Beyond the algorithms: Evaluating the risks of deploying machine learning in domestic counterterrorism: A comparison between predictive policing and counterterrorism activities

Abstract

Over the last decades, Machine Learning (ML) has been implemented in nearly every part of our daily lives. Whereas this development was heavily discussed in the area of predictive policing (PP), there was little public debate when it comes to ML's implementation in domestic counterterrorism (CT). This is due to the fact that the counterterrorism domain is a very non- transparent field. Classified information forms an obstacle to proper scholarly analysis. The thesis aims to contribute to the public debate on the implementation of ML in CT by asking the following research question: examining critiques provided by scholars on predictive policing, what are the risks of deploying machine learning tools in domestic counterterrorism? A comparative case study method supplemented by scenario-building allows for an analysis of the risks of ML/CT. More specifically, by using PP arguments as a proxy for CT, arguments can be made to show that the technical and socio-technical risks, in most cases, also hold for counterterrorism tools. The analysis highlights those risks by exploring PP arguments for three CT instruments: individual risk assessments, biometric tools (most notably facial recognition technology), and general models that predict details of future terrorist attacks. It was found that only the last...