask questions about what’s happening in their jurisdictions and to be able to understand temporal patterns, spatial patterns,” said Joel M. Caplan, assistant professor of criminal justice at Rutgers University in New Jersey, who has developed a crime forecasting program with his colleagues and is testing it at a half-dozen police agencies.

Critics say that surveillance technologies carry risks to civil liberties and that predictive policing, in particular, can additionally perpetuate a self-fulfilling cycle of bias. That is to say, an area with historically high rates of crime gets greater police attention, which results in more arrests, which in turn the algorithm uses to deem that neighborhood an area where crime is more likely to occur.

“It comes with inherent biases and prejudices that can be worse than the help it offers,” said Jason M. Schultz, who studies the privacy implications of new technologies at the School of Law at the University of California, Berkeley. “It kind of reinforces its own data by redirecting resources to those areas.”

Because predictive policing is still in its infancy, there is little independent evidence to suggest whether data mining in law enforcement is really effective.

But for companies that make the analytics tools, it is already a large and fast-growing market.

For instance, I.B.M., which has spent billions of dollars in the last year acquiring analytics start-ups, offers a software package of its own for predictive policing. Its social media analytics tool can be used by police agencies to vet Facebook and Twitter chatter. And its Coplink software allows one investigating police agency to mine another’s crime data to track down a wanted person: police officials in Mesa, Ariz., said they used this technology to find a man suspected of murder who was picked up for jaywalking hundreds of miles away.

Microsoft teamed up with the New York Police Department to develop one of the most ambitious crime analytics tools in the world. Known as the Domain Awareness System, it can analyze video from the more than 3,000 police surveillance cameras across the city and comb through a variety of other databases, from license-plate readers to sensors that can pick up heightened radiation levels to arrest records. In announcing the program last August, the police said it could, for instance, find a car associated with a possible crime or lawbreaker and analyze where that car had been seen over the last several weeks.

Law enforcement agencies, especially gang units, mine social networks like Facebook and Twitter to gather intelligence on criminal networks. The Department of Homeland Security uses an analytics tool that combs through Twitter for key words that might signal trouble. The list includes “pipe bomb,” “plume” and “listeria.”

Seattle’s predictive policing program grew out of an earlier experiment in data analytics. It began with a software program that culled 911 calls from the previous week, identified crime “hot spots” and assigned officers to patrol each one for at least 15 minutes every hour. Not all 911 calls turn out to be about criminal activity, though, and the department cut back on the practice of these so-called directed patrols. Earlier this year, Seattle bought a new software package, known as PredPol, short for predictive policing, and developed by academics at the University of California, Los Angeles. It initially deployed the software in two precincts, and in May announced its expansion citywide.

Acting Lt. Bryan Grenon said it was too early to know whether it was working. In any case, property crimes have plummeted in Seattle in the last year, by as much as 19 percent.

PredPol, according to P. Jeffrey Brantingham, a U.C.L.A. anthropologist who helped develop it, does not rely just on the previous day or previous week’s incidents. It looks at both short-term and long-term patterns in property crimes, he said, and uses a proprietary algorithm to weigh their relative importance and forecast risk in the next few hours. It has been used by 11 other police agencies, including those in Los Angeles and Santa Cruz. It has not been independently tested.

Sometimes, his algorithm has a pretty good idea of where someone might cause trouble, but not necessarily what kind.

In late May, Sergeant Clenna recalled, the PredPol map divined danger in a tiny patch of downtown that is dotted with liquor stores and loiterers. Sure enough, there was a robbery right there, though not one that a computer program could have predicted: a thief walked into a Chinese restaurant and made off with a live crab.