Digital Data and Algorithms in Law Enforcement

Some pointers for responsible implementation

Abstract

Digital data and algorithmic tools have over the past years found their way into law enforcement contexts, including modes of biometric identification and matching, automated surveillance capacities, short-term situational predictions, AI-supported analysis for large amounts of data, and the interoperability of large-scale databases and platforms for data exchange and investigation. These tools can help to increase the effectiveness and efficiency of law enforcement operations on the strategic, tactical, and operational level. They do, however, also come with a number of concerns that must be acknowledged and addressed in order to realize their potential and avoid unintended side-effects and societal frictions. Based on a multi-year research project on predictive policing in Germany and Switzerland, this paper provides a systematic perspective on the challenges involved in implementing new and emerging technologies in law enforcement contexts. Specifically, it will address (1) the nature of data, i.e. how data are socially constructed and present a particular account of the world, inevitably leading to “biased” results; (2) transparency in algorithms and AI, i.e. how “black boxes” undercut human capacities to understand and retrace processes and create problems for public accountability; (3) automation and human control, i.e. the question how human operators can retain meaningful influence over analytical processes; (4) decision-making processes and automation bias, i.e. how humans can be empowered to critically question and override system recommendations; and (5) strategic and societal implications, i.e. the fact that digital tools should not be misused to displace larger programs that address the root causes of crime.