Protect, serve, predict?

Some think the moral structure of policing is closely analogous to defensive harm: it is permissible to impose proportionate harm on others when necessary to defend against threatened unjust harm. Many think that whether a policing strategy is justified depends primarily on whether it efficiently reduces the crime rate without directly violating any stringent rights. This paper rejects both of these views. It presents policing as a project in social risk management, in which we accept some risk of suffering rights-trespass by police activity in order to greatly reduce our exposure to risk of rights-trespass by private agents. It then evaluates the use of location-based algorithmic prediction tools to guide arrests or police deployments. This paper advances two core claims, one normative and one empirical. The normative claim is that decisions about how to police (and what tools to use) must be justified by broad principles for the fair distribution of risk, never only by the net effect on the observed crime rate. The empirical claim is that predictive algorithmic tools likely can’t clear this bar in a context like the present-day United States.