Conditions of benefits and risks when algorithmic technology is implemented for public sector policing and fraud detection: a systematic literature review

Across the globe, cities, states, and nations are moving forward with implementing artificial intelligence and machine learning technologies for policing and fraud detection, motivated by an optimistic view that technology can improve the functioning of government. Yet, their implementations have sometimes led to disastrous outcomes for citizens, leading many to view these technologies with deep skepticism. We bridge this divide by conducting a multi-disciplinary systematic literature review (n = 157) focusing on specific conditions for benefits and risks associated with the implementation of artificial intelligence and machine learning technologies in predictive policing and fraud detection and prevention. We find that the social science literature generally and, more specifically, the field of Public Administration focuses on the risks of algorithmic technology, and we identify conditions of risks, such as when algorithmic systems are designed with a model of threat. The engineering literature instead focuses on potential benefits, and we identify conditions of benefits, including alignment with legal and policy objectives. We integrate these conditions into a socio-technical governance framework that conceptualizes technical system quality, human–technology interaction, and institutional context as interacting mechanisms shaping both decision outcomes and institutional legitimacy. This framework highlights the importance of closer collaboration between data scientists and social scientists to design and implement policing and fraud detection systems that better align with public and societal values.