Criminology: Predicting police enforcement bias in major US cities | Nature Human Behaviour | Nature Portfolio

Police responses to crime in eight major US cities may be biased by neighborhood socio-economic status, according to a modelling study published in Nature Human Behaviour. The finding — revealed by studying changes in patterns of crime using AI — suggests that socio-economically disadvantaged areas may receive disproportionately less policing attention with resources drawn away to their relatively better-off neighborhoods, especially when and if crime spikes.

Event-level crime prediction has the potential to preempt the time and place of future criminal infractions, making it a popular tool amongst police and government agencies. Such policing practices have raised concerns, however, regarding whether they might propagate or codify biases, especially in diverse communities, and practical success of such efforts in the field have been historically limited.

Using data from eight major US cities (Chicago, Philadelphia, San Francisco, Austin, Los Angeles, Detroit, Portland and Atlanta), Ishanu Chattopadhyay and colleagues developed an algorithm that could predict crime in a highly accurate manner, which allowed the authors to perform an analysis of potential policing bias. Focusing on Chicago, for example, the algorithm was able to predict criminal events with a less than twenty per cent false-positive rate, outperforming existing prediction algorithms. When predicted policing behaviour from the model was analysed, however, the police were found to be less responsive to crime in less affluent areas than in those that were more affluent, suggesting an enforcement, and perhaps a policy, bias.

The authors stress the importance of using their findings cautiously, discouraging the application of their algorithm to guide policing. In an accompanying News & Views, Andrew Papachristos emphasizes that it is unlikely that police and government agencies will discontinue the use of crime prediction models. Instead, he suggests that a “major next step for using such technology as a way to monitor police behaviour involves figuring out who would use these tools to monitor the police and what sorts of ramifications are even feasible when problems within departments or abuses towards the public are detected.”