Police in major metropolitan areas now use “predictive policing” technologies to identify and deter crime. Based on algorithmic forecasts from past crime patterns and individual criminal risk factors, police claim to be able to identify places and persons more likely to be involved in criminal activity. This data-driven approach impacts police patrols, investigations, and public health—like strategies to disrupt and monitor forecasted criminal activity.
The early success of predictive policing has led a few prosecutors’ offices to adopt quasi-“predictive prosecution” strategies. Predictive prosecution involves identifying and targeting suspects deemed more at risk for future serious criminal activity, and then using that information to shape bail requests, charging decisions, and sentencing arguments. The potential problem, however, is that the data used to inform predictive prosecution strategies may be subject to the same vulnerabilities currently limiting predictive policing. Data can be bad, biased, or based on erroneous correlations. Data-driven justice challenges values of transparency, accountability, and autonomy. And, while these problems matter when it comes to questions of where to send a patrol car, or even whom to investigate, they matter much more when data directly impacts a prosecutor’s decision about individual liberty.
Fortunately, prosecutors, more so than police, may have the institutional capacity and power to ensure an equitable and accountable use of predictive technologies. Prosecutors, due to their ethic “to do justice,” may be in a better position to ensure that issues of accuracy, transparency, validity, error, and exculpatory information are addressed before widespread adoption. Prosecutors may be able to capitalize on the innovation of predictive analytics and promote stronger accountability mechanisms that could benefit the entire criminal justice system.
This Essay sets out the preliminary questions that prosecutors should ask before adopting any type of quasi-predictive prosecution system.
