How the Manhattan DA’s Use of Big Data Targeting Risks Changing the Rules of Prosecution
A book excerpt from The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement (NYU Press 2017)
In downtown Manhattan, an experimental prosecution unit has begun rethinking how to reduce violent crime. Under the leadership of district attorney Cyrus Vance Jr., the Manhattan District Attorney’s Office created the Crime Strategies Unit (CSU) to target the bad apples in communities and take them out by any means necessary. Call it the “Al Capone” approach to crime, only the targets are young men suspected of violence, not national mob bosses. Dubbed “intelligence-driven prosecution,” police, prosecutors, and analysts target individuals for incapacitation and thus removal from problem areas of the city.
Analytical and aggressive, CSU prosecutors build cases against the primary crime drivers in a neighborhood. First, crime data allows prosecutors to isolate high-violence areas for scrutiny. By crunching police crime data and mapping neighborhoods, prosecutors identify particular hot spots of violence. These areas become known as “Bureau Based Projects” (BBP). A small team of prosecutors oversees each BBP and coordinates intelligence gathering in the area. These prosecutors work closely with the main CSU staff and may or may not take the cases that result from NYPD making arrests in the hot-spot areas. A “violence timeline” is created for each area, highlighting the past pattern of violence between groups, gangs, and individuals. The timeline lists details of each shooting with suspects, victims, and facts, along with time, location, and date.
Second, particular individuals are identified for police attention. Each BBP selects ten or so “priority targets” — “people whose incapacitation by the criminal justice system would have a positive impact on the community’s safety and/or quality of life.” Field intelligence officers, detectives, and patrol officers help identify the priority targets for removal. These individuals have at least five criminal convictions and a history of violence. Some have been uncooperative victims of past shootings. Others are associated with gangs or criminal groups. A “target tracker” of each young man populates the data system with photos, prior criminal history, and other personal information. These individuals become the targets. Like Al Capone, who eventually faced prosecution for tax-evasion charges rather than the more violent crimes he engaged in, these priority targets do not have outstanding arrest warrants and cannot be arrested based on existing evidence.
Prosecutors input the names of priority targets into an “arrest alert system.” This arrest alert system then allows prosecutors to know if a target has been arrested. Routine fingerprinting keyed to a person’s criminal history (rap sheet) triggers the alert. Under the old system, if a target got arrested for scalping Broadway tickets or simple assault (or some other minor offense), line prosecutors would have no way of knowing the level of threat posed by the individual. Now, alerts (usually an email) inform prosecutors throughout the office that a wanted target has entered the criminal justice system. The arrest alert system triggers a process whereby all the power of the prosecutors’ office can be used to incapacitate the individual. Enhanced bail applications can be used to argue for pretrial detention. Additional charges can be added to ratchet up pressure to plead guilty. Harsher sentencing recommendations can be sought to increase punishment. Even after sentencing, prosecutors are alerted to a defendant’s release, so that the Manhattan parole system can monitor reentry back into society.
Data sharing also allows a more comprehensive intelligence-gathering operation. A new data system allows over 400 prosecutors to prosecute 85,000 cases a year. Information about cases, suspects, neighborhoods, witnesses, gangs, nicknames, rivalries, crimes, tips, and a host of other data is coordinated through shared searchable databases. Prosecutors debrief individuals in the arrest alert system, looking for more information about networks of violence. Photos of criminal associates, social media postings, and other tips become part of the data-collection system. NYPD police commissioner William Bratton called it a “seamless web” of shared data between police and prosecutors and termed the partnership one of “extreme collaboration.” Inspired by the sabermetrics approach to baseball and finance, Cy Vance Jr. likened it to a “Moneyball” approach to crime fighting.
On a few occasions, this person-based targeting has led to large-scale arrests and prosecutions. Utilizing the intelligence-driven prosecution platform, the Manhattan District Attorney’s Office has prosecuted several violent gangs in New York City. In one case, the District Attorney’s Office in collaboration with the NYPD studied violence patterns, gang activity, and even social media before indicting 103 members of local crews. In 2014, the prosecution of these youth gangs for homicides and shootings in West Harlem stood as the largest gang conspiracy indictment in New York City history.
New York City has seen record low crime levels both before and after implementation of the intelligence-driven prosecution methods. Violence rates remain low in the targeted microareas, and shootings have dramatically declined. As a result of the initial success in Manhattan, the concept of intelligence-driven prosecution is being replicated across the country. In Baltimore, San Francisco, Philadelphia, Richmond, and Baton Rouge, intelligence-driven prosecution is using data to target the bad apples for removal from society.
The idea behind person-based targeting is both old and new. Police have always known the bad apples in a community. Prosecutors have regularly targeted them. Yet a policing philosophy that uses data and predictive analytics to prioritize the crime drivers in a society signifies a new approach. [Three] main changes emerge from these technologies — insights that will shape the future of who gets targeted.
[P]roactively targeting violent social networks will change how local police respond to crime. Traditionally, local police might react to calls for service, rely on observations on patrol, or respond to community complaints. With person-based predictive targeting, police can instead target suspects for surveillance or deterrence before needing to respond to a call. For local prosecutors, this represents a significant change. As the former head of the Manhattan Criminal Strategies Unit stated, “It used to be we only went where the cases took us. Now, we can build cases around specific crime problems that communities are grappling with.” Big data policing makes police more proactive. In many ways, intelligence-driven prosecution and policing at the local level are really just mirroring some of the approaches federal investigators and federal prosecutors have used for years. While the FBI and U.S. Attorneys regularly investigate completed crimes, they also focus on surveillance and investigation of criminal networks to prevent or disrupt future crime. For local police, the study of gang networks means a similar change from reactive policing to proactive policing.
[M]oving from traditional policing to intelligence-led policing creates data-quality risks that need to be systemically addressed. Intelligence-driven systems work off many bits of local intelligence. Tips, crime statistics, cooperating witnesses, nicknames, and detective notes can get aggregated into a large working data system. Yet the quality of the data is not uniform. Some tips are accurate, and some are not. Some biases will generate suspicion, and some informants will just be wrong. An intelligence-driven policing or prosecution system that does not account for the varying reliability and credibility of sources and just lumps them all together in the name of data collection will ultimately fail. Just as national security intelligence agencies have layers of intelligence analysts to examine incoming information, so police departments must develop similar structures to vet this intelligence-like data. Blind data collection without information about sources, reliability, or testability will result in an error-filled database. Systems must be designed — before adopting data-driven technologies — to source, catalogue, and make the information useful for officers. Especially when these systems are used to target citizens for arrest or prosecution, the quality-control measures of black-box algorithms must be strong.
Other data-integrity concerns may arise when detectives, gang experts, or police intelligence officers control the target lists. While these professionals have close connections to the community and valuable knowledge of local gangs and potential targets, the ability for risk scores to be manipulated by police interested in prosecuting certain individuals opens up questions of the objectivity and fairness of the lists. If a gang detective can put someone on the list and there is no process to change or challenge the list, the system could be abused. If there is one thing that has been demonstrated regularly with the proliferation of gang databases, these lists are rife with errors. After all, without formal criteria to be a member of a gang, rumor, assumptions, or suspicion can be enough to be labeled part of a gang and thus result in an elevated risk score. Worse, there is usually no easy way to get off the list, despite the fact that circumstances change, time passes, and the data grows stale.
Finally, big data policing may distort traditional roles of prosecutors and police officers. Prosecutors seeking to incapacitate individuals on their “primary target” list can bump against ethical lines. During a training on intelligence-driven prosecution, one supervising prosecutor spoke of a case involving a young man (a primary target) running toward a street fight holding a lock in a bandana. While the man was likely up to no good, holding a lock in a bandana is not necessarily a crime. But prosecutors chose to charge the man with carrying a dangerous weapon with the intent to use it (despite equivocal evidence of intent). Such a serious criminal charge might not have been pursued if the suspect had not been on the primary target list and may not even be supported by the facts. But when incapacitation is the goal, the prosecutor’s power to use charging, sentencing, and bail determinations aggressively can distort the traditional focus of the prosecutor. Such a distortion is not necessarily bad. If the prosecutor was correct and the primary target was a violent risk to the community, maybe such aggressive predictive prosecution makes good sense. But if this type of human targeting is inaccurate and if it is misused or even if it is unchecked, it can be damaging to the perception of fairness in the justice system.
Excerpted from The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement (NYU Press 2017).