November 20, 2015 by

Taking the Edge Off Of Predictive Policing

PolicingProjectIcon_DataOver at the New York Times Room for Debate, experts discussed whether “predictive policing”—the use of data to target possible violent offenders before they act—works, and if it can be done without infringing on civil liberties. These are important discussions. Policymakers and police chiefs alike might take note of how Richmond, California has addressed some of those concerns—by combining carrots rather than sticks with the predictive techniques.

The Room for Debate conversation follows the recent Times report on the trend of police departments turning to black-box computer programs that crunch data—including information about friendships, social media activity, and drug use—to identify individuals deemed at a high risk of committing crimes. Police then seek to dissuade them from engaging in criminal activity by engaging in heightened surveillance or threatened increased punishment for future crimes. For example, as the Times detailed, the Kansas City police called a meeting of individuals who were not accused of any crimes who had been identified by their algorithm, and warned them that they would be targeted with the harshest punishments possible for even petty future offenses. (Yes, you last saw this technology in the film Minority Report.)

These predictive policing programs are constitutionally suspect. The Constitution requires the government to provide notice and an opportunity to contest actions that can result in an individual’s deprivation of freedom. But these protections are altogether lacking when it comes to predictive policing. Targeted individuals are identified by proprietary algorithms based on undisclosed variables weighted in unknown ways, and may ensnare individuals based on their race, neighborhood, or friends. The algorithms are created by private companies, so their accuracy is never tested. And yet, being added to one of these lists negatively affects a person’s interaction with the criminal justice system, and there is no opportunity for those selected to contest—either formally or informally—their inclusion.

In San Diego, for example, citizens have been arrested and charged as gang members because of the neighborhood in which they live. Chicago, where police claim their program will “become a national best practice” and have received more than $2 million to test the program, has used the program to justify surprise drop-ins from the cops to inform individuals they’re being watched. Meanwhile, Memphis floods neighborhoods with both marked and undercover police cars while increasing traffic stops where their system predicts an increased likelihood of crime.

In addition to the constitutional concerns, creating a class of marked citizens who receive special police attention stands in stark contrast to the President’s Task Force on 21st Century Policing’s recommendation to build trust and legitimacy through procedural justice and transparency while creating community based partnerships to reduce crime. And the data’s value is dependent on the wisdom of the people doing the interpreting. Larry Samuels, CEO of PredPol, the maker of predictive policing software sold to police departments around the country, warns that predictions will vary in accuracy, and are only as good as the officers using them.

At least one city—Richmond, Ca.—has found a way to employ these new technologies in a way that takes some of the edge off the due process difficulty, and is experiencing success along the way. Rather than relying on threats, Richmond’s system is based on rewarding at-risk individuals for achieving personal and professional goals.

In 2007, Richmond’s City Council launched the Office of Neighborhood Safety (ONS). The office operates completely independently from the Richmond Police Department, which has no access to the program’s list of at-risk individuals. ONS targets the 50 Richmond citizens that its algorithm identifies as most likely to be involved in gun violence and offers them a spot in a support program that includes a stipend of between $300 and $1,000 per month to avoid dangerous behavior. The program provides a road map, with concrete goals for progress, as well as assistance in taking steps towards increased stability such as obtaining a GED or driver’s license. ONS offers a path to safety instead of a threat of increased police attention. The program has coincided with a reduction in violent crime in Richmond, where homicides have fallen from 47 in 2007 to 16 in 2013. This 66% decrease far outpaced the 16% decline in murders across the United States over the same period, according to FBI data.

Police departments have choices to make in how to reduce crime most effectively. While new technology may offer insights into which citizens are at risk of violating the law, these systems are in their infancy and pose serious problems of accuracy and invasion of privacy. At the least, departments can follow Richmond’s lead in using the systems to implement positive incentives, rather than employing threats.