1 Review: Simon Egbert, Matthias Leese. Criminal Futures. Predictive Policing and Everyday Police Work. Routledge, 2021, 231 pp

With their extensive study on predictive policing, Egbert & Leese (2021) offer an important contribution to the discussion on the ongoing digitization and algorithmization of police work. The aim of the book is to understand predictive policing as a set of related socio-technical practices. The insights into these practical realities are based on empirical material collected in Germany and Switzerland, and include ethnographic fieldwork, interviews, and document analysis. Thus, the presented arguments are grounded in a broad and rich collection of qualitative data, which the authors discuss extensively throughout the book. This vast collection of insights is one of the book’s core strengths, as the authors were at the heart of the action and – as Latour stresses in his writings – followed the actors. Latour is indeed an important point of reference, as the collected material is discussed in the context of Actor-Network Theory (ANT), trying to connect insights from (critical) criminology, sociology, and Science and Technology Studies (STS). With this approach, Egbert and Leese aim to reconstruct how predictive policing is co-constituted as an institutionalized and organized practice by technology, practitioners, and discourses.

The book is structured according to what the authors consider to be a chain of translations. The chapters follow each other as a logical progression from developing the algorithm to installing it into the socio-technical structures of policing organizations and the wider societal context at large. This structure is convincing and allows the reader to follow the making of the algorithmic system, where each step provides the reader with new insights that connect back to what s/he learned so far.

While the first chapter offers a general introduction into the book, the second sets the scene from a criminological perspective by situating the development of predictive policing in the wider context of the “preventive turn” (Crawford & Evans, 2017) and the general shift towards intelligence-led policing. The authors show that the emergence of predictive policing as a technology is not changing the work or the culture of policing in a radical way. Instead, the utilization of such tools is the result of an ongoing convergence of innovations and developments within the field. The authors argue that “predictive policing can be understood as yet another step in this quite long lineage of anticipatory and managerial developments” (Egbert & Leese, 2021, S. 23).

The third chapter then shifts the perspective to the fields of sociology and Science and Technology Studies (STS). Drawing from the former two concepts, the authors display convincingly how the introduction of new technologies into new contexts is a complex, messy, and often unpredictable endeavor. As a result, technological change almost always also implies organizational change that has not been anticipated during technological development. At this point the authors introduce STS, more specifically references to ethnomethodological work (Lynch, 2008; Suchman, 2006) and ANT (Callon, 1986; Latour, 2005). This change in perspective shifts the attention from the practices of policing on the streets to predictive policing and comparable applications as inherent elements in constituting the institution of police.

Chapter four deals with the problem of data production for predictive systems in the context of predictive policing. Following the notion that “raw data is an oxymoron” (Gitelman, 2013), the authors explain what kind of data is used and what problems occur when analog social phenomena are translated into algorithmically readable symbols. They also describe the link between the (perceived) practices of the criminal subjects and the overall system of predictive policing. From such a perspective, deviant behavior is not just being policed, but becomes co-constitutive to the system of predictive policing. The chapter creates an interesting and often less explored bridge between different collectives by understanding data as an emergent product of the coordination between these asynchronous practices.

Chapter five takes up and develops this argument. Instead of asking questions about how data is produced, the authors refer to insights from Human-Computer-Interaction (HCI) and neighboring fields to take a closer look at the multiple interactions between algorithmic output and the operators of predictive policing systems. Against the myth that predictive analytics are completely automated (see e.g. Rouvroy, 2013), they reconstruct the many different negotiations and (re-)evaluations of algorithmic output in terms of an algorithm-human configuration. Confronting the often-raised critique of algorithmic systems neglecting the importance of context information, the chapter shows how domain expert knowledge has been added into the picture after the calculations.

One of the most prominent features of predictive policing is the system-produced risk visualization map. Chapter six examines how predictive policing applications turn data into visual representations of risk. In their description of the transformation of criminal data into maps, the authors connect insights from ANT and STS with longstanding questions in criminology and the sociology of policing. They focus on how maps are produced in different stages, and how each moment of translation solidifies the projected crime risk as a fact, where “visual representation allows the police to make the future, while not yet here, tangible and relatable” (Egbert & Leese, 2021, S. 129). The inquiry opens up several perspectives on the topic, as the authors follow the artefact of the risk map to the context in which it is used. As such, this chapter is the part of the book that perhaps most clearly connects concepts and theories from different disciplines, and shows the productive potential of such an interdisciplinary perspective when examining phenomena like predictive policing.

Chapter seven then takes the action to the street, so to say. The book illustrates how predictive policing, although often understood as a disruptive technology to policing, fits in a long trajectory and history of patrolling the streets as part of everyday police work. Thus, this chapter portrays the subtler changes in the way how police work is being carried out under conditions of predictive policing systems. Again, applying ANT sensitivities, the authors also reconstruct how the entire system of predictive policing depends on the police forces ‘out there’ using the risk maps and predictions produced by the algorithmic system. Without being successfully integrated into the practices of the officers, the system loses its ability to enable any form of agency. The authors contextualize these enrolments in a larger debate of scientific rationality vs. gained experience and how these blend into a culture of policing.

Chapter eight touches on a question that has been asked since predictive policing came into the focus of public discourse: Does it work? Predictive policing itself comes with many uncertainties and difficulties with showing causal effects, often due to the complex setup of the empirical field and the lack of a ceteris paribus-like situation for these studies. The chapter, however, takes an interesting turn, asking not if predictive policing works, but rather how the empirical field (i.e., the police force) define success and how they account for it. Relying on a field-specific formulation of success might not enable a broader discourse about the usefulness of predictive policing as such – which is, as argued, a complex endeavor in itself – but shows the challenges this technology poses to the police forces. Thus, and maybe surprisingly, effects like the technological and organizational integration of police departments in order to make predictions possible in the first place have been framed as a success.

The last chapter I discuss here, chapter nine, discusses normative and ethical challenges posed by predictive policing. Classical examples, especially for algorithmic systems, are fairness, accountability, and transparency. These issues are well known also from other domains, and while it is worthwhile to take them on, it is hardly surprising. Having said that, this part also changes the perspective, not only asking how predictive policing is changing the practices of policing, but also how established practices of policing are changing the production of data and therefore the calculations of predictive policing applications. By this approach, the book closes the cycle of the translation processes, bringing us back to chapter four, but with a different perspective. A perspective that focuses on the value-ladenness of data production and its stabilization within the system of predictive policing.

The different stations and sites in the book present a well-rounded and comprehensive picture of predictive policing. The book thereby offers an empirically rich and well-grounded analysis of what predictive policing means on the level of actual practices. By digging deep into the different layers of translation, and by following different actors, the authors clearly narrate how predictive policing changes the practices of policing and vice versa. The book also connects the different disciplinary perspectives of (critical) criminology, sociology, and STS. Especially the STS approach proposed is very convincing, adding a novel and important perspective to the field. The use of this perspective, however, leads to one minor issue in the book. Because it is a crucial element of the argument and the contribution the book makes, it would have been useful to expand on the discussion about the processes of translation and to which theoretical background ANT is speaking to. This would have provided a better foundation for readers not familiar with STS.

However, this is only a minor issue in an otherwise great book that connects different strands of thinking to explore the phenomenon of predictive policing. It is by far the most comprehensive and best-informed study on predictive policing I have encountered so far, and it presents a nuanced contribution to the academic discussion on these algorithmic technologies. The authors illustrate that we should understand predictive policing as a complex and interrelated system of socio-technical practices. This is also true when they reflect on the broader social implications of a technology that mostly works preventively through police presence. Predictive policing is a system that treats symptoms but leaves out the causes. For the latter, a socio-technical theorizing of the political economy of crime would be necessary, which an algorithm simply cannot do. One is instantly reminded of Tony Blair’s slogan ‘tough on crime, tough on the causes of crime’ and how the second part of this phrase is all too often lost in translation (Lea, 2015). As such, this book marks an important and dearly needed contribution to the discussion and can also be understood as a welcome starting point for further studies and discussions on predictive policing.

Interessenskonfliktstatement

Die Autor*innen erklären, dass ihre Forschung ohne kommerzielle oder finanzielle Beziehungen durchgeführt wurde, die als potentielle Interessenskonflikte ausgelegt werden können.

Literatur