Epidemic and Insurance: Two Forms of Solidarity

Introduction

While epidemics are a millennia-old reality, epidemiology is a science born in the 19th century. The World Health Organization defines epidemic as ‘the occurrence in a community or region of cases of an illness, specific health-related behaviour, or other health-related events clearly in excess of normal expectancy’ (World Health Organization, 2008: 26, emphasis added). Epidemic would thus imply a statistical knowledge of the population, and the determination of thresholds as key indicators for the detection of outbreaks.

Yet epidemiology remains also anchored in medicine and the reality of clinical cases. This original combination of individual and collective knowledge is the object of this paper. More precisely, if one admits that statistical tools gave shape to the population as an object of knowledge and government (Foucault, 2009) and that insurance mechanisms exemplify this new governmentality (Ewald, 1991), epidemics offer a point of resistance to this statistical mindset. I would like to suggest here that the governance of epidemics challenges commonly held conceptions of community and solidarity by forcing the individual and intersubjective level within statistics. This challenge is very much in line with the disruption introduced by new algorithms in traditional perceptions of the collective (Barry and Fisher, 2019), that is also shaking insurance (Barry, 2020).

Solidarity is often defined in Stjernø’s terms as ‘the preparedness to share resources with others by personal contribution to those in struggle or in need through taxation and redistribution organized by the state’. It is the expression of a ‘political altruism’, that is currently jeopardized by ‘the triumph of markets’ (Stjernø, 2005: 2). Solidarity is thus strongly associated with welfare and social rights, that Ewald (1986) has demonstrated to be propelled by insurance mechanisms. Epidemics by contrast imply the sharing of other vital elements, one of them being the virus that circulates despite our will. This kind of imposed sharing displaces the meaning of solidarity in a manner that needs to be scrutinized. Because the insurance pool is focused on the collective level and the averaging that erases individual specificities, its solidarity is traditionally understood both as social rights and as a collective duty towards the individual. The epidemic by contrast forces us to consider individual duties towards a collective suddenly endowed with rights of its own; it thus engages us to reconsider the crucial balance between collective interest and individual freedom within the concept of solidarity (Stjernø, 2011: 157; 2005: 85–9; see also Baldwin, 2005: 10). ¹

This is not to say, however, that epidemiology or insurance practices are each limited to one conceptual form of solidarity: indeed, as soon as epidemics became a public concern they triggered welfare responses (Cohn, 2011); reciprocally, Ewald shows that early 20th-century concepts of insurance solidarity and welfare are inseparable from Pasteur’s discovery of contagion, and the collective consciousness of the mutual risk to which we expose one another (Ewald, 2011: 79).

This is not to say that solidarity would be limited to such abstract and conceptual definitions; Brunkhorst (2005) has shown how the modern concept, with roots in both Roman civic citizenship and Judeo-Christian ethics of brotherliness, has been politicized as an egalitarian and secular project in the 19th century. Solidarity became (in Europe at least) a norm for action for social democratic and Christian democratic parties alike, with an accent on redistribution for the former and on personal responsibility for the latter (Stjernø, 2011: 169; see also Wilde, 2007). This paper does not discuss the possible conceptual link between epidemic solidarity and any of those traditions, although one could argue that it calls for a displacement of our ‘norm for action’. Its scope is limited to isolating the kind of social bond that comes into play in epidemics in contradistinction to insurance, and its historical formalization by epidemiology.

The first section shows how the epidemic is scientifically re-constructed in the 19th century at the intersection of statistics and clinical medicine. The next section details how epidemiology thus sets up a conception of the collective in relation to, but also as a counterpoint to, the insurance pool, therefore implying a different kind of solidarity. I then examine how this conception guides the first statistical models of contagion, showing that epidemiology precociously attended to model the individual within the collective. The last section focuses on the management of contagion as a social phenomenon, through tracing algorithms, highlighting a paradox: although epidemiological and algorithmic outlooks on collective phenomena are conceptually congruent, tracing apps have remained largely debated and seldom adopted. This debate is usually presented as an ethical dilemma between fears of surveillance on the one side and the containment of the epidemic on the other. It also highlights, I suggest, that digital apps usually nurture a responsibility for oneself, rather than the civic responsibility for the community that the epidemic solidarity requests. The widespread reluctance to use these technologies for the sake of epidemic containment thus illustrates, I argue, a confusion between the two kinds of solidarity found in insurance and epidemics.

The Epidemics’ Prism on the Individual and the Collective

This section adopts the perspective of illness as a social and historical construct (Carmichael, 1991; Coleman, 1982; Conrad and Barker, 2010; Foucault, 2003a; Rosenberg, 1989), with a specific focus on epidemics. Despite some recurring ‘modernist biases’, public health measures to counter contagion did exist in the premodern era (Geltner, 2019: 11). Yet I argue that the statistical understanding of contagiousness at the end of the 19th century displaces the imaginary of the relationship between the individual and the collective in epidemics. In order to do so, I briefly expose below how contagious diseases were perceived, before and after statistics imposed their gaze on collective phenomena and their government.

Until modernity, the perception of epidemics as a divine punishment of collective and/or individual sins seems to remain constant (Carmichael, 2008: 21): the plague that descends upon Thebes in Sophocles’ Oedipus Rex is the sign of a moral fault, and the wrath of the gods that it unleashes (Kousoulis et al., 2012). Healing is a purification that requires the identification of the criminal, the culprit that must be found and exiled. The miasma, or pestilence, as Hippocrates would also say, is seen as the vector of the disease through smell (Paillard, 1998). Similarly during the Justinian plague, Pope Gregory evokes ‘God’s wrath’ (Little, 2006: 30). Divine punishment is seen again as causing the Black Death in the Middle Ages (Carmichael, 2008: 21), and its recurring waves until at least the 16th century (Cohn, 2011: 78).

The divine will does not mean passive acceptance, though (Geltner, 2019: 30). Cohn (2011) thus mentions hundreds of medical tracts written in Italy from 1348 to 1600, ² exemplifying remedies and treatments proposed by doctors and physicians, with recurring references to (either to confirm or reject) Hippocrates’ and Galen’s writings (Cohn, 2011: 9–32; see also Nutton, 1983: 16–17). Cohn interestingly notices a shift in those tracts with the 1575–8 plague wave: doctors do not address the patient anymore but the prince, thus transforming the epidemic from an individual to a political threat (Cohn, 2011: 90). Governing the plague now meant (non-exhaustively) building hospitals and lazarettos, thus separating the sick from the healthy (Bashford, 2016); establishing quarantines and provisioning large populations deprived of work and freedom of movement (Cohn, 2011: 106); cleaning streets, disinfecting houses, and killing domestic pets and animals suspected as carriers (Cohn, 2011: 82); more traditionally, organizing fasts and processions. ³ Because plague circulates with trade, so do these practices. Varlik (2015) thus depicts the spatial entwinement of the Ottoman Empire’s expansion and the propagation of plague. Similarly, Bashford points to the convergence and increasing uniformity of quarantine knowledge and practice in ‘a network built over interconnections of trade and human movement’ (Bashford, 2016: 11). ⁴

The choice of suitable political measures actually depended on finding the vector of contagion (Carmichael, 1991). Was the disease transmitted by contact or through corrupted air? Such questions fill the debates among doctors during the 15th and 16th centuries (Nutton, 1983, 1990). If contact with an infected person was the cause, isolation was necessary, opening the way for the stigmatization of specific, supposedly ‘infecting’ groups: the Jews for plague, the soldiers and prostitutes for syphilis, the poor, etc. (Cohn, 2012). Concomitant to the will of understanding and controlling the propagation, death registers started to be established; ⁵ Carmichael (1991, 2008) situates the first necrologi in Milan around 1450. Prior to that (during the Black Death), private doctors might have kept some records, but not in a systematic way (Cohn, 2011: 65). Earlier on, as concerns for instance the Justinian plague, beyond evocations of ‘large scale mortality’ (Carmichael, 2008: 19; see also Kousoulis et al., 2012: 156), almost no quantified information exists (Le Goff and Biraben, 1969: 1499; see also Little, 2006).

The mortality bills established during London plagues allow John Graunt, towards the end of the 17th century, to exhibit regularities in collective phenomena (Bayatrizi, 2008; Franklin, 2015: xv; Hacking, 2006; Le Bras, 2000). The population thus emerges as an epistemological object, thanks to data collected for the sake of containing epidemics. This statistical rationality affirms that aleatory events cannot be predicted or managed at the individual level but exhibit some regularity when considered in the aggregate (Ewald, 2011). This statistical gaze was central to the development of insurance mechanisms (Ewald, 1986; Foucault, 2009: 76–9; Hacking, 1990), ⁶ but also to public health and working-class hygiene in the 19th century (Coleman, 1982), and later to what Armstrong (1995) calls ‘surveillance medicine’ that investigates the population to determine risk factors and statistical models of the probability of illness. Epidemiology takes shape in this context, as a statistical knowledge of epidemics that significantly displaces their conceptualization.

However, because epidemiology combines statistics with medical knowledge it has to address some specific resistances. During the 19th century, opposition to statistical medicine was voiced by the clinical school that insisted on the patient’s centrality to medical practice (Foucault, 2003; see also Armstrong, 1995: 393; Desrosières, 1998: 82). From the perspective of this paper, I would like to reductively oppose here collective versus individual approaches in the study of epidemics. On the one hand, hygienists were proponents of the aggregate view. A case in point is Virchow who claimed, against contagion theory, that the main cause of the 1848 typhus epidemic in Upper Silesia was the dire living conditions of large parts of the population (Schütz, 2020), hence it could be contained thanks to social, economic and educational measures alone (Azar, 1997: 70). On the other hand, John Snow’s ⁷ study, focused on an 1854 cholera epidemic in London, demonstrates that the cause of the disease is found in the pollution of drinking water. In his essay, Snow highlights the geographical distribution of cases around a specific water pump (Snow, 1855: 46). But he is not satisfied with this global approach; he also details the symptoms and gives a biography of the infected individuals, aimed at understanding the spread of the disease (Snow, 1855: 14–15).

Epidemic versus Insurance Solidarity

Snow’s example highlights the fact that the aggregate view propelled by statistics is in part inadequate to grasp the contagion processes of an infectious disease. This conflict between two views on the social probably reached an apex with Pasteur’s theory of microbial contagion at the end of the 19th century. For Latour, the Pasteurians indeed imposed a change of scale that ‘renewed medicine without ever taking disease as the object of study and renewed politics and hygiene without ever taking the poor or the social outcast as the unit of analysis’ (Latour, 1988: 104, emphasis added). The theory of microbial contagion thus triggered a renewed questioning on the nature of the social (Latour, 1988: 35). My argument here is that epidemiology’s focus on contagion with a statistical gaze also propels a distinct conception of the population and of solidarity that is best contrasted to how insurance statistics grasp the same objects.

At its core, insurance consists in the pooling of a shared uncertainty. Building on the regularity of collective phenomena, it assesses the overall risk cost on the group and shares it between its members, thus creating a solidarity between them. In Durkheim’s terms, the accident is therefore treated as a social fact, which can be managed in collective terms. Thus, risk ‘characterizes an aggregate that retains its identity despite the variations that may affect its parts [. . .] the regularity of risk is independent from the conduct of individuals’ (Ewald, 1986: 285, personal translation, emphasis added). Individual specificities are indeed averaged out in the process of displaying the collective regularity. It is therefore the group that bears the burden and responsibility of random events. Solidarity finds expression in the capacity of the group as such to compensate for individual misfortunes: the group is perceived as what saves its members, so to speak, from the randomness of accidents.

The epidemic, by contrast, is the typical case of an uninsurable systemic risk. One of the first epidemiologists indeed defines contagion in the following probabilistic terms: ‘the chance of occurrence of fresh cases in any year increases with the number of cases which have already occurred in that year’ (McKendrick, 1926: 102, emphasis added). In the epidemic, the statistical smoothing that is crucial for insurance does not take place; cases add up – and accordingly increase individual risks – without their ever cancelling each other out on the average. Moreover, the group does not provide protection but risk exposure, and solidarity is an inescapable fate rather than an opportunity. The epidemic thus constructs the population as a group of individuals who are identically exposed but are also crucially dependent on each other’s risk. By contrast, insurance mechanisms, secured by the law of large numbers, fundamentally assume independent and identically distributed individuals.

Reflecting on the meaning of community, Roberto Esposito contrasts communitas, or the sharing of munus as both gift and duty, with immunitas, or the modern social contract as a way to avoid the obligation of the munus (Campbell, 2006; Tierney, 2016a). Although reductively, this conceptual distinction can fruitfully be applied, I suggest, to the two kinds of solidarity and interdependence at stake in this paper. The munus, that etymologically holds both communitas and immunitas together, is a specific kind of gift, distinguished by its obligatory, non-reciprocal character: ⁸ ‘although produced by a benefit that was previously received, the munus indicates only the gift that one gives [. . .] it is a ‘pledge’ or a ‘tribute’ that one pays in an obligatory form’ (Esposito, 2009: 5, emphasis added; Campbell, 2006: 4). The sharing of the munus in communitas thus creates the community on an essential lack – the common being ‘not what is proper but what is improper’, and on a negation of the subject – as ‘subjects do not find “in the community” a principle of identification’ (Esposito, 2008: 50; 2009: 7).

The epidemic easily fits with the communitas as an imaginary of the collective where the loss of self (Tierney, 2016a) is a constant possibility: like the munus, the virus is a ‘constitutive danger of our co-living’ (Esposito, 2009: 8), passing from one individual to the next, threatening their identity and very existence (Tierney, 2016a: 60). Moreover, the loss of control implied by contagion forces the metaphor of life as an impersonal flow, that Esposito refers to Deleuze in his own definition of life as bios: the living is ‘an impersonal singularity which [. . .] rather than being imprisoned in the confines of the individual, opens those confines to an eccentric movement that “traverses men as well as plants and animals independently of the matter of their individuation and the forms of their personality”’ (Esposito, 2008: 194). In this strand of thought, the virus brings to the forefront life as an impersonal force that challenges the modern concept of the subject (Esposito, 2012; Saidel, 2014).

For Esposito, the conceptual frame of the modern polis since Hobbes has been the paradigm of immunitas, that lifted (or masked?) the threat imposed by the communitas (Esposito, 2008: 45–77). Immunitas is a kind of communal life that, just like the immunity from the disease, keeps the individual safe from the threat of the common. Community as immunitas holds, however, an antinomic tension at its core: it protects the community by negating the communitas and restoring ‘the borders that were jeopardized by the common’ (Esposito, 2008: 50). Protection from others is indeed obtained by instituting an absolute, transcendental sovereign, that negates the immanent circulation of the munus in communitas (Esposito, 2008: 60–1). Moreover, the modern basic rights all derive from this immunitary logic: with individual property, the ‘appropriating subject’ creates itself in the act of appropriation of ‘one’s own that is not common’ (Esposito, 2008: 63–8, emphasis added). Locke hence contends that property starts with one’s own body – in complete denial of contagious disease processes. As a ‘condition and outcome of the process of appropriation’ (Esposito, 2015: 116–17), modern liberty is likewise ‘that which ensures the individual against the interference of others’ (Esposito, 2008: 72, emphasis added). ⁹

Interestingly, Ewald contends that welfarism and insurance solidarity redefined the social contract ‘in opposition to the individualistic vision, originating in Hobbes or Rousseau’s natural law’ (Ewald, 2011: 79; see also 2020: 138). Yet the insurance pooling rests on a hypothesis of initial independence that denies the danger inherent in the munus of the communitas, and thus remains very much in line with the methodological individualism incepted by Hobbes. Moreover, although Léon Bourgeois had microbial contagion and mutual responsibility in mind when he defended ‘solidarism’ and the institution of social rights (Ewald, 1986: 361–2), those ended up strengthening the individual protected sphere. Solidarity has indeed become a right, ‘the democratic realization of individual freedom’ (Brunkhorst, 2005: 3, emphasis added; see also pp. 59–60). As Esposito puts it, society now ‘associates individuals through their functional separation’ (Esposito, 2011: 87, emphasis added). Similarly, Supiot contends that national social security systems propel an anonymous and financial kind of solidarity, that makes of the social state the only debtor (Supiot, 2002: 16). Mistaking the welfare state for a large insurance company, Stjernø (2005: 341) confirms, contributes to undermining ‘universalism, redistribution and solidarity’.

This antinomy between two kinds of solidarity has found an interesting expression in the historical debate between Tarde and Durkheim on the meaning of the social (Candea, 2016). In Durkheim’s holistic view, the social remains external to the individuals it connects; as Karsenti (2016: 99) puts it, for Durkheim ‘the proper way to reach the whole is not to build it up from the parts, but to grasp it in a qualitative leap, by fusion or integration’. Durkheim adopts therefore a statistical gaze that discards individual variations for the sake of a sociological viewpoint. For Tarde by contrast social phenomena are best understood as intersubjective effects of imitation (Tarde, 1979) – also defined as ‘the epidemiology of ideas’ (Latour, 2010). Durkheim’s holistic view is rejected as a form of mysticism, as it grants the social a life of its own (Candea, 2016: 37). Latour interestingly compares Tarde’s stance to epidemiology:

Tarde does for social theory what Pasteur had done in epidemiology: in the same way as bacteriology allows one to move from a regional theory of miasmas to a point-to-point and person-to-person theory of contagion through a specific vector (cholera bacillus, Koch’ bacillus, etc.), Tarde moves from an aggregated cloud of collective qualities to a highly specific point-to-point, person-to-person ‘contagion’ of ideas each of them having its own peculiar effectivity. (Latour, 2010: 163, n. 9, emphasis added)

There is therefore on the one side a top-down approach, which holds that social facts are graspable at the population’s level of reality, and on the other side, a bottom-up approach which contends that social reality is built up from individual interactions. As far as insurance mechanisms are concerned, the condition of possibility of risk measurement is the aggregate, top-down approach; epidemics by contrast cannot be understood without keeping an eye on each specific case as a potential vector of propagation. Interestingly, the 19th century solidarity evoked by Ewald appears as a combination of both forms of solidarity and communities; it indeed promises insurance solidarity, underscoring the newly gained social rights of each member of the pool. But it cannot escape the drawbacks of its methodological individualism: for this reason, insurance is sometimes considered as encouraging license, or moral hazard (e.g. Ericson et al., 2000). Not surprisingly then, Bourgeois’ solidarism borrowed from the metaphor of the epidemic to emphasize the duties of each person in the collective and praised a secular morality inspired by the categorical imperative (Ewald, 1986: 361–2). Likewise Durkheim (2002) strived to balance the threat of individualism by his ‘moral individualism’.

A Genealogy of Epidemiological Models

In the purely statistical approach, one ignores individual variations to gain an understanding of the aggregate. Porter (1996: 85) thus contends that ‘19th century statisticians liked to boast [that] their science averaged away everything contingent, accidental, inexplicable or personal, and left only large-scale regularities’. Epidemiology, by contrast, cannot efficiently reflect the reality of the contagion with this statistical smoothing that disregards the individual specificities; in the debate between Durkheim and Tarde, the epidemiologists would therefore take sides with Tarde. Yet the epidemic, paradoxically, only takes shape with a collective gaze. As Greenwood put it: ‘the epidemiologist’s unit is not a single human being, but an aggregate of human beings’ (Greenwood, 1935: 15, emphasis added).

This tension between the individual and the collective is perceptible in epidemiological models and their historical elaboration. The genealogy proposed below highlights the constant negotiation between these two levels to be held together. From the outset, these are global models, often very theoretical, whose aim is to account for the population, or assess the impact of specific public health policies. But unlike classical statistical models, they cannot do without the individual level (Jacob, 2010). An epidemic is thus characterized by some factors, such as the R0, the basic reproduction rate, or the number of people that an individual will in turn contaminate; can such a single figure be valid for the whole population, like Quetelet’s ‘average man’? Nothing is less certain. This genealogy shows the striving over the years for the construction of a global vision in a bottom-up approach, where the bottom cell gets refined with technological capacities.

One of the first mathematical theories of epidemics was developed by Ross, a British physician working on malaria at the beginning of the 20th century (Heesterbeek, 2002; Smith et al., 2012). Ross highlights a critical mosquito density rate for the spread of the disease, without translating it into a critical reproduction rate. ¹⁰ But the main family of the current so-called compartmentalized models results in fact from some work initiated by a pupil of Ross, McKendrick, and one of his close collaborators, Kermack. Together they developed the SIR model in a seminal 1927 paper (Kermack and McKendrick, 1927). The model calls for dividing the population into compartments (susceptible, infected and recovered) and modelling the transition from one to the other. The authors thus demonstrate that based on disease-specific parameters (in a very simplified manner – an indicator of the duration of the disease in the infected person, and the infectivity rate), there is a population density threshold below which the disease does not spread. The model also makes it possible to visualize the evolution of the epidemic, with its peak followed by a return to normal. Moreover, depending on the parameters of the disease, only a given proportion of the population will be affected.

This deterministic model segments the population into three sub-groups and thus seems to have abandoned the individual level for the sake of uncovering these macro issues. For the epidemiologists, however, this aggregation constituted its main limit, since it did not give a sense of the importance of individual contacts in the potential spread of a disease. To do this, stochastic models generalizing the Kermack and McKendrick equations were needed. McKendrick (1926) had actually drawn the lines of such a model in a 1926 paper, but real pioneering work in stochastic modelling was performed in the 1950s, notably by Bailey (1950), Bartlett (1949, 1953, 1956) and Kendall (1956). Bailey’s rationale for this work was that ‘a considerable degree of chance enters into the conditions under which fresh infections take place’ (Bailey, 1950: 193), hence the need for stochastic models that would account for the randomness of contagion. Moreover, Bailey adds, simply modelling the probability of infection by a global ratio of infectious to susceptible does not reflect local differences; in his view, the level of communities (the city) is preferable to the national population. Even at this local level though, infected individuals vary in their level of contamination since their contacts can range from 10 to 50 people, depending on their activity. The complexity that the stochastic model tries to account for thus comes from the necessary articulation of the individual dimension – infected persons’ contacts with others – and the global dimension of the epidemic. At the individual level, or when the population under consideration is small, stochastics cannot be dispensed with and allows for a reliable construction of the global level (Bartlett, 1956: 82).

In a review of models published in 1952, Serfling complains about the poor quality of the data and the over-simplification of the models (deterministic and stochastic): the parameters should be further refined to account for cultural, social, environmental, seasonal, or other determinants in the frequency of contact (Serfling, 1952: 163). Models of this kind have actually become available since the early 1970s. Criticizing existing models, Elveback et al. (1976) propose to add population compartments in order to account for the multiplicity of contacts and types of propagation according, in particular, to the age of the individuals:

While oversimplification is an inherent characteristic of epidemic models, we have reduced it considerably by introducing families and neighborhoods as specific mixing groups; by permitting the length of latent and infectivity periods, occurrence of illness, and response to vaccination to be random variables; by maintaining the identity of each individual; and by allowing young children to behave as young children. The basic model [. . .] can be used for the study of any agent dependent on person-to-person spread. (Elveback et al., 1976: 153, emphasis added)

Their model is stochastic: each individual contact may or may not result in an infection, depending on a random draw parameterized by a transmission probability specific to each type of interaction (estimated on past influenza epidemics). The aim is therefore to reconstruct a global vision from the micro dimension of a theoretical community (1000 people in this model), an approach reproduced in more recent studies (e.g. Longini et al., 2004, for 2000 people).

In the latter part of the 1970s, models were finally developed that account for the spatial structure of propagation (Mollison, 1977: 314), a point that in 1956 seemed difficult to resolve (Bartlett, 1956: 195). Subsequent refinements result from increasing technical capabilities, but conceptually at least, current models seem to be in place since that time. Noteworthy are recent models that attempt to reproduce the spatial structure of propagation based on the actual geography of a country (Longini et al., 2005, for 500,000 people in Southeast Asia; Ferguson et al., 2005, for Thailand; see also Ferguson et al., 2006, for the United States and Great Britain, with a model of air travel in the United States); these are stochastic, individual and geographic models of the population. Built for the sake of comparing alternative public policy actions, these models seem to obliterate the micro dimension, which is nevertheless present: the goal is to assess the overall impact, by reconstructing bottom-up its individual components. Until very recently at least, these models did not capture individual differences, but rather assessed the chance of an infection within a given segment defined by its matrix of interactions (e.g. an age group, or a geographical area).

Epidemiological models thus attempt to describe in a mathematical way the complex articulation of the individual and the collective. The population is not taken as an abstract whole transcending the individuals that make it up; the collective is here nothing other than the aggregate of its parts, hence dependent on its parts. A case in point is a March 2020 paper by the Imperial College Covid-19 Response Team (Ferguson et al., 2020), which is often considered – together with the Chinese response in Wuhan – a milestone for the global confinement that followed (Flahault, 2021: 26). The researchers use an individual and geographical stochastic model similar to those described above, and compare the impact of different non-pharmaceutical intervention policies on selected indicators, such as the total expected number of deaths, and of intensive care beds needed – suddenly turned critical (Di Domenico et al., 2020; Ferguson et al., 2020; Massonnaud et al., 2020).

The authors distinguish between mitigating policies, which aim to attenuate the peak without altogether preventing it, and more drastic measures to suppress the disease in the short term (without being able to prevent a second peak after the measures are lifted). Each of the social distancing measures considered in the paper is associated with a compliance rate, as if to reflect an inevitable drift in its practical application. This again highlights the relevance of individual behaviour for the understanding of epidemics. For Surico and Galeotti (2020) contagious diseases are indeed particularly sensitive to ‘externalities’: differences in exposure of different subgroups, or even differentiated individual application of the instructions are paramount to the overall result.

Global measures are therefore in a way incommensurable with the micro-reality of the virus and its transmission on the one hand, and discrete individuals on the other, contagion always starting with specific ‘primary’ and ‘index’ cases (Giesecke, 2014). In this strand of thought, some recent models focus on the dispersion around the mean of the propagation number; some epidemics might be explained by a few ‘superspreaders’, that bear most of the contagion (Lloyd-Smith et al., 2005; Riou and Althaus, 2020). Here again, the ‘average based approach’, so central for insurance solidarity, is criticized for its inaccuracy. This incommensurability between global prophylactic measures and the very granular reality of contagion further implies a paradox. Compared to previous major pandemics, Covid-19 indeed occurred when the technological capacity to isolate specific individuals is mature, as shown by the proliferation of tracing apps, thus in theory reducing the need for global measures. Yet, as will be shown below, these apps are also perceived as most infringing upon individual rights, illustrating the complex relationship between containment practices and their political implications (Baldwin, 2005: 524 –63). Epidemics thus continue to trigger ‘some of the most fundamental and perduring dilemmas in the contradiction between individual rights and the demands of society’ (Baldwin, 2005: 10).

Prospective Digital Epidemiology

There is a deep congruence between the epidemiological outlook on collective phenomena described in the previous section and current algorithmic technologies of personalization (Lury and Day, 2019). One of the specificities of new algorithms as compared to traditional statistics is indeed their bottom-up approach (Barry and Fisher, 2019), and their capacity to isolate individual patterns as anomalies and exceptional cases. Working on these anomaly detection algorithms, Aradau and Blanke (2018) interestingly demonstrate that they reverse traditional statistics; instead of averages and large numbers, these techniques aim at detecting outliers, or points that were usually discarded as not fitting into the normal curve. For Latour (2010: 159), big data and digital traceability thus mark Tarde’s vindication of Durkheim.

As set out in the previous sections, with epidemiology also, the group never transcends its part in a holistic, Durkheimian manner, and remains the aggregate of its individual components. Epidemiologists’ recommendation for the containment of epidemics has thus traditionally been the isolation of infected cases and the tracing of their contacts for quarantine. This practice was successfully applied on yellow fever in the 19th century (Cohn, 2011: 139), or later on with tuberculosis (Roberts and Buikstra, 2013: 15) and typhus (Sansonetti, 2020). Contact tracing and testing has also been the experts’ preferred strategy against Covid-19 (Di Domenico et al., 2020: 12; see also Ferguson et al., 2020: 15). The flourishing of contact tracing apps in the context of this pandemic should therefore not come as a surprise. They range from governmentally run applications to privately owned ones, from centralized to decentralized schemes, and full location or limited encrypted data solutions (Montjoye et al., 2020). Yet the controversies that accompanied their potential use crystallize, I suggest, the issues and confusions between epidemic and insurance solidarities.

The most emblematic application – if not the most discussed – is the one proposed by Yoshua Bengio and his team. They offer a tool for assigning a risk score both to persons, based on their own contacts and meetings, and to locations, based on their risks concentration. This double scoring would allow users to ‘know’ their risk and optimize their trajectory to avoid further exposure. On Bengio’s blog, one reads:

Imagine that an app in your phone would keep track of the probability that you are infected based on where you have been and the encounters you made and would share that risk information with people’s phones you encounter so their app could update their own risk estimation. (Moayed and Bengio, 2020)

Regularly updated data would allow continuous training and dynamic adjustment of the model to the situation and the person. Even from this very sketchy description, one can guess that it couples digital tracing, epidemiological models, and state of the art artificial intelligence algorithms to calculate scores. By contrast with more simple apps, the aim is indeed no longer to give a ‘contact-case’ alert to users, but to assign a score to each person. Conceptually, this is equivalent to attributing a probability of infectiousness on an individual basis. The dichotomous split into risky/non-risky is here replaced by a continuous assessment. It further exemplifies Ewald’s (2011) distinction between risk and data: aleatory events used to be perceived as risks, that is pooled and measured at the collective level; but as data, they are assumed to predict individual results.

This tracing app is relevant in the perspective of this paper for a couple of reasons. First, by individualizing the probability of infectiousness, it seems to bring the endeavour of the epidemiological stochastic gaze to its end. The app indeed materializes the bottom-up construction of the population and its intersubjective interactions. Second, the community is, however, not perceived as communitas, but remains anchored in the immunization paradigm and its methodological individualism, as will be shown below.

The proponents of the app indeed highlight that it promotes an accountability where each person is responsible for the risk he or she runs:

Imagine that when you meet someone, you are able to know what risk of contamination they carry, so that you can choose to keep your distance, or not let them into your shop, house or car [. . .] the aim is not to blame or identify, but rather to provide citizens with the information they need to minimize the risk of being contaminated by the virus. (Moayed and Bengio, 2020, emphasis added)

From this perspective the app conveys the idea that the others are a risk to oneself, against whom one should seek protection. Moreover, as with other scoring devices, infection scores thus bear the risk of ostracization: individuals who are algorithmically found as riskier than others would suddenly become the ‘others’ of the system (Aradau and Blanke, 2018), echoing ancient practices of purification.

Responsibilizing individuals for their health, a direct result of methodological individualism applied to public health, has already been observed for non-contagious diseases. Infectious ones would then enter this area of health that is perceived as ‘a sort of transactional zone between political considerations for the nation’s physical fitness and personal techniques of selfcare’ (Rose, 2001: 3; 2007: 223). The conflation of contagious and non-contagious diseases masks, however, a significant gap between being held responsible for one’s own health and the actual responsibility for the others’ health that occurs in the epidemic. Hence the paradox of current debates around the use of tracing apps; some refuse to be traced in the name of basic human rights of freedom and privacy, as if not using the app was a risk they were willing to take for their own self alone. In this strand of thought, it is very symptomatic that the tracing apps are most often presented as a means to protect oneself, and very rarely as one’s contribution to the protection of the community. Reversing the immunization paradigm during the epidemic would require seeing oneself as a danger for others, and not the other way around.

Conclusion

Epidemics appear to lie at the heart of the emergence of statistical tools and the crystallization of the concept of population in the 19th century, while also imposing a point of resistance to this aggregate approach. With epidemics, the actuarial and global models, in which some cases are compensated for by others, do not hold, nor does insurance solidarity as the pooling of hazards. The control of contagion requires indeed the detection and isolation of cases, giving full importance to individual specificities that statistics traditionally tended to ignore. In this respect, the epidemiological view, that already in the 1950s strived to impose stochastic models of individual behaviors, seems in perfect conceptual congruence with current big data technologies’ bottom-up approach to collective phenomena.

Digital tracing apps are thus the point of encounter between the new technologies and the tradition of epidemiological tracing, testing, and isolating practices. The acute public debate around their use – some journalists speaking of a ‘coronopticon’ as a generalized state of surveillance pretexted on the pandemic – exposes the sensitivity of the subject (‘Countries are using apps’, 2020). It invites us to rethink the strengths and weaknesses of the algorithms. Their major strength lies in the leverage effect associated with targeted measures, in an environment where resources are scarce – and in the case of Covid-19 tragically so – and where global lockdowns have dramatic socio-economic impacts. But it is also their main weakness, or their danger; for the more effective the technology will be in differentiating and setting apart, the greater the temptation will be to divide as if between pure and impure, these old categories that always border on the imaginary of disease.

They also demand to question anew both the practice of scoring and the imaginary of self and others it propels. While health scores usually nurture a discourse of responsibility as selfcare, the reality of the propagation reminds us of a form of solidarity long forgotten: it is Esposito’s communitas as common dangers and duties, rather than the immunitas where each member stands immune from the risk of others. In this strand of thought, the debates around the tracing apps crucially enlighten the ‘modern immunization paradigm’, according to which community and solidarity are conceived as deriving from independent, dissociated actors, as ‘men associated in the mode of reciprocal dissociation’ (Esposito, 2009: 27; see also Tierney, 2016a). Waiving privacy and individual rights against the containment of contagion thus reveals the confusion of epidemics with the experience of private sickness under insurance solidarity.

In this context, Esposito’s ‘affirmative biopolitics’ can fruitfully dispel the confusion. In Esposito’s reading, Foucault’s concept of biopolitics indeed oscillates between an affirmation of life and its negation (Esposito, 2008: 32–44), leading him to a puzzle: how can a power that founds its legitimacy on the protection and expansion of life have actually propelled so much death (Foucault, 2003b, 260; 1990, 149)? Introducing the munus as the unilateral gift and obligation toward others is Esposito’s specific contribution to the reflection on biopolitics (Tierney, 2016b), that also helps to answer Foucault’s puzzle. Both poles actually rely on the immunization paradigm, where protection is granted either by excluding/exterminating in the case of Nazism or, if one admits that the welfare state belongs to the affirmative pole, by assuming independent subjects of rights and risks. A truly affirmative biopolitics for Esposito cannot be obtained without reconciling the community with the communitas, without redefining life as bios, as what is ‘constitutively improper, and for that reason common, as pure difference can be’ (Esposito, 2008: 193). The concept of munus also helps to grasp the kind of solidarity at play in the pandemic: not only, not mainly the pooling of resources of otherwise independent individuals, but also a solidarity where property (of one’s health, or one’s data) gives way to the common as what is opposed to ‘the proper, private and immune’ (Esposito, 2013), as what truly belongs to no-body.