5

Community Responses to Predictive Policing

Andrea Headley, Georgetown University and planning committee member, moderated a roundtable discussion that explored community perspectives on predictive policing approaches. Introducing the panel, she explained that the committee developed this session based on the executive order that initiated the workshop, which highlights the impacts of community critiques of policing approaches as well as challenges around community trust in law enforcement. Earlier, David Weisburd recalled the impact of community critiques, specifically on law enforcement use of predictive policing approaches, and noted the importance of establishing a shared foundation of the history of community response to predictive policing to situate the workshop. Key messages from speakers included the following:

• Previous applications of predictive policing approaches in New York and Chicago raised communities’ concerns about disparate racial impact. (Martinez, Rahman)
• Application of predictive policing approaches involves political decisions about allocation of resources. (Martinez, Scott)
• The perceived objectivity typically associated with big data and algorithmic predictions could contribute to the expansion of police power or be used to justify violations of civil rights or civil liberties. (Rahman, Scott)
• Predictive policing technologies can reinforce assumptions that certain locations and certain people are inherently dangerous. These labels could influence law enforcement behavior in danger-labeled places/with danger-labeled people. (Rahman, Scott)

• Community members, particularly those living in high-crime areas, are often uniquely well positioned to evaluate predictive policing approaches. Their lived experiences could identify and remedy potential shortcomings of predictive policing systems. (Rahman, Martinez)
• Technology alone cannot rebuild trust in policing, and predictive policing alone does not address deeper political and historical issues affecting trust. (Rahman)
• Improving public trust in new technologies and approaches requires critical evaluation, transparency, community input, and proof of non-discrimination, all prior to implementation. (Scott)

COMMUNITY DEFINITIONS OF PREDICTIVE POLICING

Headley invited panelists to share their definitions of predictive policing. Jeramie Scott, Electronic Privacy Information Center, gave the simple definition of “an algorithm analyzing data in order to predict where a crime will happen or who will commit a crime [or] be a victim of a crime.” This definition, noted Scott, can include a wide range of technologies. Freddy Martinez, Lucy Parsons Labs, said that while it is possible to give a technical definition of predictive policing, it is also important to recognize the connections between modern-day predictive tools and less technological historical counterparts. Historical approaches rely on the foundational assumption that dangerous individuals or locations exist and need to be monitored, suggested Martinez. Decisions about what and who to monitor are political ones, he said.

Shakeer Rahman, Stop LAPD Spying Coalition, expressed his opinion that predictive policing is more of a “branding term” than a specific technology, similar to “community policing” or “smart policing.” The exact definition of predictive policing is less important than the way the term has been used to justify allocation of resources to the police, to deflect criticism, and to engage scientists, academics, and industry in the institution of policing. While Rahman noted that the term “predictive policing” is falling out of favor, he suggested that similar methodologies and systems—particularly the use of mass data algorithms and machine learning—continue to play a part in everyday policing. Predictive policing is often promoted as a tool for police effectiveness, Rahman said, but he expressed concern that use of such tools expands police resources, discretion, and power, and can enhance the legitimacy of conduct that would be viewed as troubling if it were not associated with advanced technology and big data.

COMMUNITY EFFORTS AND CONCERNS

Communities have been at the forefront of bringing concerns about predictive policing programs to light, said Headley. She asked panelists to share their experiences with community organizing around law enforcement use of predictive policing tools. Rahman told participants about the Stop LAPD Spying Coalition, which organized to oppose what might now be considered the first-generation predictive policing programs developed by the Los Angeles Police Department (LAPD). He noted that one of these programs, Los Angeles Strategic Extraction and Restoration (Operation LASER), was developed using federal funding and included both place-based and person-based approaches. Stop LAPD Spying Coalition used public records research and collective study to document and uncover the impact of these programs and found that almost half of the individuals identified as chronic offenders were Black (Stop LAPD Spying Coalition, n.d.). A 2019 audit conducted by the Office of Inspector General concluded that of the 637 people in the LAPD “chronic offender” database, 112 had zero points under LAPD criteria (Office of Inspector General, 2019). Significantly, the Stop LAPD Spying Coalition campaign was led by nonprofessional community members, Rahman said. This defies the common assumption that innovations such as predictive policing are too complex to be understood and addressed by community members, suggested Rahman. To the contrary, it is vital to develop expertise around these programs within the most impacted communities.

Martinez discussed his work at Lucy Parsons Labs, which focused on the Chicago Police Department’s (CPD’s) use of person-based predictive policing. He explained that CPD used person-based predictive policing to create a list of the “hundred most violent offenders in Chicago,” referred to as the Strategic Subject List (SSL). Through public record requests and analysis, Martinez and colleagues found that the SSL eventually included almost half of all young Black men in the city (Kunichoff & Sier, 2017). Martinez highlighted concerns with implementation of the SSL, including processes for notifying listed individuals, policies related to determining the severity of charges for individuals who committed crimes after being notified, and the potential for unintended consequences of police contact with notified individuals. Eventually, the SSL was disbanded. Martinez agreed with Rahman that community members are capable of informed evaluation of predictive policing, noting that his work engages many individuals who are experts because of their lived experiences. Discussions about predictive policing often focus on the need to protect people who live in high-crime neighborhoods, said Martinez, but these are often the same people who are raising informed and nuanced critiques of predictive policing programs.

Scott provided an overview of major community concerns related to predictive policing. A key concern in communities, he said, is that focusing on specific crime prevention or response technologies can cause people to lose sight of foundational solutions to crime. He expressed concern that predictive policing could be used as a “band-aid” to address inadequate social policies. Predictive policing also raises concerns about the expanding nature of surveillance and data collection. Scott suggested that technologies relying on data will likely engender arguments for collecting more data to improve the system. “Where do we draw the line?” he asked, when collecting and analyzing increasing amounts of data on members of the public. Scott noted that expanding data collection can also draw resources away from programs that could potentially make a long-term difference in public safety and community well-being.

Communities are also concerned that predictive policing could lead to unnecessary interactions with law enforcement and could justify interactions that might otherwise raise questions about civil rights or liberties violations, Scott noted. In addition, he expressed concern that if an algorithm informs police that a person or place is dangerous, this label could influence police response. For example, said Scott, when police use a technology that purports to detect gunshots, they show up to the scene prepared for gunfire. However, predictive policing tools can be inaccurate and could direct police to enter a situation under the incorrect assumption that gunfire has occurred. Similarly, a tool telling police that an individual may be armed or dangerous could influence the way police interact with that individual. Finally, said Scott, predictive policing could be used to label young people as crime committers. The criminal legal system acknowledges that young people who commit crimes typically desist as they age, but predictive policing algorithms may not account for the potential for young people to cease criminal activity as they mature. Thus, labels could influence the way police interact with young people and could undermine a young person’s ability to reform.

Scott encouraged decision makers to think broadly about the costs and benefits of predictive policing technologies. It is the role of the larger community, said Scott, to engage in broader thinking, evaluation, and decision making.

PUBLIC TRUST

Given the historical context of policing, Headley asked panelists to discuss how taking public trust seriously could shape decisions about the use of predictive policing technologies. Rahman expressed his perspective that certain communities historically lack trust in the institution of policing, and technology alone cannot change that. From his perspective, science-

and technology-based tools created for policing have tended to increase the resources devoted to policing and expand the institution of policing rather than making policing more fair or less oppressive. Noting the history of law enforcement’s involvement in racial violence in the United States, Rahman argued that this history makes it difficult to strike an objective balance between the costs and benefits of predictive policing technologies. Questions about such costs and benefits are political questions without objective, scientific answers. It is a mistake to try to resolve such questions through technocratic expertise, said Rahman, instead of attempting to understand them in their full context.

Scott agreed that improving public trust is challenging but said that options exist to increase the accountability and transparency of policing. Public input, welcomed at all stages of technology development, could allow community members to share concerns, perspectives, and questions before a technology is deployed, he said. Moreover, key questions remain unanswered, said Scott, such as how data will be collected and secured, and how potential expansion and “mission creep” will be curtailed. Furthermore, positive proof is critical for demonstrating that new approaches are nondiscriminatory before implementation. Martinez expressed his perspective that little meaningful oversight of law enforcement use of predictive policing technologies exists currently. Scott noted that the actions he described will not immediately result in public trust, but they could increase transparency and accountability, and potentially prevent the use of discriminatory technologies.

ALTERNATIVES TO PREDICTIVE POLICING

If predictive policing is not a good tool for ensuring public safety, asked Headley, what are the alternatives? How should resources be allocated? Rahman argued for investing in solutions that make communities stronger and safer, such as housing or education, noting that these are political choices. Martinez agreed that provision of resources could address the root causes of crime, noting that in Chicago, the same communities classified as high-crime areas are impacted by lead in water, poverty, and closed schools—“it is always the same map,” he said. Emphasizing that non-law-enforcement actions could impact crime, Martinez noted that allocating resources to address root causes requires political will. Scott added that technological solutions are often perceived as easier and quicker to implement than long-term efforts to address root causes of crime. Achieving a society with drastically lowered crime and violence necessitates addressing disparities in wealth, education, mental health, healthcare, and housing in a meaningful and long-term way, he said, noting that every other approach is “a band-aid.”

A workshop participant asked panelists about whether predictive policing tools could enable police to perform non-traditional police work, such as connecting people with resources or improving public spaces. Rahman replied that while investments in the community are needed, police may not be the appropriate institution for delivery of these resources. In disenfranchised communities, police tend to be the “starting point” when considering solutions to community problems, though they might not be well trained or equipped for such work, or even responsible for addressing such problems. Their law enforcement roles may even deter those in high-crime communities from accepting police support, Rahman said.

FUTURE PREDICTIVE POLICING

Andrew Ferguson asked panelists to consider likely future iterations of predictive policing. Specifically, he asked panelists what they would ask policymakers to consider when making decisions about whether to implement new predictive policing tools. Rahman noted that efforts are being made to consolidate predictive policing technologies, moving police departments toward centralized law enforcement technology data platforms. Future platforms could include prediction technologies along with standard technologies such as body cameras and gunshot detection tools. Rahman emphasized the value of broad community engagement around policing, to address the system comprehensively rather than focusing on a single technology.

Scott offered specific examples of questions that he would like to see answered before future predictive policing technologies are implemented:

• What data are being used?
• Are these data considered historically biased?
• How can biases be addressed?
• Is the technology provably non-discriminatory?
• Have there been third-party audits of the technology?
• How accurate is the technology?
• Is there a better, simpler solution to the problem?

The roundtable discussion on community responses to predictive policing highlighted notable concerns about the potential for racial bias, expansion of police power, and reinforcement of harmful assumptions about certain locations and individuals. While predictive policing technologies are often promoted as tools for enhancing police effectiveness, the panelists emphasized the need for critical evaluation, transparency, community input, and addressing root causes of crime through investments in education, housing, and other social services rather than relying solely on technological solutions.