Counter-terror is an algorithmic mode of governance that reinscribes racial capitalism’s biopolitical control---their representations are transformed into racist predictive policing measures.

Sharma and Nijjar ’23 [Sanjay Sharma, Associate Professor at the University of Warwick. Jasbinder S. Nijjar, lecturer in sociology at the University of Surrey; “Post-racial politics, pre-emption and in/security”; 05-01-2023; European Journal of Cultural Studies, vol. 27, no. 2; doi.org/10.1177/13675494231168177; accessed 07-25-2024] dliang

The objective of preventing crime, rather than simply responding to already committed crime, has a long history in Western nations. Some of the earliest forms of militarized policing in Britain’s colonies reflected a ‘commitment to a preventative function’ (Brogden, 1987: 11), as part of a broader strategy of racial regulation. However, it was during the 1970s in the USA that prevention became an explicit strategy in attempts to reduce rising levels of crime and make policing more efficient and effective. This heralded a future-orientated intervention-based approach premised on the prediction and pre-emptive management of crime.

The impact of advances in digital technology on policing over the last few decades has accelerated and entrenched a belief in the possibility of data-driven crime prediction. What is now commonly referred to as ‘predictive policing’ has been understood as the proactive use of algorithmically mediated data analysis for the purpose of finding patterns in datasets, based on which risk estimates are produced for either individuals or locations and are operationalized in the form of targeted prevention measures (Egbert and Leese, 2021: 19).

The growth of predictive policing is connected to an increasing number of crime-prediction software applications. In particular, the for-profit company, PredPol (re-named Geolitica), pioneered the adoption of these kinds of crime-forecasting tools among police departments in the USA and UK.

The PredPol software was developed in 2010 through a collaboration between the Los Angeles Police Department and researchers from the University of California. A co-founder of the company, the anthropology professor Jeff Brantingham, used a time- and place-based approach to identify urban geolocations that were supposedly more at risk of certain crimes. Brantingham, working with other colleagues, repurposed algorithms used for predicting earthquake aftershocks to anticipate the location of crime ‘hot spots’ based on ‘near-repeat’ offenses and other behavioral criminology theories (Shapiro, 2017). However, often overlooked are the militarized origins of the development of crime-prediction software programs. Brantingham was motivated to develop PredPol through his original Pentagon-funded work, which forecasted battlefield casualties in Iraq. Other projects included research funded by the US Department of Defense which attempted to model ‘gang’ violence in California via Iraqi insurgent attacks (González, 2015).1

A key innovation of crime-prediction software has involved utilizing machine learning algorithms, increasingly referred to as artificial intelligence (AI) systems, which are claimed to ‘learn’ from data.2 Machine learning is based on approaches including ‘super- vised’ and ‘unsupervised’ learning in which an algorithmic model is trained on data to predict outcomes. Supervised approaches depend on labeled (structured) data and known outcomes to learn or model the relationship between inputs and outputs. In contrast, unsupervised learning uses unlabeled (unstructured) data and discovers statistical correlations or patterns (significant features) to generate a model without knowing the outcomes in advance.

Predpol appears to use supervised machine learning to make area-based predictions of crimes (Haskins, 2019). It relies on ‘hot spot’ mapping and ‘near-repeat’ theory to produce risk estimates by analyzing existing police data (date/time, location and crime type).3 More recently, there have been innovations in forecasting crime that move beyond predefined police datasets to (dubiously) claim greater accuracy. HunchLab has been a leader in deploying ‘big data’ predictive analytics to generate geospatial crime predictions, based on utilizing both supervised and unsupervised approaches (Mittelstadt et al., 2016).

Hunchlab diversified data sources involve infrastructural and environmental characteristics including risk terrain modeling (based on weather patterns, locations of amenities and socioeconomic indicators). In HunchLab’s own words,

The system automatically learns what is important for each crime type and provides recommendations of where to focus the resources that you have available. If you don’t have particular datasets (such as bars or bus stops), the system simply adapts to use the data available in a given jurisdiction (Azavea, 2015: 10).

Notwithstanding the problematic nature of ‘crime theories’, HunchLab implicitly expounds an ideology of ‘big data’ that is predicated on ‘data science’ empiricist claims of greater objectivity and accuracy in the algorithmic modeling of complex social phenomena. In the case of crime forecasting, infrastructural and environmental data supposedly mitigate against the possible historical partiality of police data or the politics of police knowledge.

Azavea, the original company behind HunchLab, sold the latter to ShotSpotter in 2019, who renamed HunchLab as ‘Community First Patrol Management Software’. The software claims to mitigate bias by using ‘objective, non-crime data and purpose-built mechanisms’ (ShotSpotter, 2022), which acknowledge what is known as the ‘garbage in, garbage out’ computing problem of automated decision-making systems. This approach purportedly addresses how institutional racism has created conditions for the disproportionate targeting and overpolicing of racially defined communities, thus producing crime data determined by racialized relations of power. If these data are only used to train crime prediction algorithms, it leads to a further entrenchment of racism–what has been described as the problem of ‘runaway feedback loops’ (Ensign et al., 2017).

ShotSpotter advocates that by improving (expanding and diversifying) datasets and deploying advanced machine learning algorithms, predictive policing will deliver neutral, scientifically objective crime forecasting. The company promotes an AI dynamic modeling of crime premised on fallible notions that the system improves as it ‘self- learns’. What ShotSpotter fails to address is that data in the form of neighborhood boundaries, types of amenities and other socioeconomic indicators, when correlated, can operate as discriminatory proxies of race. It may lead to amplifying risk estimates of crime in racialized and marginalized areas (Chun, 2021).

It is worth momentarily pausing to clarify that what is called ‘crime’ cannot be grasped outside the function of policing and its relationship to race-making and racial regulation. To elaborate, classic and contemporary work examines how police actively and strategically mobilize race as a social category in the production and publication of crime statistics (see Bridges, 2015; Bridges and Gilroy, 1982; Gutzmore, 1983). This mutual process of racializing crime and criminalizing race legitimizes racist patterns of policing. While such relations of power are harmful in and of themselves, their point is ‘not just to punish Black communities but to mark them’ (Kelley, 2016: 28). In other words, racial regulation (re)produces formal police knowledge that (re)institutionalizes race as a law- and-order problem. As Joshua Scannell (2019: 108) maintains,

Policing does not have a ‘racist history’. Policing makes race and is inextricable from it. Algorithms cannot ‘code out’ race from American policing because race is an originary policing technology, just as policing is a bedrock racializing technology.

Scannell draws on the work of the abolitionist scholar, Ruth Wilson Gilmore, when characterizing the criminal justice system as organized through racism. Gilmore (2007: 247) defines racism as ‘the state sanctioned and/or extralegal production and exploitation of group-differentiated vulnerability to premature death’. From this standpoint, Scannell (2019) argues that,

In the inverted world of predictive policing, group differentiated vulnerability is translated into probable criminal ‘risk’. Predictive policing software uses almost every conceivable measure of vulnerability and victimization under American racial capitalism. (p. 111)

Our point is that the long, enduring and institutionalized relationship between racism and criminal justice makes the pursuit of ‘objective’ crime prediction impossible. The elusiveness of such an endeavor is further compounded by the biopolitical (and hence racialized and militarized) concept of statistical risk, on which predictive policing is predicated.

Dan Bouk (2015) traces the connections between prediction, risk and race in the history of actuarial science and life insurance policy in the USA. A burgeoning post-Civil War insurance industry effectively created–from what we now call ‘big data’–a ‘statistical individual’ as a commodified risk. Significantly, the operation of ‘white data politics’ materialized via the construction of an exclusionary, normative ‘white’ category in the classificatory practices of insurance companies. Black populations, in contrast, were racially marked not only because of perceived biological differences but also in terms of a black pathology characterized by their itinerancy and indeterminacy (e.g., lacking legitimate medical or property records). Bouk notes how ‘these systems perpetuated inequality in the ways they made risks, in their systematic preference for making African Americans into substandard, subprime risks’ (p. 185).

What emerged from the production of Black populations through notions of probability and prediction was biopolitical, statistical racism, based on (ethno)racially coded ‘otherness’ as ‘risky’, to manage uncertainty, pre-empt future outcomes and inform pre- sent actions. Furthermore, the relations of exclusion, isolation and marginalization that materialized from such biopolitical arrangements determine contemporary algorithmic calculations. For instance, financial policies that redlined African American neighbor- hoods deemed risky ‘have greatly influenced modern algorithms because they generated massive datasets that consist of decades of information built on exclusion and discrimination’ (Allen, 2019: 234). Today, the data used to calculate credit scores, approve or deny mortgages, determine interest rates on loans and make other automated decisions include not only financial transactions but also a host of ‘suspect, race-related data’ (p. 238). This marks a process of ‘digital redlining’ where algorithmic calculations rest on, and reinforce, long-standing relations of racial denial, exploitation and segregation (Noble, 2018).

With race being a key determinant of risk production, contemporary technological practices of predicting and mitigating criminal risk are bound to institutional racism in the criminal justice system and manifest the racial logics and consequences of earlier forms of ‘corporate risk-making systems’ (Bouk, 2015). It should be of no surprise that the ‘big data’ opaque machine learning models of crime-prediction software programs have been entangled with the innovation of predictive analytics of the financial industry. For instance, IBM spent over $14 billion developing analytics software for corporate, law enforcement and state security. The techno-solutionist turn to manage uncertainty has led to the ‘language of race’ being displaced by a ‘language of risk’ (Wang, 2018: 251). This coding of racial vulnerability as risk, and of risk as racially determined, is critical to the construction of a ‘digital carceral infrastructure’. And, as Jackie Wang observes,

It is important that we pay attention to this paradigm shift, as once the ‘digital carceral infrastructure’ is built up, it will be nearly impossible to undo, and the automated carceral surveillance state will spread out across the terrain, making greater and greater intrusions into our everyday lives (p. 251).

We have sought to emphasize that race is central to the constitution of risk and its attendant practices of predictive policing. The widespread adoption of racially deter- mined predictive policing systems, in turn, is integral to the intensification of a ‘digital carceral infrastructure’. Such developments are part of the ‘biopolitics of securitisation’ (Nijjar, 2022), where the racialization of risk, and the production of race as risk, determines racial and martial regimes of police surveillance, analysis and pre-emptive action. As we discuss in the next section, the relationship between race, surveillance, analysis, intervention and securitization is central to instigating an inimical development from prevention to pre-emption in the post-racial ‘calculus of risk’ (Amoore, 2013).