The following section aims to provide a snapshot of the themes and ideas with linkages to the issue of predictive policing. In this exposé of ideas, I will proceed as follows: First, I reference a body of work that concerns itself with the wider context in which a practice like predictive policing can occur and makes sense. In the subsequent sections, the focus will become narrower and consequently, I will end this introductory section with a sketch of the works that focus on predictive policing in Germany.
The idea here is not to represent fully or exhaustively the concepts of the mentioned authors. I raise no claim to completeness, rather, my aim is to embed the practice of predictive policing in the wider context of twenty-first century surveillance practices, i.e. practices predicated on the watching from above with the aim to predict, suppress or incentivise the behaviour of the watched.
The Precrime Assemblage
Writing about a twenty-first century policing technique such as those that came to be clustered under the label of predictive policing demands a certain degree of contextualisation. Predictive policing practices are fundamentally embedded in what Haggerty and Ericson 2000 described as the surveillant assemblage. Building on the work of Deleuze and Guattari 1989, they argue that surveillance and social control is no longer carried out only by discrete institutions like the panoptic prison or other state-run security forces. Writing at the cusp of the twenty-first century, they describe a convergence of these multiple surveillance systems into an integrated and rhizomatically expanding electronic ecosystem predicated on the abstraction of human bodies into a series of digital “flows” (Haggerty and Ericson 2000, 606). These digital flows can then be reorganized into distinct “data-doubles” (Haggerty and Ericson 2000, 606) that can be subjected to scrutiny or intervention.
Predictive Policing works according to this blueprint: First, Data flows about human behaviour from a multitude of corporate and state-run systems are integrated. The data that is gathered and used in predictive policing systems varies from system to system but always includes geographic data about where and how crimes took place in the past. Often, however, these systems go much further and integrate much more granular data about the human experience in certain quadrants of the city (socioeconomic indicators, number of children, meteorological data etc.). Next, these flows are reassembled into an abstract landscape of likelihood - often visualised on one single zoomable map - allowing for scrutiny and intervention, for example in the form of temporarily increased police presence. In this vein, Mantello 2016 aptly characterises the complex of predictive techniques in policing as the precrime assemblage.
From post-crime to pre-crime
Another critical building block in approaching predictive security practices come from theoretical criminology. In her seminal article, Zedner 2007 eloquently captures the temporal shift implicit in much of today's predictive security practices: The shift from looking at crime after it has happened, i.e. post-crime, to crime as a future possibility to be managed, i.e. pre-crime. In her own words:
“In a post-crime society there are crimes, offenders and victims, crime control, policing, investigation, trial and punishment”. She continues: “Pre-crime, by contrast, shifts the temporal perspective to anticipate and forestall that which has not yet occurred and may never do so. In a pre-crime society, there is calculation, risk and uncertainty, surveillance, precaution, prudentialism, moral hazard, prevention and, arching over all these, there is the pursuit of security” (p. 263).
The consequences of this shift are profound and wide-ranging. In pursuit of some common denominator of the pre-crime logic, Zedner goes on to theorise a central notion of the pre-crime logic, i.e. the logic of security. Apart from the aforementioned temporal shift, Zedner expands on two other contouring aspects of the security logic, namely the notion of risk and the economic focus on minimising loss. Risking oversimplification, risk or risk logic is characterised by the “control of the irrational by rational means” (Ericson and Haggerty 1997, 86), in which fear always plays the dominant motivational role. The raison d'être of risk technologies is thus the taming of chance, the domestication of the unknown or - in Foucauldian terms - “the biopolitical taming of aleatory events” (Aradau and Blanke 2017, 376). In this context, prisons for example, are not institutions of reform or punishment any more but rather become “carceral warehouses for those categorized as posing the highest risk” (Zedner 2007, 265).
Second, security is constructed as a commodity, i.e. something that can be bought and sold. Consequently, security and its absence can be marketed (Ericson and Haggerty 2002) allowing private actors to engage in the profit-driven provision of (in)security (Bigo 2002 argues that security can never be thought of without its antithesis and thus coined the term (in)security). This, in turn, means that to the degree with which private and corporate actors are becoming central providers of security, there will be market-driven discrimination in the distribution of risk and (in)security. Zedner 2007 cautions that “[w]here security is a saleable commodity, accountability to the democratic polity is liable to be usurped by the more powerful demands of narrower constituencies, be they bounded political communities, consumer groups or shareholders” (p. 266). On top of this, the more one relies on commodified security products, the more one craves them as the potential good that they promise is inherently limitless, i.e. there is no perfectly secure state to strive towards, just more threats to counter and more risks to mitigate. Ericson 1994 sums this up beautifully when he writes: “Security is marketed within a system seen as having limitless potential, and this system therefore augments insecurity. As with all forms of commodification, the more one experiences security products the more they become objects of desire and insatiable appetite” (p. 171).
The shift from post-crime to pre-crime and the security logic that underpins the latter are essential foundations in which any contemporary predictive security practice and especially predictive policing practices are embedded. Predictive policing practices in this sense are particularly clear examples of what Massumi 2015 calls the post-9/11 “primacy of preemption” that has come to define “a political epoch in as infinitely space-filling and insidiously infiltrating a way as the logic of “deterrence” defined the Cold War era” (Massumi 2015, 5).
Politics of patterns and regimes of “between-ness”
In their landmark study, Kaufmann, Egbert, and Leese 2018 sum it up well when they write that “patterns are the epistemological core of predictive policing” (p. 1). Drawing from the experience of 48 in-depth interviews in multiple countries, they proceed by laying out four ideal-typical styles of pattern-based predictive analytics - each with its own distinctive assumptions about humans and spaces that are more or less prone to criminal behaviour. In other words, patterns in policing are never the result of a singular logic and are always political in the sense that they “give form to and formalize different understandings about crime, which are in turn based on specific ideas of governing crime”. However, pattern-based logics tend to insulate themselves from scrutiny as they often fail to reveal their most basic assumptions about human nature and the nature of criminality and thereby solidifying their authority as incontestable forecasting devices.
Their analysis clarifies and adds much-needed contour to Aradau and Blanke's ) notion of between-ness as a essential element of how time and space is configured within governmental apparatuses (Aradau and Blanke 2017). According to Aradau and Blanke, inherent in practices of predictive policing (and all other forms of predictive analytics) is a certain geometrical understanding of the world and its features. In this vein, between-ness is defined as the shortest path between two data points in any given feature space. This geometrical primacy expresses itself in the models that underpin predictive policing software. According to these models, a crime leads to a higher probability of crimes in similar feature-space configurations. Thus, “[r]esearch into predictive policing is […] focused on how to featurize this spatio-temporal regularity in order to produce smaller hotspots and near-real-time decision-making” (p. 384). From there, Aradau and Blanke go much further and claim that within a “governmental apparatus of big data […] between-ness is distinct from the ‘docile bodies’ of discipline or the population of biopolitical government as it is the production of pure relationality, of geometrical connection as simultaneously similarity and difference” (p. 385). Whether the notion of between-ness deserves to take centre stage in the critical examination of contemporary practices of government is outside the scope of this paper.
For now, it suffices to acknowledge that pattern-based geometrical reasoning cannot separate itself from the political roots of any governmental practice. Perhaps the strongest element of Aradau and Blanke's contribution is the notion that, currently, the social sciences lack the vocabulary to effectively question the geometrical primacy that practices like predictive policing are predicated on. Against this backdrop, Kaufman et al.‘s contribution outlined above seems to be even more vital.
The aesthetics of predictive policing applications
Predictive policing programs - notwithstanding the varying degrees of complexity that enable their functionality in the background - always seek to present their prediction products in one single interface. Depending on the stage of development and funding of the program more generally, these end-user interfaces can be more or less sophisticated but they always seek to condense highly complex probabilistic reasoning in one single and appealing overview. For example, the rather low-budget KrimPro program in Berlin delivered its predictive insights as a static PDF-document displaying a color-coded heat map which was passed down to beat cops through the usual bureaucratic channels. In contrast, the much better funded SKALA program in North-Rhine-Westphalia cooperated with the Data Analysis and Visualization group (Link) at the University of Konstanz in order to set up a much more direct link with the patrolling policemen by allowing them to use a tablet computer to access the newest predictions interactively and in real-time. Thus, interfaces are - in Rehak's words - “ideological” (Rehak 2003). They work to remove themselves from awareness, seeking transparency or at least unobtrusiveness - as they channel agency into new forms” (p. 122). For another text that argues in favour of realising the inherent political nature of information visualisation see Sack 2011.
Naturally, this “scientification of suspicion and speculation” (McCulloch and Wilson 2016) risks obfuscating the heuristics underpinning the the predictive algorithms. A problem generally known under Latour's (1987) label of “black-boxing” . However, this practice faces another problem: It risks alienating the police officer who - before the introduction of predictive analytics into police work - made the decision where to patrol himself and could thus could take the credit once her “hunches” paid off. Coupled with fears of excessive monitoring, this deprivation of personal responsibility is among the strongest forces that the advocates of predictive policing face within the institution of the German police.
The issue of how knowledge is presented and visualised within the pre-crime assemblage already brings me one step closer to the research interest at the heart of the paper, namely on what grounds and according to which logic the probabilistic insights of predictive policing programs form the basis for police intervention in the twenty-first century. After completing this preliminary sketch of related research, my research interest will be made more concrete in order to allow a more thorough and more guided analysis of the German pre-crime assemblage in part II.
Predictive policing in Germany
Apart from the broader theoretical strokes aiming to capture the context of current predictive security practices, it is important to recognize - for the purposes of the present investigation - that predictive policing programs in Germany have already been subjected to intellectual scrutiny from a variety of disciplines with varied research interests. There are more general overviews of the issue area (Egbert 2017, 2018b; Rolfes 2017), more specific interrogations of particular aspects such as purely empirical policy-evaluation style investigations on the (unclear) impact of predictive policing on crime rates (Gerstner 2017, 2018) and even more specific investigations into the possible explanation for the so-called near-repeat victimisation that forms the basis for much of the predictive policing software (Gluba et al. 2016).
Among the most notable contributions from what might be called critical contemporary criminology is the already mentioned study by Kaufmann et al. that includes in-depth interviews also with German officials, but does not exclusively rely on material from Germany. One of the study's co-authors, Simon Egbert, has published extensively on the issue of predictive policing in the German context (Egbert 2017, 2018b, 2018a). Perhaps most interestingly for the present investigation, Egbert 2018b builds on the work of sociologist Rainer Keller and his _Sociology of Knowledge Approach to Discourse* (Keller 2013) in order to zoom in on one particular part of the discourse around predictive policing, namely the statements of interior ministers and their official rationale for the need to implement predictive policing software. He finds that apart from economic (increased efficiency of policing) and technological (advancements in analysing vast amounts of data) rationales, there is a distinctively political rationale which involves the framing of domestic burglary as a security problem necessitating a political response. The main discursively constructed threat to be countered via by introduction of predictive policing programs are highly mobile crime gangs from Eastern that specialise in serial domestic burglaries.
After this foray into the variegated literature that deals more or less directly with predictive security practices, it seems necessary to tie these insights together to present the background against which the research interest of present investigation is made visible.
Firstly and most fundamentally, Predictive policing practices in Germany and elsewhere are embedded in the broader context of a rhizomatically expanding pre-crime assemblage that abstracts human bodies into flows of digital data. These flows are then reassembled such that they can be subjected to scrutiny and intervention.
Secondly, predictive policing practices are exemplary in that they exhibit and render intelligible the shift from a post-crime to a pre-crime rationale and the inherent security logic of the latter. Apart from the fundamental temporal shift, security is predicated on the economic imperative of mitigating loss and minimising risk. Further, security is commodified in that it can be bought and sold. Its limitless profit-driven expansion therefore hinges on the construction of insecurity.
Thirdly, Kaufmann, Egbert, and Leese 2018 probed the epistemological foundation of predictive policing practices, namely, patterns. They show that predictive policing practices can follow a variety of pattern-based rationalities each with its own way to formalise how to think and what to do about crime. Hence, pattern are political and it is of primary importance to develop the vocabulary and methodological toolbox needed to excavate the inherently political assumptions that ground the geometrical abstractions of pattern-based reasoning.
Fourthly, predictive policing software seems to exhibit a certain aesthetisation of information visualisation, the existence of which warrants closer critical examination of its (philosophical) premises. As the visualisation of the predictive insights form the basis for real-world intervention, their aesthetisation must be taken into consideration when the aim is to excavate the logic inherent in the conceptualisation and implementation of predictive policing programs in Germany.
Lastly, I provided a short overview over the research that concentrates on German predictive policing programs, zooming in on more critical and discourse-centered approaches to the issue-area. Against this colourful background, I will now attempt to clarify my own research agenda.
Aradau, Claudia, and Tobias Blanke. 2017. “Politics of Prediction: Security and the Time/Space of Governmentality in the Age of Big Data.” European Journal of Social Theory 20 (3):373–91. http://journals.sagepub.com/doi/10.1177/1368431016667623.
Deleuze, Gilles, and Felix Guattari. 1989. “A Thousand Plateaus: Capitalism and Schizophrenia.” Journal of Interdisciplinary History 19 (4):657. https://www.jstor.org/stable/203963?origin=crossref.
Egbert, Simon. 2017. “Siegeszug Der Algorithmen? Predictive Policing Im Deutschsprachigen Raum.” Aus Politik Und Zeitgeschichte 67 (32-33):17–23. http://www.bpb.de/apuz/253603/siegeszug-der-algorithmen-predictive-policing-im-deutschsprachigen-raum.
———. 2018a. “Predictive Policing in Deutschland. Grundlagen, Risiken, (Mögliche) Zukunft.” Räume Der Unfreiheit. Texte Und Ergebnisse Des 42. Strafverteidigertages Münster, 2. - 4.3.2018. Berlin: Organisationsbüro Der Strafverteidigervereinigungen 2:241–65. https://www.strafverteidigervereinigungen.org/Schriftenreihe/Texte/Band%2042/Egbert%5F241%5F265%5F42SchrStVV.pdf.
———. 2018b. “About Discursive Storylines and Techno-Fixes: The Political Framing of the Implementation of Predictive Policing in Germany.” European Journal for Security Research, January. http://link.springer.com/10.1007/s41125-017-0027-3.
Ericson, Richard V. 1994. “The Division of Expert Knowledge in Policing and Security.” The British Journal of Sociology 45 (2):149. https://www.jstor.org/stable/591490?origin=crossref.
Ericson, Richard V., and Kevin D. Haggerty. 1997. Policing the Risk Society. Toronto: University of Toronto Press. https://www.jstor.org/stable/10.3138/9781442678590.
Gerstner, Dominik. 2017. “Predictive Policing Als Instrument Zur Prävention Von Wohnungseinbruchdiebstahl. Evaluationsergebnisse Zum Baden-Württembergischen Pilotprojekt P4.” Forschung Aktuell - Research in Brief 50 (1). https://www.mpicc.de/de/forschung/forschungsarbeit/kriminologie/predictive%5Fpolicing%5Fp4.html.
———. 2018. “Predictive Policing in the Context of Residential Burglary: An Empirical Illustration on the Basis of a Pilot Project in Baden-Württemberg, Germany.” European Journal for Security Research 3 (2):115–38. http://link.springer.com/10.1007/s41125-018-0033-0.
Gluba, Alexander, Eva Groß, Nina Hermes, and Laura Hoppe. 2016. “Einmalige Vs. Mehrmalige Wohnungseinbrüche. Ein Test Der Flag-Hypothese Zur Erklärung Wiederholter Viktimisierungen.” Kriminalistik 6 (6):393–401. https://www.researchgate.net/publication/304353848%5FEinmalige%5Fvs%5Fmehrmalige%5FWohnungseinbruche%5FEin%5FTest%5Fder%5FFlag-Hypothese%5Fzur%5FErklarung%5Fwiederholter%5FViktimisierungen.
Haggerty, Kevin D., and Richard V. Ericson. 2000. “The Surveillant Assemblage.” British Journal of Sociology 51 (4):605–22. http://doi.wiley.com/10.1080/00071310020015280.
Kaufmann, Mareile, Simon Egbert, and Matthias Leese. 2018. “Predictive Policing and the Politics of Patterns.” The British Journal of Criminology, December. https://academic.oup.com/bjc/advance-article/doi/10.1093/bjc/azy060/5233371.
Mantello, Peter. 2016. “The Machine That Ate Bad People: The Ontopolitics of the Precrime Assemblage.” Big Data & Society 3 (2):205395171668253. http://journals.sagepub.com/doi/10.1177/2053951716682538.
Rolfes, Manfred. 2017. “Predictive Policing: Beobachtungen Und Reflexionen Zur Einführung Und Etablierung Einer Vorhersagenden Polizeiarbeit.” Potsdamer Geographische Praxis 12:51–76. https://publishup.uni-potsdam.de/frontdoor/index/index/docId/10344.
Sack, Warren. 2011. “Aesthetics of Information Visualization.” In Context Providers: Conditions of Meaning in Media Art, edited by Margot Lovejoy, Christiane Paul, and Victoria Vesna. Bristol: Intellect.
Zedner, Lucia. 2007. “Pre-Crime and Post-Criminology?” Theoretical Criminology 11 (2):261–81. https://doi.org/10.1177/1362480607075851.