Granular biopolitics: Facial recognition,pandemics and the securitization of circulation

Abstract

The COVID-19 pandemic has provided opportunities for facial recognition technology and other forms of biometric monitoring to expand into new markets. One anticipated result is the wholesale reconfiguration of shared and public space enabled by the automated identification and tracking of individuals in real time. Drawing on data from several industry trade shows, this article considers the forms of ‘environmental’ governance envisioned by those developing and deploying the technology for the purposes of security, risk management, and profit. We argue that the ‘contactless culture’ that emerged during the COVID-19 pandemic anticipates the normalization of a form of mass-customized biopolitics: the ability to operate on the population and the individual simultaneously through automated forms of passive identification. This form of governance relies not just on machinic recognition, but on the real-time reconfiguration of physical space through automated access controls and the channelling of both people and information.

A New York City attorney received widespread media coverage when she was singled out by an automated facial recognition system and denied access to see the famed Rockettes with her daughter at Radio City Music Hall during the Christmas holidays. Reportedly, the venue’s parent company, which owns several New York City venues, including Madison Square Garden, had a database of the employees of law firms engaged in legal action against and was systematically excluding them from events (Hill and Kilgannon, 2022). The incident was, in the scheme of things, relatively minor, but it highlighted the use of facial recognition technology for the management circulation in a way that is likely to become increasingly common as the technology spreads. We open with this example, because the control of access and circulation with automated facial recognition was a recurring theme in our research on the deployment of facial recognition technology during the COVID-19 pandemic. The multiplication of boundaries and checkpoints during the pandemic lent itself to automated firms of identity and status verification. As in the case of the New York attorney, individuals in a crowd could be identified and singled out – based on a range of information including COVID-19 symptoms (such as elevated temperature), past exposure, and their vaccination or quarantine status. Although the focus of this article is on the framing of the utility of facial recognition during the pandemic, the logic of automated governance we examine has broader relevance in the era of remote, real-time biometric identification.

Indeed, the imperatives of the COVID-19 response – social distancing and contactless-ness – accelerated the development and deployment of passive forms of tracking and detection that enable increasingly individualized forms of social control. As soon as physical proximity came to be viewed as a threat, technologies that provide ‘at-a-distance’ services were enrolled to replace face-to-face activities so as to reduce the potential for viral contagion. At the same time, the goal of preserving as much circulation as possible led to the replacement, in many contexts, of blanket forms of quarantine by targeted forms of screening and sorting. The goal was to allow ‘safe’ forms of circulation while identifying and curtailing avenues of potential contagion. As it transpired, the social distancing imperative became a selling point for the emerging smart-camera and facial recognition industry, which mobilized the promise of efficient, passive, mass-customized monitoring. The result was, as one news account put it, ‘a lucrative market for facial recognition manufacturers’ (O’Donnell, 2020).

The widespread highly publicized response to the pandemic thus spurred, ‘novel uses of biometric technologies to limit contagion and maintain economic opportunities’ (Van Natta et al., 2020: 1). Our field work in security industry trade shows suggests that for the promoters and vendors of the technology, the pandemic provided additional impetus for highlighting the personalized logics of governance and control already envisioned by the technologies they have been developing and promoting. ‘Frictionlessness’ – for example, could be reframed not just as a means of easing passage through existing checkpoints (such as secure locations, ticketed venues and transit turnstiles), but as a way of managing the proliferating array of borders and access points associated with pandemic management. Office buildings, apartment complexes, shopping malls, and public facilities sprouted checkpoints to monitor vaccine status, symptoms, and potential exposure risks. Some systems regulated access to workspaces based on pandemic occupancy requirements. This ‘thickening’ of the borders to fill a growing number of spaces – and even to enable continuous real-time monitoring – heightened the need for automated forms of identity and status verification. The ability to deploy automated recognition systems at the level of individual movement and access control represents an emerging scale and temporality of the governance of circulation with implications that, while revealed by the pandemic, extend beyond it.

Drawing upon field research on the use of facial recognition during the pandemic, we describe this form of governance as a ‘granular’ form of biopolitics – a formulation meant to highlight the customized forms of intervention it envisions. A more precise, though perhaps more obscure, formulation would be to describe the mass-individualized governance of shared spaces as a granular form of ‘environmentality’. This term invokes speculative observations by Michel Foucault (2008: 259) about forms of control that operate not at the level of subjectification (as in the case of disciplinary practices), but at that of the environment, or ‘milieu’. Typically, the milieu refers to a shared environment, however, the novelty of automated identification is that it enables the individualization and customization of the milieu itself. The result, we argue, is the mass-customized management of populations at the level of the individual, without necessarily relying upon the attendant forms of subjectification that mark the disciplinary ‘pole’ of biopolitics. As Han (2022) puts it in his reflections on ‘infocracy’, ‘. . . disciplinary power gives way to smart power, a power that does not give orders but whispers, that does not command but nudges. In other words, it pokes us with subtle tools that influence our behaviour’ (p. 5). As we shall see, a range of imperatives including heightened acceleration and norms of efficiency can be built into the material and informational environment without necessarily being ideologically internalized. We may not consciously desire to continually accelerate our production of electronic communication, but the systems we rely upon make this process all but inevitable. Processes of subjective internalization can, in this respect, be displaced or bypassed by feedback in the physical and informational environment. Rouvroy et al. (2013) make a similar point in their work on algorithmic governmentality, which, they argue, ‘produces no subjectification, it circumvents and avoids reflexive human subjects, feeding on infra-individual data which are meaningless on their own, to build supra-individual models of behaviours or profiles without ever involving the individual’ (p. 169).

This shift marks a historical development in the deployment of biopower anticipated by the widespread deployment of automated forms of real-time identification at-a-distance, and, relatedly, in the development of customizable environments including virtual and augmented reality.

Background: pandemic security marketing

The official response to the pandemic in many jurisdictions highlighted the risks of circulation and of touch, helping to frame them as matters of public attention and self-care.

Thus, manufacturers promoting the convenience and efficiency of facial recognition responded to the pandemic as further incitement to a future permeated by smart cameras that do everything from securing our streets to eliminating house keys and credit cards. The industry’s strategic positioning of facial recognition and other biometric technologies in response to COVID-19 suggests the shape of what Kelly Gates (2011) has dubbed ‘our biometric future’ – although, as we shall argue, this future does not envision a uniform ‘us’ but rather a highly stratified one. Such technologies promise to transform our experience of shared and public space in ways that align with emerging forms of ubiquitous and comprehensive monitoring that structure much of the online world.

It is this version of the future that we encountered in the fieldwork described in this article: one in which physical space becomes interactive and deformable in ways akin to the real-time customization of the online environment. Facial recognition applications that vendors promoted at biometrics and security trade shows enabled the real-time monitoring and customization of physical space at the level of individual identification and access. Some systems coupled facial recognition with surface body temperature tracking to identify symptomatic individuals – and those with whom they had been in contact. Others tracked the circulation of employees through the workplace, linking their identity with relevant data about their rights of access. The picture that emerged was one in which control systems link automated identification with automated response tailored to individuals.

As the case of the pandemic response demonstrates, facial recognition technology plays a crucial role in this customization process, enabling automated systems to adjust the physical environment in real time to discrete individuals in a crowd. We locate these strategies of mass-customized securitization during the pandemic within a broader logic of governance ushered in by automated identification and response. In a world equipped with automated facial recognition, our faces become identifying metadata tags (or ‘headers’) linked to our actions, movements, and interactions. In combination with automated forms of response, ranging from turnstiles, smart billboards, and security systems, smart cameras facilitate the sorting, stratification, and management of public circulation in real time. The result is that individuals experience the same space differently – some are channelled in one direction (e.g. towards the VIP lounge) and others elsewhere (towards the cheap seats, the sidewalk outside Radio City Music Hall – or, perhaps, the waiting police officer; Hutchins and Andrejevic, 2021). Even the appearance of physical spaces can be customized, as in the case of Walgreens’s cooler doors that modify displays in real time according to the inferred demographics of individual consumers, (Kuligowski, 2019).

Often concerns about the rise of facial recognition technology are framed in terms of the loss or reconfiguration of privacy (Gray, 2003; Whitaker, 1999). The concept of privacy can be a vexed one: in many jurisdictions, we do not have a legal expectation of privacy in public spaces, where we can be seen by others, photographed, and captured on closed circuit television cameras (Rothenberg, 1999). There is in many contexts, however, a de facto sense of relative anonymity in public space. Facial recognition technology has the capacity to transform this expectation, invoking the prospect of mass identification in shared and public spaces. The findings presented here, based on trade show research conducted above 18 months, explore the dimensions of a potential future in which automated forms of biometric recognition become increasingly ubiquitous. We situate the findings from this fieldwork within our broader study of the current deployment of facial recognition technology, which draws on a range of approaches including a survey of the trade literature and interviews with stakeholders including industry representatives, government officials, and academic researchers.

We developed the notion of ‘granular’ biopolitics following an in-depth engagement with the marketing of facial recognition in response to the COVID-19 pandemic at industry trade shows. The industry shows provided some concrete examples of leading edge applications of the technology and how its benefits were framed in response to the pandemic – and to broader issues of securing the circulation of employees, travellers, shoppers, and clients. In order to frame the discussion of our research findings, we start with a brief overview of the context of our field work, which focused on the promotion of facial recognition technologies at security and biometrics tradeshows during the COVID-19 pandemic. We follow this with a discussion of what Rabinow and Rose (2006: 204) describe as the two ‘poles’ of biopolitics: ‘the molar and the molecular’ – the populational and the individual. We draw on this distinction to elaborate upon the notion of a ‘granular’ form of control that tailors strategies once targeted to the population to fit particular individuals. We use this formulation to describe a process whereby automated identification connects directly to the automated reconfiguration of the physical and informational environment. This formulation emerges from the findings in the fieldwork section, which identifies several inter-related logics of granular biopolitics, including the proliferation of borders and checkpoints that create additional sources of ‘friction’ addressed by automated identification and the real-time modification of physical and informational space.

Our goal in forwarding the concept of a ‘granular’ biopolitics is to supplement recent literature on crowd-level governance with a consideration of how individuals within a crowd can be identified and acted upon in real time (Lall and Wahba, 2020; Nishiyama, 2018). Intervening in crowd management at the individual level means linking biometric data collection with the customized reconfiguration of physical and informational space. This modality of governance presages forms of control associated with emerging technologies of virtual and augmented reality. The so-called ‘metaverse’, for example, envisions forms of control that operate in digitally customizable environments, uniquely shaped for each individual. Augmented reality pushes further, providing the physical world with an interactive overlay that loops back into physical modulations from customized informational displays to the individualized canalization of physical space.

The exercise of what might be described as ‘biopower-at-a-distance’ enabled by ‘smart’ cameras results in a reconfigured hybrid of disciplinary (bodily) and population-focused forms of control. However, it dispenses with the strategies of subjectification characteristic of the former – and with the shared or collective character of the latter. At the same time, as Epstein (2008) puts it, ‘. . . biometrics hits both at once the objectives of discipline and biopower. For it features as the ultimate individualizing technology; yet it is also deployed to regulate entire populations’ (p. 184).

Research context: the trade shows

In this article, we explore the role played by facial recognition technology in securing mobility through examples drawn from biometrics and security trade shows we attended (virtually) between July 2020 and December 2021, including the Biometrics Institute United States congress (the United States, 21–22 September 2020), the Biometrics Institute Global Congress (the United Kingdom, 7, 13, 21,27 October 2020), the Nippon Electric Company (NEC) Immersion + showcase (Australia, 22 July 2020), the Identity Expo (Australia, 3–4 November, 2020), the International Security Conference (ISC) West trade show (the United States, 5–7 October 2020), the Security and Government Expo (Australia, 12 November 2020), and the Second Annual Facial Recognition Summit (Malaysia, 7–8 April 2021). These events showcased existing technologies and previewed prototype applications for future use. We attended these trade shows in order to learn how facial recognition was being marketed for the purposes of surveillance and security in contexts including public safety, workplace management, and commerce. However, because of the timing of the research, we encountered the pivot of the marketing of the technology in response to the COVID-19 pandemic.

We registered for and attended all but one of the events online, because of COVID-19 restrictions, recording relevant sessions, and taking notes and screenshots. Participants in the trade fairs and congresses included commercial vendors as well as representatives from government, the academy, civil society, and law enforcement. Although these shows covered four continents, including Europe, the United States, Asia, and Australia, there were clear commonalities in the approach to using biometrics to secure circulation in the face of the pandemic and other perceived security threats. These commonalities – and the way their connection to more general strategies for managing circulation in shared and public space using facial recognition technology is the subject of this article. Because of the nature of the online presentations, we were not able to directly interview trade show participants, but we were able to record presentations and panel discussions, and to download supplementary materials.

All told, we accumulated close to 1000 pages of transcripts of presentations, discussions, and Q&A sessions covering a wide range of practices and issues associated with the past, present, and future deployment of the technology. We also collected links to supporting material including promotional brochures, presentation slides, and videos. Our guiding question, in embarking on this research, was to discover the emerging proposed and actual uses of facial recognition technology. We organized the materials according to shared themes related to the management and control of circulation. These inter-related themes form the bases for the four ‘findings’ sections on ‘frictionlessness’, temporality, the multiplication of borders/checkpoints, and the automated modulation of the physical environment.

Security trade fairs provide a useful site for exploring the actual and anticipated range of security and military technologies because they showcase a market largely invisible to the general public. Stockmarr (2015) highlights the importance of trade fairs in setting the terms according to which technological and market solutions identify and frame risks, since the exhibitions, ‘provide a window on how security technologies are described and sold by producers, and then adopted by users who frequently come to endorse the producers’ visions and understandings of security and risk’ (p. 187). In this respect, as Hockenhull and Cohn (2020) argue, the speculations and pronouncements presented at such events play an important role in the contemporary production of sociotechnical imaginaries (Jasanoff and Kim, 2015).

The development of automated face recognition dates back to the late 1950s (Perkowitz, 2020) but it has been accelerated by the rise of social media and the troves of images captured by large media platforms (Acquisti et al., 2014). Testing by the US National Institute of Standards and Technology indicate dramatic recent increases in overall accuracy (Crumpler, 2020), although issues of both bias and accuracy have accompanied the development of the technology (see, for example, Magnet, 2011; Pugliese, 2012 and Browne, 2015). A Georgetown Law study revealed that by 2016 half of US adults were in a police face recognition database (Garvie et al., 2016), and large numbers have images available online. The use of the technology is widespread in countries including China, Russia, and the United Arab Emirates, and Australia has assembled a nationwide database from official IDs (Bischoff, 2021). The commercial start-up Clearview AI claims to have built a database in excess of 10 billion face images, with the goal of reaching more than 100 billion globally. If your image has appeared on a publicly available website or social media post, the chances are that it is in Clearview AI’s data set (Harwell, 2022). Although other forms of automated biometric identification continue to be developed, face recognition has the advantage of being able to draw from existing data troves, both public and private. It can also rely on the installed base of existing CCTV systems. As Introna and Wood (2004) observe, facial recognition systems, ‘have become a prime focus for the security establishment’ because,

Unlike other biometrics facial recognition can operate anonymously in the background. The targets do not need to surrender their face image, as they would their fingerprint, or their iris scan. A face can be captured and (de)coded without the consent or participation from those being targeted. (p. 178)

The deployment of facial recognition technology follows a familiar pattern in which the conveniences and burdens are unevenly distributed. Like other surveillance technologies, facial recognition is often deployed for the purposes of policing and control over those with the least power in society. Sites including public housing projects, homeless shelters, and school are being subjected to the technology often without notification or deliberation (MIT Technology Review, 2020). By definition, granular forms of sorting and channelling circulation operate unevenly – and this unevenness reproduces and exacerbates existing hierarchies of power and control.

Quarantine light: preserving circulation

Given the timing of our research, it will come as no surprise that the trade show presentations focused on facial recognition as a tool for combating the spread of COVID-19. Biometric identification offers the perceived benefit of linking a body directly to an identity (unlike other device-based forms of tracking). Facial biometrics, in turn, emerged as the remote ID system of choice, in part because companies had already been developing the technology for a range of security applications in which the face doubles as both a means of identification and a locus of expression – and therefore of inferences about mood, intent, and interest (see, for example, Gray, 2003).

For the promoters of facial recognition, the technology offered a response to the pandemic that reinforced long-standing trends in mass-customized monitoring. Pioneering surveillance studies scholars Oscar Gandy (1993) and Gary Marx (1985) noted early on the role played by emerging technologies in broadening the reach and scope of monitoring to sort and categorize populations. As subsequent surveillance scholars have noted (see, for example, Lyon, 2003), this broadened reach coincided with a proliferation of categories for sorting the population into ever more narrowly defined categories. Recent scholarship has highlighted the forms of bias, exclusion, and asymmetric power associated with surveillance-based social sorting, noting how historical forms of oppression are reinforced by contemporary practices of data-driven monitoring and automated surveillance (Browne, 2015; Erwin, 2015; Pugliese, 2012; Ziadah, 2021).

The crucial additional aspect of what we are calling granular biopolitics is the connection between these forms of social specification and automated systems for customizing and modulating informational and physical environments in real time. Thus, automated face recognition literalizes what Torpey (2018: 44) describes as the ‘embrace’ of individuals by the state – and other monitoring and sorting entities. He draws on the etymological meaning of the term ‘embrace’ to suggest ways in which identificatory technologies allow authorities to ‘grasp’ or ‘lay hold of’ individuals – that is, not to ‘penetrate’ them (as in the sense of ideological control), but to encompass them. Although Torpey uses the term in a figurative sense, we are interested in the ways in which this embrace is materialized in physical form.

The pandemic provided a stark reminder of the economic impact of stasis. Restrictions on circulation coincided with the threat of an economic downturn: shops closed, airplanes mothballed, venues shuttered. The political fallout generated debates that pitted the health benefits of quarantine against its economic costs. The policy response in some jurisdictions, as well as the marketing of biometric solutions for securing mobility offered a response to this perceived trade-off: perhaps circulation could be reinstated if it was coupled with comprehensive monitoring to track contacts, monitor social distancing, and detect emergent symptoms. Even if some forms of human circulation, and by association, labour, had to be curtailed, the economy need not grind to a halt. Significant forms of informational and material circulation could continue unabated – or even at an accelerated pace – thanks, in part, to the immediacy of virtual interaction.

The technological promise of the first pandemic of the platform era was that always-on connectivity might facilitate forms of instantaneous intervention that would allow circulation to continue under conditions of enhanced monitoring and control. For example, workplaces that require in-person co-presence implemented employee tracking systems to monitor social distancing and maximum occupancy regulations (Trachtenberg and Boškov, 2020). Airports installed remote symptom monitoring systems coupled with portable facial recognition systems (Freed, 2021), and some jurisdictions used automated facial recognition systems to enforce quarantine (Reevell, 2020).

In keeping with biopolitical strategies, this approach treats the virus as ‘endemic’ – existing in the population, but potentially manageable through heightened hygiene measures, monitoring, and, when necessary, sequestration (Foucault, 2003). The coupling of identification and symptom monitoring at a distance became a strategic tool underwriting the biopolitical imperative of maximizing the productive powers of the population, while placing differential restrictions on that same population relative to disease prevalence, risk, and proximity. This economic imperative relies, as Foucault (2007) notes on, ‘making possible, guaranteeing, and ensuring circulations: the circulation of people, merchandise, and air . . .’ (p. 51). All of these became crucial concerns during the course of the pandemic, as lockdowns reduced consumption, disrupted supply chains, and directed attention to air filtration and circulation systems.

Granular biopolitics

The challenge posed for those who would deploy automated systems for managing productive forms of circulation is that physical space has yet to catch up with the ‘recognitive’ capacity of the online world. We cannot (yet) simply ‘log in’ to the physical world. We still need to pull out our IDs, our bank cards, and transit passes because the physical environment has remained largely ignorant of who we are and where we are going. The promise of rapid, mass identification ‘at-a-distance’ is to abolish this ignorance: to subject physical space and the activities that take place within it to the automated, real-time tracking available online. Our faces become machine readable meta-data for the physiological ‘packets’ of our bodies as these circulate through physical space.

The imperatives highlighted by the response to the coronavirus pandemic accelerate the process of real-time identification for the purposes of managing and regulating social interaction, contact tracing, symptom detection, and quarantine enforcement. Characteristic of remote, automated tracking is the introduction of digitally facilitated individuation into strategies for the control and management of circulation in the name of productivity and well-being. The goal is to couple techniques for managing risk at the population level – including, for example, routine disinfection, social distancing, and mask wearing – with increasingly individuated forms of identification, monitoring, and tracking. Automated systems provide the ability to determine, in real time, whether these routines are being followed by specific individuals and to respond accordingly.

Theoretical work on biopower influenced by Foucault’s ‘sketchy suggestions’ (Rabinow and Rose, 2006: 197) has emphasized the historical specificity of techniques for governing populations in the name of both biological and economic productivity (Hardt and Negri, 2000; Lazzarato, 2002) and the securitization of circulation (Aradau and Blanke, 2010). In the following analysis, we draw in particular on two key insights about biopower: first, that, as Rabinow and Rose emphasize, the specific forms taken by biopower are grounded in historical conditions (and vary accordingly), and, second, that multiple strategies and technologies of power co-exist without one regime entirely displacing the others.

With respect to circulation, we take up the challenge and the analysis proposed by Gilles Deleuze’s (1992) discussion of the transition between the ‘molds’ of a crisis-stricken disciplinary society and the ‘modulations’ of a coming control society. Deleuze (1992) invokes his co-author Félix Guattari’s description of an

imagined a city where one would be able to leave one’s apartment, one’s street, one’s neighborhood, thanks to one’s . . . electronic card that raises a given barrier; but the card could just as easily be rejected on a given day or between certain hours,

a city where ‘what counts is not the barrier but the computer that tracks each person’s position-licit or illicit -and effects a universal modulation’ (p. 7). Deleuze (1992) adds though that within this new diagram of power, ‘[i]t may be that older methods, borrowed from the former societies of sovereignty, will return to the fore, but with the necessary modifications’ (p. 7). It is within this uneven and opportunistic amalgam of modular, computational methods of real-time sorting, with the residue of ‘harder’, more archaic modes of punishment, confinement, and inscription, that we can find the articulation of what we term ‘granular biopower’. The fantasy of an infinitely and universally modulable ‘mesh’ of control mechanisms is accompanied by the reassertion of the significance of the barrier, of proliferating barriers and borders which ‘increasingly [crop] up in unlikely places – the laundromat, the grocery store, the bus station’ – and, we might add, the convenience store and the concert hall (Johnson et al., 2011: 65).

Attending to the historical specificity of biopolitics in the era of ‘smart’ spaces and streets, we consider the ways in which remote biometrics such as automated facial recognition reconfigures the relationship between governance at the level of the body and of the population. Michel Foucault’s (2003, 2007, 2008) work on biopolitics frames this distinction in terms of the relationship between disciplinary and populational regimes of governance. The former, exercised upon the individual, he describes as an ‘anatamo-politics’ (Foucault, 2003: 43) concerned with the individual-as-body, or as discrete, bounded unit. It manifests in the form of the regulation of bodily movements and dispositions: drilling, exercising, and standardizing (see, for example, Foucault, 2007: 248–249). In order to obviate the need for comprehensive surveillance, it relies on logics of subjectification – what Foucault describes in terms of the process whereby those subjected to the monitoring gaze become its bearers, adopting for themselves to its priorities and prescribed practices (Foucault, 1995: 314). At the populational level, by contrast, the focus is on overall statistical regularities: fertility, morbidity, and mortality rates, establishing standards and baselines for identifying irregularities and maximizing outcomes (Foucault, 2003: 43). The subsequent literature on biopolitics tends to treat these differing strategies of power as operating along a spectrum from the population to the individual (Rabinow and Rose, 2006). Nonetheless, they operate in distinctive ways at their respective levels. If a disciplinary focus on bodies draws on processes of subjectification, populational forms of governance operate through the environment, tending towards the modification of the ‘milieu’ that Foucault describes in terms of ‘environmentality’. Thus, intervention in the populational register takes place at the environmental and regulatory level: draining swamps, imposing hygiene guidelines, instituting health insurance systems, and so on. As Foucault (2003) puts it,

Both technologies are obviously technologies of the body, but one is a technology in which the body is individualized as an organism endowed with capacities, while the other is a technology in which bodies are replaced by general biological processes. (p. 249)

For Foucault (2003), ‘the two sets of mechanisms – one disciplinary and the other regulatory – do not exist at the same level. Which means of course that they are not mutually exclusive and can be articulated with each other’ (p. 250, emphasis added).

As technological conditions change, so too do modalities of governance. Consider, for example, the case of automated surveillance. Panoptic forms of control rely on the mobilization of uncertainty to amplify the monitoring gaze. In a pre-automation context, tracking an entire population in real time would be impossible (hence, the need for those under surveillance to become themselves the bearers of the monitoring gaze). However, with the rise of automated forms of monitoring (in the online space, for example), the amplifying role of uncertainty could become superfluous – the knowledge of being comprehensively monitored displacing the uncertainty that one might be seen at any particular moment. Combined with the prospect of automated intervention, even processes of subjectification might be bypassed. If an automated response can be calibrated to intervene pre-emptively, there is less need to rely on the internalization of standards of ‘correct’ behaviour. This temporality of automated pre-emption is implicit in the development of automated technologies designed to intervene pre-emptively in forms of anti-social or illegal behaviour by, for example, detecting when a fight is about to break out or when a theft is occurring (Hutchins and Andrejevic, 2021).

At the other end of the biopolitical spectrum, automated systems make possible the prospect of disaggregating populations. For Foucault (2003: 246) population-level biopolitics operates on regularities at the aggregate level (even when individual outcomes seem aleatory). Increasingly, comprehensive forms of data collection, however, focus on individual-level probabilities – an ongoing quest to isolate and contain the aleatory. Making sense of this level of data collection requires automated systems. Automation also enables the disaggregation of the ‘milieu’ and the customization of the ‘rules of the game’ in real time. In Foucault’s (2007) formulation, the milieu – the medium of population-level governance – is a shared space. The milieu is the context in which, ‘circulation is carried out’ (Foucault, 2007: 20–21). Foucault (2007) describes it in terms of, ‘natural givens rivers, marshes, hills – and a set of artificial givens – an agglomeration of individuals, of houses, etcetera. The milieu is a certain number of combined, overall effects bearing on all who live in it’ (pp. 20–21). As such, it is the ‘field’ in which a form of intervention distinct from the logics of disciplinarity take place, one in which, ‘instead of affecting individuals as . . . bodies capable of performances, and of required performances – as in discipline – one tries to affect, precisely, a population. I mean a multiplicity of individuals who are . . . bound to the materiality within which they live’ (Foucault, 2007).

Environmental forms of governance do not necessarily spell the disappearance of disciplinary control. Rather, the form of subjectification entailed by disciplinary measures is not crucial to their operation. It is the resulting combination that we seek to identify in the formulation of ‘granular biopolitics’: on one hand, individualized control without the imperative of internalization associated with discipline; on the other hand, the customization of the milieu.

Automated identification: the sociotechnical imaginary

We turn, in this section, to the trade show research to explore some implications of the automated securitization of circulation. In the presentations and literature we studied, the connection between passive, biometric data collection and automated intervention in the informational and physical environment is a recurring theme. Opportunity and security go hand-in-hand as means of promoting the potential benefits of the technology. In each of these sections, we draw on our fieldwork to explore the promise of ‘frictionlessness’; the related reconfiguration of operational ‘tempo’; and the ‘thickening’ and multiplication of borders associated with the exercise of granular biopower. We conclude with a consideration of how these developments align with automated forms of governance.

‘The new F-word’

The facial recognition industry has learned from the success of so-called ‘surveillance capitalism’ (Zuboff, 2019) that barriers to monitoring can be levelled by the promise of convenience. As the former CEO of ICANN put it, ‘It has been demonstrated time and again that individuals will trade privacy for convenience’ (Rainie and Anderson, 2017). In the face of perceived concerns about the intrusive character of passive biometric data collection enabled by facial recognition technology, the offer of convenience is framed as the royal road to its widespread normalization. As one participant in the Biometrics Institute’s US Congress put it, ‘It’s pretty difficult to argue against the proposition that convenience is going to be the driving factor in making all of this come to fruition and making the general population, if you like, the users of digital identity’. ¹ This promise of enhanced convenience – often paired with security – was a recurring theme in the trade shows, since many of the consumer-facing uses under consideration had to do with enabling secure transactions, whether speeding cash-free purchases or facilitating access to homes, offices, shopping and air travel.

Convenience, in turn, was repeatedly framed in terms of ‘seamlessness’ – the ability to interact and transact speedily and with minimum use of any credentials other than one’s own face. As the CEO of a company called FindBiometrics put it, ‘Friction has become the new “F-word,” as a smooth user experience has taken priority, bringing with it better customer and employee satisfaction with lower administration costs’. Since seamlessness refers to passive forms of data collection, it shifts control over the authentication process away from individuals and towards the institutions that control the sensor infrastructure. Automated seamlessness relies on smart cameras and other sensors to capture biometric data at a distance and match it to stored databases. Shops, turnstiles, doors, rooms, and various other checkpoints would continuously scan individuals as they move through the course of their day. People may not even realize that information has been captured about them until some type of intervention or interaction takes place. ² In an echo of ‘friction-free’ capitalism (Shade, 2020), friction refers to anything that slows the pace of transaction, interaction, access, or detection.

Early in the pandemic, when vectors of contagion remained uncertain, seamlessness was coupled with the imperative of touchlessness. Speed and hygiene were aligned in promoting ‘distanced’ forms of monitoring. Doorknobs, handles, and buttons as well as swipe cards and cash – all the things we need to touch for access or to conduct transactions – were to be avoided. As a marketing representative from a company called Cognitec put it,

Using your face instead of fingerprints, ID documents or other tokens not only speeds up access control procedures, it also minimizes the contact with surfaces. More people are wearing gloves, especially in hospitals, doctors’ offices and elderly homes, and would like their face to easily unlock doors, computers and phones.

An executive from a company called IDEMIA made a similar observation, noting that passive recognition contributes to a frictionless workplace by allowing people to verify their identity: ‘. . . without having to touch something or hand your credentials to somebody for manual validation, it . . . enables that kind of smoother workflow with you know, a less touched environment’. He noted that the pandemic had led to a, ‘a surge in this type of surveillance market, similar to what we saw post-911’. The comparison is a suggestive one, insofar as both moments mark a concern with the hazards of circulation, both address the spectre of the undisciplinable, and both underwrite the proliferation of monitoring, tracking and intervention only possible through automated systems.

‘Operational tempo’

Automation and acceleration go hand-in-hand in the contemporary context, although there is nothing inherently speedy about automated processes in the abstract (Wajcman, 2019). In practice, however, automated technological processes are developed and deployed within social contexts where they are shaped by the imperatives of those who invest in and control them. Under current conditions, as Armitage and Graham (2001) argue, in their discussion of the political economy of speed, the imperative is to accelerate processes of circulation and exchange: ‘International trade and its imperatives for ever-accelerating productive activities is the organising logic of the “globalised” society’s tempo’ (p. 11).

The promises of ‘frictionlessness’ and ‘seamlessness’ deployed by the promoters of facial recognition technology align themselves with this accelerated temporality. If in spatial terms, there is an ongoing fixation upon the problem of ‘the last mile’ (in delivery and transport systems), in temporal terms, the goal is to subtract the surplus seconds from routine interactions and transactions (Lee and Whang, 2001). At the Biometrics Institute Congress, for example, one participant described the ability of facial recognition to accelerate the ‘operational temporality’ of identity detection and verification. This term applies more broadly to the potential of the technology in the workplace, in retail outlets, and in security and transport applications. No more waiting in checkout lines, no more swiping IDs, transit cards or passports.

The goal of increasing the pace of operational temporality was redoubled by the pandemic, which added an element of risk to any form of interaction that involved being around other people for a prolonged period of time. The inconvenience of crowded lines and the ‘friction’ of keypads, touchscreens, and even pens took on an added element of risk. As the executive in charge of biometrics applications at Visa put it during the Identity Week trade show,

The past two years have been all about implementing contactless because it was seen to be faster . . . So I think the pandemic helped to spur the movement towards contactless that was already in motion anyway. It dovetails very nicely.

The self-reinforcing progression of, on one hand borders and checkpoints and, on the other hand, technologies for dealing with them efficiently, marks a convergence between the online world and the physical world. Indeed, there is a sense in many of the vendor presentations of the attempt to enable the physical world to keep pace with the informational. The Internet, combined with the expansion of network bandwidth and the exponential growth of processing power, has not only accelerated the deployment of data, but also the conduct of a range of activities that depend on it. As an IDEMIA executive at the ISC West trade show put it, ‘we’ve seen over the past several years, a stronger and stronger adoption of, you know, how do I do things faster? How do I do things with less contact?’

Acceleration is equated with seamlessness, convenience and efficiency. The recurring example in the trade shows drew upon one of the more familiar established uses of facial recognition technology: the deployment of smart cameras to enable automated identity verification at border crossings. This hallmark of globalization heralds the reduction of friction points for the movement of some people (and, conversely, the rapid detection and detention of others). The model of ‘frictionless’ border crossing helps normalize the technology in the name of convenience, as a participant in the Biometrics Institute Congress suggested,

I think indeed the public becomes more and more acquainted with the use of their biological traits with a face being used for easy access to their phone or for maybe crossing the border. So, in general, it’s presented to the public in a way that it can be easy and smooth and can offer alternatives for long waiting lines.

The accelerated circulation facilitated by seamlessness and frictionlessness evokes the chronopolitics of resilience described by Chris Zebrowski (2019). Speed as resilience works, ‘to diminish the emergency’s temporal duration, geographic scope and destructive potential so as to secure a speedy return to “normality”’ (p.160). If a symptom can be detected as quickly as possible – perhaps even anticipated – contagion can be prevented or limited. The IDEMIA executive emphasized the importance of regularly tracking individuals’ vital signs to discern potentially significant deviations:

the challenge becomes how do I do that without you having to wear a smartwatch or a, you know, electronic ring . . . how do I do that at a distance, you know, with people that are just . . . passively coming in.

As previous work on pandemics suggests, the surveillance dimensions of disease containment address, ‘a deep anxiety about the timeliness of response’ (French and Mykhalovskiy, 2013: 175). The response to this anxiety is, ‘an immense effort to detect, pre-empt or rapidly respond to health events to prevent them from having trans-local effect’ (French and Mykhalovskiy, 2013: 175). Managing heightened levels of circulation, then, requires increased data collection and decreased reaction time – both addressed by the promise of automation.

Multiplying borders

The corollary of seamless circulation is the deployment of efficient tools for sorting populations in order to discriminate between those permitted and denied access. The paradox is that securing seamlessness relies on the proliferation and expansion of borders and checkpoints. Some jurisdictions, for example, have been creating photo databases of undocumented immigrants to identify them using facial recognition wherever they might be (Harwell, 2019). The use of these and similar technologies for passive, at-a-distance identification re-inscribes borders wherever they are deployed: ‘the body, in effect, becomes the carrier of the border as it is inscribed with multiple encoded boundaries of access’ (Amoore, 2006: 347–348).

The ‘normalization’ of facial recognition as a border crossing technology is repeatedly invoked in the trade shows to support the extension of facial recognition technologies to realms beyond international travel. Wherever the cameras crop up, new borderlands emerge. NEC’s I:Delight facial recognition system, for example, erects biometric checkpoints at the entrance to office buildings, shops, and office suites (NEC, 2020). Those without the proper clearance – or who might be manifesting pandemic symptoms – can be identified and denied access or service. Enterprise solutions for the workplace offer systems that track employee density and alert people to avoid ‘hot zones’ that are already at capacity. Juniper Networks, for example, has partnered with NEC to combine its motion tracking systems with facial recognition technology to ‘reduce risk in return-to-work’ (Bau, 2020). The borders traced by these smart systems can be erected around individuals who have been identified as having pandemic symptoms, such as fever, so they can be isolated and their past movements traced to determine which spaces need to be quarantined or cleaned, and which contacts notified.

Quarantine and other forms of contagion control (such as the mobility restrictions imposed during lockdowns) erect invisible, malleable borders that require ‘seamless’ and efficient control. In the midst of the pandemic, a Russian company called NtechLab developed a system called ‘FindFace’ that was used in Moscow for contact tracing and quarantine enforcement. In South Australia, authorities contracted with a company that uses global positioning system (GPS) tracking and facial recognition to determine whether people are complying with self-quarantine requirements (Jarrah, 2021). An NEC executive observed that the pandemic resulted in people’s newfound acceptance of ‘conditions on their access to stadiums and sports events, to some shopping centres and other places’ – conditions that rely on multiplying checkpoints rendered ‘seamless’ by automated recognition systems and temperature sensors.

The expanding deployment of facial recognition systems envisions the stratification of physical space with an array of malleable borders – and their associated checkpoints – for sorting people in the name of risk and opportunity. For example, the CEO of a company called TrueFace noted that in addition to forms of automated access restriction, facial recognition technology provides ‘opt-in’ opportunities for sorting people in real time: ‘You know, it’s the fast movement in line at the airport or expedited check-in at a stadium or hotel’. This is a technology that is already being developed in casinos and sports stadiums to detect and exclude risky individuals while targeting special offers and enhanced access to big spenders (FindFace, 2021; Hutchins and Andrejevic, 2021). The network of borders envisioned by ubiquitous forms of passive identification requires linking identification and verification systems to the real-time, automated modulation of physical space.

Modulated space

One of the striking features of the ISC West conference was the number of sessions – three in total – devoted to automated doors. Many more sessions focused on ways to link physical access points such as turnstiles and elevators to facial recognition systems. More broadly, various ‘access solutions’ were a recurring theme in the trade shows, highlighting the convergence of security, convenience, hygiene and mobility in pandemic-era marketing strategies. Making sense of these developments we build on McCosker and Wilken’s (2020) observation that machine vision systems rely on the development of ‘powerful new platforms that connect and operationalize visual data’ (p. 30) by considering how the visualization process connects with the automation of physical space. ‘Operationalization’ in this context suggests that imaging technology is not simply a matter of representation, but of automated response. As Harun Farocki (2004) puts it in his description of ‘operative images’, these are images that act: they become ‘part of an operation’ (p. 17).

Operationalizing facial recognition ‘seamlessly’ relies on the construction of physical infrastructures for sorting flows of both people and information in real time. This infrastructure may include new physical checkpoints – such as turnstiles to enter elevator foyers in office buildings. Since the ability to sort the movements of individuals in real time depends on the deployment of automated infrastructure, facial recognition is frequently paired with existing access systems including automatic doors, elevators and turnstiles. At the Second Annual Facial Recognition Summit, for example, a representative from the Schindler Elevator group presented the case study of a high-rise Singaporean office building that has installed 22 high-speed elevators connected to facial recognition access points. The system requires all tenants and visitors to enrol their faces in a system that records the floors to which they have access. People approach the elevators through an entry portal not dissimilar from those through which international travellers pass for face verification. During times of high traffic, the portals manage the incoming flow to ensure the efficient allocation of elevators and to prevent crowding. As the company’s promotional material puts it, ‘by allocating passengers to individual floors at controlled, efficient times. The user never has to touch a thing, the entire journey is done using intuitive sensors and their personal smartphone’ (Boon Edam, 2021).

The result is what might be described as the high-resolution rationalization of spatial mobility. Against the background of this level of automated control, ‘dumb’ doors and elevators look somewhat anarchic: once someone is allowed into a building, they might go to any floor they choose – perhaps creating a security risk. Schindler’s system not only controls access, but keeps records of who travels where and when, in order to predict traffic flows and the movement preferences of individuals – ‘remembering’, for example, which floor someone usually goes to, and when they are likely to come and go (Analytics India, 2020). As in the case of many of the other technologies presented at the trade shows, the advantages of this system were framed not just in terms of efficiency (seamlessness), but also hygiene: the entire process can take place without individuals having to touch elevator buttons (or anything else) at any point during their journey.

More generally, the ‘access control industry’, which builds infrastructures for secure buildings and spaces, pivoted towards facial recognition during the pandemic. As the website for one of the industry leaders, Southwest Solutions, put it: ‘Preventive measures such as facial recognition with a mask and temperature detection have become standard options for clients inquiring about access control. Biometrics has served as a great partner in monitoring entrance control safely and efficiently’ (Southwest Solutions, 2020). The company has built biometric access systems for office buildings, airports, retail outlets, courthouses, and universities.

This level of investment suggests that biometric ‘solutions’ are likely to outlast the pandemic. They are already being framed as an improvement upon practices that, in hindsight, appear distressingly unhygienic, anticipating the future spread of other endemic diseases, such as the seasonal flu. To the extent that such systems achieve the goal of easing and speeding access, they will also be able to capitalize on the promise of convenience. At the same time, they shift the balance of control over identity and circulation to those who operate the access controls and manage the link between bodies, borders and checkpoints.

To say that the body becomes the carrier of the border is not just to suggest that checkpoints rely on biometric information to make access decisions, but also to invoke the ways in which biometric identification links data about a range of activities to these decisions. The fact that someone is automatically identified at a pandemic exposure site, for example, could be used to flag them as a risk when they approach access points at shops, office buildings, or apartment complexes. Activities that take place far away from physical checkpoints can be folded into the decision of how such checkpoints channel and sort populational flows, extending the reach of the security barrier. By the same token, as automated checkpoints multiply (when a growing range of doors get ‘smart’), they can collect and share information with one another to regulate access. As Pallister-Wilkins (2016) notes, ‘As devices of data capture, security barriers come to produce the data that are often used, at a later time or in another place, to govern movement and wider (in)securities’ (p. 158). In a self-stimulating spiral, the checkpoints help generate the data needed to facilitate their more granular management of circulation.

From pandemic to endemic

The marriage of identification-at-a-distance technology with other forms of biometric tracking sets the stage for a pandemic response that blurs the distinction between epidemic and endemic disease management. The international response to the spread of COVID-19 in different nations oscillated between these strategies. The quarantine lockdown in Wuhan, China, for example, where people were kept in their homes for several weeks to stop the virus in its tracks, followed epidemic protocol (Burki, 2021). In Sweden, by contrast, the focus was on the preservation of physical circulation through social distancing, mask wearing and enhanced hygiene rather than quarantine (Davies and Roeber, 2021). The attempt to preserve the rhythm of daily circulation treated the virus as endemic: something that, even while it existed in the population, could be managed through changes in behaviour and infrastructure.

Situating mass-targeted forms of population management within the history of biopower outlined by Foucault highlights the unique aspects of such an approach enabled by the development of automated identification. In his lectures on the emergence of biopolitics, Foucault (2003) highlights the process whereby the population becomes an object of political management – in the name of heightening its productivity and maximizing a ‘state of life’ (p. 247). The goal is to treat the maximization of the population overall as a political and biological problem, which takes shape against the background of emerging forms of datafication that are in turn conditioned by the imperative of economic growth and stability. Foucault locates the emergence of the population as a problem in the second half of the 18th century, but its ongoing salience is revealed by the response to COVID-19 which turned on the relationship between the biopolitical management of the population and, as the media and policy response emphasized, the fate of the economy. As pandemic restrictions closed down businesses and unemployment rose, the public and policy debate over various pandemic response measures framed the issue in biopolitical terms: not as a trade-off between the need to save the economy and to save lives, but as an assessment of whether the economic impact would have a more devastating impact on the population than the virus (see, for example, Bazelon, 2020 and Rotman, 2020).

Tellingly, Foucault (2003) emphasizes the focus of biopower not just on the exceptional moments of the ‘famous epidemics’, but on, ‘what might broadly be called endemics, or in other words, the form, nature, extension, duration, and intensity of the illnesses present in a population’ (p. 243). Biopolitics encompasses an ongoing process of environmental engineering. As Foucault (2007) puts it, tracing the history of biopolitical forms of intervention, ‘it will no longer be the problem of exclusion, as with leprosy, or of quarantine, as with the plague, but of epidemics and the medical campaigns that try to halt epidemic or endemic phenomena’ (p. 10).

This development is not so much a response to changing health circumstances – the threat of plague always loomed – but to the changing economic role of the population with respect to the imperative of economic growth. Technological and social developments in the realm of transport and production helped define the population in relation to the growing reach and interdependence of trade and commerce. Problems of ‘territory’ (fixing and demarcating it) were supplemented by those of mobility:

of allowing circulations to take place, of controlling them, sifting the good and the bad, ensuring things are always in movement, constantly moving around, continually going from one point to another, but in such a way that the inherent dangers of circulation are cancelled out. (Foucault, 2007: 37)

The defining biopolitical issue raised by the pandemic then, is how to manage circulation – that is, how to craft the ‘medical campaigns that try to halt epidemic or endemic phenomena’ (Foucault, 2007: 10). The shock of the COVID-19 moment is that it reintroduced the spectre of plague-era quarantine and resulting forms of productivity-threatening stasis.

Conclusion: post-populational logics

In his lectures on biopolitics, Foucault (2008) discerns signs of, ‘a massive withdrawal with regard to the normative-disciplinary system’ in which ‘discipline-normalization’ is displaced by interventions that modify ‘the terms of the game, not the players’ mentality’ (pp. 261–262). Environmental-level governance does not require the standardization of behaviour – or even thought – as long as changes in the ‘rules of the game’ result in an overall population-level effect.

As Foucault’s analysis suggests that the response to COVID-19 fits within the more general trajectory from exceptional response to endemic management. The proliferation of sensor systems that combine identification with biometric information facilitates the generalization of pandemic logics. A growing range of illnesses (and other forms of perceived risk) can be subjected to the logic of granular biopolitics – even as the statistical measures that simultaneously define and disaggregate the population are enhanced. General measures of morbidity and mortality become increasingly specified, encompassing smaller and smaller groups – resulting in customized risk scores and individualized probabilities. These, in turn, require customized, real-time response: targeted environmental modulation and pre-emption. The pandemic moment thus helps illuminate the model of granular biopolitics anticipated by the prospect of real-time mass-customized monitoring. This is a diagram of power that shifts the balance of control away from the self-disciplining subject (and away from the shared milieu) and towards forms of automated, pre-emptive intervention. In this respect, the pandemic helps provide an answer to the question of what a society characterized by the widespread deployment of automated facial recognition technology might look like: one that embraces, ‘the dream of targeted governance’ (Valverde and Mopas, 2004) through forms of automated recognition that initiate the process of individualized, real-time response. This is the biopolitical lesson of the pandemic response: the invitation to ‘return to normal’ through automated at-a-distance technologies envisions a reconfigured version of space – both virtual and physical – traversed by digital borders that micro-manage and disaggregate the population. Automated governance is not simply a matter of information systems and the decisions that flow from them, but of the real-time, automated modulation of physical space.