6

Moving Forward: Reflections on the Future of Predictive Policing

In the final panel discussion, speakers engaged in a moderated dialogue around the future of predictive policing. Session moderator David Weisburd asked panelists to address whether predictive policing is worth pursuing given the ethical, legal, and community concerns, and, if so, how the next generation of predictive policing might be envisioned. Responses are summarized beneath headers indicating the questions posed to speakers. Key messages from this session included the following:

• While there are valid concerns about predictive policing, dismissing data-driven policing approaches entirely would be a missed opportunity. Thoughtful deployment remains key. (Ludwig, Joh)
• Community trust remains paramount to effective policing, and the most effective policing is guided by community needs and priorities. (Bueermann)
• Potential risks and benefits of deploying predictive policing approaches largely depend on the specific policing response to a prediction. (Cummings)
• An independent body to evaluate costs and benefits of predictive technologies before widespread deployment would be beneficial. (Headley)
• Predictive tools are best understood as aiding, not replacing, human judgment. (Joh)
• A more effective future iteration of predictive policing could avoid issues seen in previous deployments, such as reliance on unreliable

• data sets, lack of transparency, and lack of accountability by developers and end users. (Abt)

IS THERE A PLACE FOR PREDICTIVE POLICING IN THE FUTURE OF POLICING?

Many fields—from medicine to education—have made great progress using data to drive practice, said Jens Ludwig, University of Chicago. While valid concerns exist regarding predictive policing, it would be a lost opportunity to completely dismiss the idea of data-driven predictive policing, Ludwig said. When evaluating the usefulness of data-driven policing as a tool, Elizabeth Joh emphasized the importance of considering it in comparison to alternative approaches. Policing has traditionally involved a great deal of discretion and human judgment, said Joh, so much could be gained by using data to make decisions. However, she said, it is important to recognize that data-driven predictive policing tools are designed to aid instead of substitute human decision making. Predictive tools require a carefully considered implementation plan, said Joh. Renée Cummings noted that the risks of predictive policing depend on the actions taken in response to predictions. For example, if place-based predictive policing results in more lighting or a cleaned-up park, risks of deployment are low; but if a prediction results in harassment or unconstitutional searches, risks are high—and high risk could be a key factor deterring decision maker support for these tools.

Thomas Abt argued for a place for predictive policing if it is broadly defined as an approach that involves increased use of science and analysis in law enforcement and overall criminal justice. Jim Bueermann argued that artificial intelligence (AI) and data-driven approaches are here to stay and predicted that while predictive policing tools are being renamed and repositioned, some forms of predictive approaches could soon become a standard law enforcement technology.

HOW CAN POLICING PRIORITIZE COMMUNITY TRUST?

Community trust is the number one issue in policing today, said Bueermann. The best policing, he suggested, is community led and identifies priorities based on community concerns. Community engagement poses challenges, Bueermann noted. Specifically, it can be more time-consuming than unilateral decision making, and police chiefs often serve multiple communities, which could have divergent opinions around the use of predictive policing approaches. Synthesizing differing perspectives and making practical decisions regarding subsequent actions are challenging for police leaders, he said. Understanding community concerns and making collective, trans-

parent decisions are fundamental to building and maintaining community trust in policing, Bueermann stated. Ideally, collective decisions to implement predictive policing technologies would result from collaborative conversations spanning issues of strategy, desired outcomes, and potential harms, and a plan for monitoring and evaluation would be established, said Bueermann.

Andrew Ferguson urged decision makers to listen to communities. Communities experiencing high crime rates were some of the first to identify problems with predictive policing, he said, noting the importance of listening to their concerns and perspectives before advancing new iterations of predictive policing approaches. Moving forward with these tools without forethought or community input, said Andrea Headley, could negatively impact community trust—not just trust in policing but in government more broadly. Communities are generally aware of their own problems and needs, she added, but lived experiences of community members are often undervalued as evidence.

WHAT DOES THE NEXT PHASE OF PREDICTIVE POLICING LOOK LIKE?

While the term “predictive policing” has fallen out of favor, “data-driven policing” remains popular, said Ferguson. He expressed concern that the same problems that existed with predictive policing could emerge in new forms with the use of AI, and he encouraged decision makers to carefully consider the inputs, interventions, and impacts of new programs. Philip Lukens, International Association of Chiefs of Police, and planning committee member, agreed that AI will likely be integrated into law enforcement technology, highlighting the importance of acknowledging and addressing previous potential harmful consequences now. Cummings added that successful, ethical uses of new predictive policing approaches necessitate that police officers and leaders both understand how data and algorithms are used and have the ability to identify and address potential ethical issues and unintended consequences. Ludwig listed three components he views as critical to new iterations of predictive policing: (a) a U.S. Food and Drug Administration (FDA)-like agency to evaluate costs and benefits of policing technologies, (b) a tool for measuring community sentiment, and (c) investments in human capital, leadership, and management to strengthen policing organizations. Abt advised that future predictive policing avoid the pitfalls of previous iterations, particularly the use of large, unreliable data sets in the absence of strong justifications and controls; lack of transparency; and lack of regulation. Future predictive policing approaches would be well served by embracing transparency and making information available to impacted individuals, as well as by careful evaluation and regulation of commercial products, Abt said.

Like Ferguson, Abt suggested that the term “predictive policing” is vague and outdated, has caused confusion, and is not widely used by practitioners. Bueermann also noted a shift away from the term and hypothesized that future predictive policing tools may be part of a broader ecosystem of police technology. He cautioned that as the law enforcement technology ecosystem grows and new terms are used, adequately educating police leaders, community leaders, and the public about the costs, benefits, and ethical and legal consequences of such tools could become increasingly challenging. Joh agreed that the future of predictive policing will likely involve the integration of predictive policing technologies into the everyday practice of policing, with police increasingly relying on the information or suggestions these systems generate. This could change the nature of policing, she said, and raise deeper questions about transparency and accountability.

WHAT ARE THE ROLES OF DECISION MAKERS AND CONSTITUENTS IN THE FUTURE OF PREDICTIVE POLICING?

Federal Government

Future funding for experiments in predictive policing could be tied to controls around error and racial justice harms, said Ferguson. Headley agreed that federal funds play a large role in the adoption of predictive policing technologies and added that innovations can spread quickly across law enforcement agencies. Another role for the federal government, said Weisburd, involves setting national standards for the use of predictive policing approaches. Speakers identified additional roles for the federal government. Abt noted the value of the federal government, including influencing practice and policy through grants, organizing convenings for shared learning, disseminating information about evidence-based practices, providing technical assistance, and supporting research and evaluation. Ferguson noted the possibility of requiring federal grantees to implement plans for evaluation, auditing, and controls.

Technology Developers

Public-private partnerships are often critical to innovation, said Cummings. However, said Joh, incentives and interests of private companies may not align with those of law enforcement or the public. Other challenges related to working with private companies, added Joh, include understanding and negotiating contracts, and the inability of law enforcement agencies to “shop around” for alternative products. Weisburd noted that some small police agencies may not have the resources or knowledge to hold informed, transparent conversations with commercial vendors about

products of interest. Bueermann agreed that it can be difficult for police chiefs to understand technical aspects of a contract, in part due to a lack of internal expertise in this area. The federal government could assist with this problem by developing model data use agreements, Bueermann pointed out.

Police Agencies

Bueermann noted that other agencies, such as fire departments, have internal officers tasked with monitoring internal practices. A similar role would benefit police agencies, he suggested—an employee who could ask whether a practice is effective, just, or potentially harmful. Such questions are not part of the socialization or education of police in this country, said Bueermann, and he expressed his opinion that without significant institutional changes, the existing culture of policing would win out over the best science where they conflict.

HOW WILL FUTURE PREDICTIVE POLICING BE EVALUATED?

Echoing other comments made throughout the workshop, Ludwig noted the lack of evaluation of predictive technologies as a major issue. Successful implementation should not be measured simply by reduction in crime, he said, but should include additional outcomes. Currently, Weisburd noted, policing programs are often implemented based on single studies; he agreed that rigorous evaluation of programs is critical, prior to widespread implementation. Headley reiterated an earlier point calling for independent analysis of a technology before deployment. Like drugs, predictive policing technologies could significantly impact individuals and communities, said Joh, yet an evaluation process to determine the costs and benefits of deployment is lacking. Ludwig echoed this suggestion, calling for an FDA-type agency to evaluate algorithms used by state actors, to prevent application of policies or programs that have a net negative impact. To conduct such evaluations, Ludwig noted the necessity of improved outcomes data, including data on community sentiment about policing. The federal government could help determine how to obtain high-quality, representative, high-frequency measures of community sentiment, said Ludwig, and Bueermann agreed with the importance of community-sentiment data.

REFLECTIONS AND KEY TAKEAWAYS

Based on workshop presentations and discussions, speakers and planning committee members closed by offering final reflections and takeaways regarding several key areas of concern identified throughout the workshop.

Community Engagement and Trust

Several speakers across the workshop emphasized the value of building and maintaining community trust. Weisburd suggested that community trust be a primary consideration in the deployment of predictive policing technologies. Building trust requires engaging with communities, being transparent about the uses of technologies, and ensuring that interventions are fair and just. Headley echoed the importance of considering community concerns and emphasized the potential for predictive policing to exacerbate harm. Headley and Kim Neal, National Association for Civilian Oversight of Law Enforcement and planning committee member, both underscored the ethical and social implications of predictive policing. Neal emphasized that predictive policing could decrease the human element of policing and further erode community trust, particularly in communities of color. To build and maintain community trust, Ferguson argued that place-based predictive policing is ideally implemented with democratic approval, clear ex ante policies, and meaningful audit mechanisms for community input.

Data Quality and Transparency

Data present a particular challenge for effective, equitable predictive policing, said Ferguson. Challenges are seen with methodology, transparency, accountability, security, and administration of data. The types of data used in predictive tools matter for both effectiveness and fairness, he noted. Ferguson argued that unbiased predictive policing tools would not utilize police-discovered crimes, quality-of-life crimes, or public disorder crimes as inputs, and would directly address the potential feedback loop that can occur when high-crime areas receive heightened police attention. Echoing Rashida Richardson’s presentation, Headley observed that predictive programs are not neutral or objective but instead reflect inequities that are “baked into” input data. Lukens suggested that data dashboards could be useful and could include mechanisms for accountability, transparency, and regular reporting. One benefit of data-driven policing, said Sarah Brayne, is that it gathers a significant amount of data about policing in general. These data can be used to measure the costs and impacts of policing, such as the average number of times residents in given communities are stopped by police. Formalizing such problems with data could be the first step toward resolving them, she said. However, these opportunities are only possible if independent researchers or oversight agencies can access the data, Brayne noted.

Legal Considerations

Predictive policing presents legal questions in several areas. Ferguson argued against the use of high-crime-area predictions as a factor in Fourth Amendment reasonable suspicion analysis or probable cause. If predictive policing is used in a discriminatory fashion, Lukens said, enforcement under Title VI is necessary. Headley suggested that predictive policing that lacks transparency threatens due process rights, and she noted the current absence of a legal mechanism for correcting errors or challenging findings of predictive policing approaches.

Interventions and Impacts

Ferguson also drew attention to the importance of the particular response to a predictive policing prediction—a recurring theme throughout the workshop. He emphasized the need for a holistic response to predictive policing outputs, including investment in non-policing interventions to address environmental risks. The predictive element of identifying geographic areas or people who are at risk is unproblematic, he clarified. Instead, the police response to such predictions has been inadequate for addressing structural problems and has invoked concerns around civil rights, civil liberties, and disparate impact. Drawing on his experience as a former police chief, Lukens proposed a layered approach to place-based predictive policing, which would involve identifying key constituents and decision makers, deploying a range of social services and non-profit partners, implementing environmental interventions in high-crime areas, and establishing law enforcement services and proactive crime prevention programs (see focused deterrence discussion in this proceedings for one example). “There will always be a need for police services,” he noted, “not as an occupying force but as part of the community, ensuring peace as a service.”

Oversight and Regulation

Lukens raised concerns about transparency and accountability in commercial contracts, which can prioritize proprietary interests over public accountability. Greater oversight and regulation of predictive policing technology vendors could be implemented, Lukens said, which could include ensuring that contracts are transparent and that commercial vendors are held accountable for the ethical use of their technologies. He also noted the value of additional training for police leaders, to enhance their ability to research and evaluate technologies, as well as legal support around contracting with commercial vendors. Several members of the planning committee highlighted the crucial role of the federal government in setting standards

and providing oversight. Weisburd echoed Abt’s calls for increased federal investment in research and evaluation to develop standards for the use of predictive policing technologies. Weisburd also called for federal agencies to play a more active role in ensuring that such technologies are used ethically and effectively.

Evidence of Effectiveness

Before adopting predictive policing tools, the reliability of the predictions and the effectiveness of the interventions need to be empirically tested in the field, said Ferguson. Lukens suggested that validation requirements for policing technologies be weighed against the urgency of public safety and imminency of threat. Headley urged decision makers to think about both the potential harms of the tools and the strength of evidence supporting their use. Research suggests that violence and crime can be reduced through efforts like providing mental health and substance abuse care and affordable housing—decisions about where to invest resources could benefit from taking such research into account. Weisburd echoed the statements of multiple speakers and planning committee members in calling for a regulatory body similar to the FDA, to ensure that predictive policing technologies are thoroughly tested for effectiveness and fairness before being implemented in communities.

CLOSING

Individual speakers emphasized that the future of predictive policing will depend on how the ethical, legal, and practical issues identified in the workshop are addressed. While challenges exist around engaging with communities, ensuring data quality and transparency, pairing predictions with appropriate interventions, and measuring the effectiveness of predictive policing programs, when considering whether data-driven policing is a useful tool, it is important consider the alternative. Workshop discussions provided valuable insight into ways that key decision makers and constituents could address these challenges and ensure that predictive policing technologies are used effectively, accurately, and fairly.