Predictive policing

Predictive policing is the practice of using data analysis and statistical algorithms to anticipate where crimes will occur or who will commit them. It transforms historical crime data into deployment decisions for police resources, promising efficiency over reactive patrolling. But the methodology creates a fundamental epistemological trap: the algorithm predicts crime based on where police have historically looked, and the prediction then directs police to look there again. The result is a self-fulfilling prophecy that criminalizes geography rather than reducing harm.

The most widely deployed systems, such as PredPol and Palantir, use variations on earthquake aftershock prediction models, treating crime as a spatial contagion. This framing is not neutral — it encodes the assumption that crime is a natural phenomenon rather than a socially constructed category, and that police presence is the appropriate response. The systems-theoretic insight is that predictive policing does not merely optimize resource allocation; it actively constructs the reality it claims to predict. By concentrating enforcement in historically over-policed neighborhoods, the algorithm validates its own outputs while eroding the legitimacy of the communities it targets.

Predictive policing is the paradigmatic WMD because it combines opacity, scale, and damage in a closed feedback loop. The harm is not a bug but a structural feature: the model cannot distinguish between crime and policing.