Chicago is a leader in using predictive policing, the use of data and algorithms to inform its overall strategy and on-the-ground decision making. One such tool that’s been used since 2012 is the Strategic Subject List, a computer model that purports to identify which Chicagoans are most likely to be involved in violence, either as a victim or perpetrator.
The idea of the SSL, city officials say, is to use data and technology to solve some of the city’s most intractable problems. Gun violence is one of them—this July saw 76 gun deaths, putting the city on pace to surpass last year’s homicide number. The official known uses of the SSL, which assigns individuals a score from 1 to 500, are to connect individuals to social services and to serve as an “investigative resource” for police.
If all this sounds vague, it’s because there’s very little information available about how police are using the list. Asked to provide details which social services are offered and how, or how the SSL might assist in investigations, city and police officials either declined to comment or referred reporters to other people who could not answer. In a Chicago magazine analysis, official police documents contradict multiple claims made by city officials, and some officials contradicted one another or the little public data available.
The Chicago Police Department fought to keep the list secret until this summer, when it lost a lengthy legal dispute with the Sun-Times and was forced to release a version of the database based on arrest records from August 2012 to July 2016. Among the main takeaways: only 3.6 percent of people on the SSL were a “party to violence,” meaning they have been involved in a shooting or murder as either the victim or offender.
So far only one independent analysis of the SSL has been undertaken; in a study published earlier this year, the RAND Corporation said the program had little impact on violence. At the time, police officials said the version of the SSL used by police had changed significantly since the 2013 version analyzed by RAND, but did not provide details on how.
This secrecy is one reason the tool has made national headlines and drawn comparisons to the dystopian Tom Cruise film Minority Report, where police use technology to predict crimes and arrest people before they commit them. “This is basically government decision-making turned over to an algorithm without any transparency about it,” says Karen Sheley, director of the American Civil Liberties Union of Illinois’ Police Practices Project.
In the past, police Superintendent Eddie Johnson has said the SSL has helped the CPD become “very good at predicting who will be the perpetrators or victims of gun violence,” and officials have doubled down on saying the SSL is only a risk assessment tool.
Chicago magazine delved into five findings about the list.
1. Police say they are not using the list to question or arrest people, but official documents show otherwise
Frank Giancamilli, CPD spokesperson, said in an email response on August 4 that the list was “simply a tool that calculates risk” and “is not used for enforcement and does not establish probable cause for arrest or even questioning.”
But documents show it will have a significant impact on police actions on the ground. In a contract approved this past February to increase CPD’s use of data-driven strategies in enforcement, “predictive analytics and SSL” are described as the driver of a planned “total overhaul” of mission assignments for a pilot program in two police districts. That could include, for example, district sergeants deciding which areas receive specialized units or rapid response cars.
Even before this contract was signed, the department used multiple press releases to highlight arrests of people on the SSL, a position that is contrary to statements that the list is used to help rehabilitate people, discourage them from criminal activity, and connect them with social services. The press releases regularly list the number of people arrested, nature of the charges, and number of arrestees with SSL scores. While the department has insisted the list is not used in enforcement, it denied a public records request from the Sun-Times for the list, claiming “criminals could still use the list to ‘thwart’ the police.” How criminals would “thwart” police in their use of a tool they have stated is only for social service outreach was not made clear.
Since late June, the Chicago Police Department and a city of Chicago spokesperson did not respond to repeated email requests for comments, as well as several phone calls, on how they will deploy resources based on the SSL.
2. Data suggests that more people on the SSL are being arrested than approached for social services
Social service outreach based on the SSL is managed by the Custom Notifications program at John Jay College. The program is under the umbrella of the Chicago Violence Reduction Strategy, a violence intervention model much like that used by CeaseFire (CeaseFire itself, and its parent group Cure Violence, are not affiliated with the SSL), which employs four staff members from John Jay College to apply the group’s intervention framework on the ground in Chicago.
According to executive director of the Chicago Violence Reduction Strategy program, Chris Mallette, the program sends out police officers, community members, and personal friends to convince an individual (typically gang-affiliated) that they are risking their life or their freedom by being involved in violence. “The primary goal is getting to who are the hot people right now, and who do we think is going to be hot in the future,” he says, noting that the SSL is only part of that equation, but that “everybody we talk to has a decent or high Strategic Subject List score.”
In a Department of Justice report describing strategies for custom notifications, Chicago’s program is cited as an example for using social network analysis to determine the “impact players” (people most responsible for driving violence) as targets for intervention. However, the RAND evaluation of the program found this strategy ineffective. In a response to the RAND evaluation, CPD stated that the SSL model had changed, moving away from looking at social networks to only looking at an individual’s interactions with law enforcement.
Regardless, police have declined to release data on how many people are involved in the Custom Notifications program. A Freedom of Information request from Freddy Martinez, a data activist who runs the nonprofit Lucy Parsons Labs, revealed that in 2016, 1,024 notifications were attempted, 558 were completed, and only 26 people attended a call-in, where police officers, social workers, and others offer support services. (Each of the notification attempts, says Mallette, could involve a visit to someone’s house. A completed notification, meanwhile, would include a face-to-face meeting.) The department makes it clear that for those who have interacted with the Custom Notifications program and are later charged with a crime, “the highest possible charges will be pursued.”
To put this in perspective, CPD has stated that 280 individuals with SSL scores were arrested in four gang raids during a six-month span in 2016.
3. 56 percent of black men in the city ages 20 to 29 have an SSL score
The newest data release about the SSL showed that nearly 400,000 people were assigned scores on the list, using primarily eight variables: age, the number of times an individual has been shot, the number of times they were the victim of assault or battery, gang affiliation, arrests for violent offenses, drugs, or weapons, and whether someone is getting arrested more frequently over time.
(While gang affiliation was a factor in the original data released by the Sun-Times in May, the mayor's office said that the most recent version as of this summer apparently does not. On the other hand, in August, CPD confirmed via email to Chicago that the gang database was one of eight factors used in the algorithm.)
The Chicago Police Department has stated that the inputs don’t include race, gender or location, and that the resulting scores do not overestimate or underestimate risk for any specific demographics. Still, the outputs show definite patterns. The majority of black men in Chicago ages 20 to 29 have an SSL score, meaning that 56 percent of them are considered at risk for taking part in violence in some way.
Kade Crockford, director of the Technology for Liberty Program at the American Civil Liberties Union of Massachusetts and a regular blogger about policing and technology, says the ACLU is skeptical of police claims that the list is unbiased. “In Chicago, like other large metro areas in the United States, police have focused extra special attention on young black men,” says Crockford. “The danger with the addition of technology is that some people think because a computer has told you to profile a bunch of young black men, this is race-neutral and couldn’t be a racially discriminatory program.”
Crockford says that computer programs and algorithms are as biased as the people who create them, citing examples like facial recognition programs created by white programmers that can’t recognize black people and how Facebook algorithms that determine hate speech favor white men over black children.
The city has declined multiple requests to clarify what happens to people who are placed on the list or have a high SSL number, though documents referenced above indicate that more police may be deployed to areas where individuals have high SSL scores.
4. The list is based on arrests—not convictions
The list uses the four most recent years of CPD arrest records—rather than convictions—as an input for the SSL, meaning people may end up on the list for crimes they have not committed.
Convicted in Cook, a project of Smart Chicago Collaborative’s Civic Works Project, reviewed felony cases filed in Cook County Court between 2005 and 2009 and found that about 6 percent of the felony cases filed never ended in a conviction. The number is relatively low but still significant. Additionally, nationwide about 90 percent of felony convictions result from a plea deal, a matter of contention among criminal justice reformers.
In addition, Crockford and other activists are concerned that individuals who have served their time will continue to be under suspicion. “Even when people have finished their sentences and served their time,” says Crockford, the list shows “you can never really escape the collateral consequences. It makes it really difficult to imagine how someone can escape from the criminal system.”
We also know that there is racial bias in arrest rates. For example, while white and black people do drugs at a similar rate, the latter are arrested and incarcerated at much higher rates for drug-related offenses, even in Chicago where politicians have attempted to curb low-level drug arrests.
5. Arrests are focused in already heavily policed areas
The general pattern of enforcement in Chicago is that more officers patrol the city’s minority and heavily low-income South and West side than other areas of the city.
While geographic information may not be directly included in the Strategic Subject List, the data shows that the most recent arrests of people with an SSL score map directly on the areas that are heavily policed, as evidenced through “contact cards,” the forms that police fill out after a street stop.
That creates a troubling cycle: Are police using SSL scores to determine where officers are assigned, which then leads to more arrests and higher SSL scores? Without more information from the police department about how police use the data, we can’t know.
As predictive policing has gained traction around the country, Chicago remains at the forefront. The SSL is a complex example of what happens when data is used to evaluate human behavior, and whether the public can determine how police use that information.
For Miles Wernick, the head of the team of researchers based at the Illinois Institute of Technology who created the algorithm on which the SSL is based, it’s important that people understand that the initial interest of the list was in social service outreach more than in any punitive measures. “It’s really critical that when people use these sort of tools they use them in ways that are appropriate,” says Wernick. “It should never be used to arrest people, harass people, or take any sort of punitive actions based on some computer algorithm.”