Accountability and indeterminacy in predictive policing

Abstract

‘Predictive policing’ describes law enforcement agencies’ use of artificial intelligence algorithms to predict the locations of criminal activity. Building on research with the producers of a crime forecasting software, I discuss a number of epistemological contradictions and ethical concerns that predictive policing introduces. Chief among these is what I describe as ‘performative indeterminacy’ – a measurability paradox in which the increased probability that a crime is deterred due to algorithmic prediction is offset by the increased probability that it is detected. These opposing tendencies make the evaluation of predictive systems impossible after field deployment, thus raising questions about their ethical merit. With what recourse, then, do agencies justify their use of predictive systems? I argue that police intellectuals invoke the moral connotations of ‘accountability’ to warrant questionable ethical and epistemological assumptions. However, ‘accountability’ in this context problematically conflates three distinct areas of concern: (i) police agencies’ relationship to their publics, (ii) command staffs’ supervisory control over patrol officers, and (iii) the scrutability of algorithmic decisions – none of which, I maintain, is a realistic outcome of predictive policing, as long as its epistemological and ethical hazards remain unacknowledged. I conclude by arguing that, if recognised, performative indeterminacy could actually be put towards progressive ends. Engaging the fact that the use of predictions disconnects models from ‘ground truth’, it thus becomes possible to hold police departments accountable to a new standard: to disrupt entrenched geographies of police suspicion and enact a fairer and more equitable landscape of state surveillance.