Predictive policing is sweeping the nation, promising the holy grail of policing--preventing crime before it happens. The technology has far outpaced any legal or political accountability and has largely escaped academic scrutiny. This article examines predictive policing's evolution with the goal of providing the first practical and theoretical critique of this new policing strategy. Building on insights from scholars who have addressed the rise of risk assessment throughout the criminal justice system, this article provides an analytical framework to police new predictive technologies.
TABLE OF CONTENTS INTRODUCTION I. PREDICTION AND THE CRIMINAL JUSTICE SYSTEM A. A Brief History of Actuarial Justice B. The Prevalence of Prediction in the Criminal Justice System II. THE EVOLUTION OF PREDICTIVE POLICING A. Predictive Policing 1.0: Targeting Places of Property Crime B. Predictive Policing 2.0: Targeting Places of Violent Crime. C. Predictive Policing 3.0: Targeting Persons Involved in Criminal Activity D. Reflections on New Versions of Predictive Policing III. POLICING PREDICTION A. Data: Vulnerabilities and Responses 1. Bad Data a. Human Error b. Fragmented and Biased Data 2. Data: Responses a. Acknowledging Error b. Catching & Correcting Error c. Training and Technology B. Methodology: Vulnerabilities and Responses 1. Methodological Vulnerabilities a. Internal Validity b. External Validity--Overgeneralization c. Error Rates 2. Methodological Responses C. Social Science: Vulnerabilities and Responses 1. Social Science: Vulnerabilities 2. Scientific Studies: Responses D. Transparency: Vulnerabilities and Responses 1. Transparency: Vulnerabilities 2. Transparency: Responses E. Accountability: Vulnerabilities and Responses 1. Accountability: Vulnerabilities 2. Accountability: Responses F. Practical Implementation: Vulnerabilities and Responses 1. Practical Implementation: Vulnerabilities 2. Practical Implementation: Responses G. Administration: Vulnerabilities and Responses 1. Administration: Vulnerabilities 2. Administration: Responses H. Vision: Vulnerabilities and Responses 1. Vision: Vulnerabilities 2. Vision: Responses I. Security: Vulnerabilities and Responses 1. Security: Vulnerabilities 2. Security: Responses CONCLUSION [T]he Santa Cruz Police Department became the first law enforcement agency in the nation to implement a predictive policing program. With about eight years of data on car and home burglaries, an algorithm predicts locations and days of future crimes each day. Police are given a list of places to go to try to prevent crime when they were not responding to calls for service. (1) We could name our top 300 offenders.... So we will focus on those individuals, the persons responsible for the criminal activity, regardless of who they are or where they live.... We're not just looking for crime. We're looking for people. (2) INTRODUCTION
In police districts all over America, "prediction" has become the new watchword for innovative policing. (3) Using predictive analytics, high-powered computers, and good old-fashioned intuition, police are adopting predictive policing strategies that promise the holy grail of policing--stopping crime before it happens. (4) Major cities in California, South Carolina, Washington, Tennessee, Florida, Pennsylvania, and New York, among others, have purchased new predictive policing software to combat property crimes such as burglaries, car thefts, and thefts from automobiles. (5) Data from past crimes, including crime types and locations, are fed into a computer algorithm to identify targeted city blocks with a daily (and sometimes hourly) forecast of crime. (6) Police officers patrol those predicted areas of crime to deter and catch criminals in the act. (7) In large cities such as Los Angeles, Chicago, and New Orleans, complex social network analysis has isolated likely perpetrators and victims of gun violence. (8) Social maps link friends, gangs, and enemies in a visual web of potential criminal actors. (9) Intervention strategies seek to reach these potential victims and perpetrators before the violence occurs. (10)
Law enforcement's embrace of predictive technology mirrors its adoption in other areas of the criminal justice system." New pretrial risk assessment models claim to be able to predict future dangerousness. (12) Post-trial sentencing predictions forecast likely recidivism. (13) Probability outcomes forecast likely probation violations. (14) It is no wonder, then, that predictive analytics have begun to shape policing strategies. Predictive analytics not only sounds like a futuristic solution to the age-old problem of crime, but also has the appeal of seemingly being based on empirical data free from human biases or inefficiencies. (15) Such marketing allure has resulted in a series of national news stories that have proclaimed predictive policing to be the future of law enforcement. (16)
Predictive policing thus raises some profound questions about the nature of prediction in an era influenced by data collection and analysis. The first generation of predictive policing technologies represents only the beginning of a fundamental transformation of how law enforcement prevents crime. (17) Identifying a future location of criminal activity may be statistically possible by studying where and why past crime patterns have developed over time. (18) But forecasting the precise identity of the future human "criminal" presents a far more troubling prediction. Both may be based on historical data with statistically significant correlations, but the analyses and civil liberties concerns differ. (19)
This article addresses the deeper questions behind the adoption of predictive analytics by law enforcement. The article develops a framework for how predictive technologies must be policed by legislators, courts, and the police themselves. Building off a wealth of theoretical insights of scholars who have addressed the rise of risk assessment in other areas of criminal justice, the article provides an analytical structure for future adoption of any new predictive technology.
This article offers three insights to the rather sparse literature on the subject of predictive policing. (20) First, the article situates predictive policing within the decades-long search for predictive solutions to criminal justice problems. Predictive policing may be new, but the lure of predictive techniques is not. Second, the article examines the rapid evolution from place-based property crimes to place-based violent crimes and then to person-based crimes. This evolution has largely gone unchallenged, even though the social science justifications for the different crime types remain contested. Third, and most importantly, the article uses the example of predictive policing to develop a theoretical framework to police all future predictive techniques. With the rise of big data, the Internet of Things, intelligence-driven prosecution, and as yet uncreated surveillance tools, law enforcement will continue to adapt and innovate. (21)
Part I situates the debate over predictive policing within the larger context of prediction in the criminal justice system. Prediction has been a "new thing" for decades and significant scholarly work has been done demonstrating its effects on other aspects of the criminal justice system. (22) From pretrial release to parole, predictive mechanisms now control many aspects of the criminal justice system. Predictive policing is but the next iteration of this move toward actuarial justice. (23)
Part II examines the evolution of predictive policing techniques from placed-based property crime to place-based violent crime. I call this the move from Predictive Policing 1.0 to Predictive Policing 2.0, in which the insights of a rather rigorous empirical and scholarly approach to studying property-based crimes have been adopted without equivalent empirical studies to the problem of violent crime. (24) While similar logic prevails, equivalent research does not. Part II also analyzes a separate technique focusing on the identification of individuals predicted to be involved in crime. (25) This is what I call Predictive Policing 3.0, with a focus away from places to persons. In cities like Chicago and New Orleans, sophisticated data programs are mapping shootings and studying the underlying human connections. (26) Mirroring a public health approach to disease, this focus on societal violence targets both potential shooting victims and offenders. (27) Targeted individuals are identified and interventions conducted to address (and hopefully prevent) future violent acts.
Part III then develops an analytical framework to evaluate police prediction. Specifically, I study nine core issues that must be addressed before adopting any predictive policing technology. These fundamental issues--data, methodology, social science, transparency, accountability, practical implementation, administration, vision, and security--present substantial risks and vulnerabilities for adopters of the technology. Because of the industry's rapid growth, police administrators and agencies have not adequately addressed these risks. The goal of this section is to move beyond Predictive Policing 2.0 or 3.0 to address universal concerns that will affect the next generation of technology, and all future predictive techniques.
The foundational insight of predictive policing is that certain aspects of the physical and social environment encourage quite predictable acts of criminal wrongdoing. (28) Interfering with that environment or those connections will--the theory goes--deter crime. (29) Predictive policing, thus, is less about blind fortunetelling, and more about divining hidden crime-inducing environmental conditions which can be deterred by an intentional police response. The same deterrence principle also can be applied to the predictive technologies themselves. This article seeks to...