Abstract
During the course of the COVID-19 pandemic, a wide range of digital technologies and data analytics have been incorporated into pandemic response models globally, in the hope of better detecting, tracking, monitoring and containing outbreaks. This increased digital involvement in disease control has offered the prospect of heightened effectiveness in all of the above, but not without raising other concerns. This paper contributes to ongoing discussions of the digital transformation in disease control by proposing a materialist analysis of how such control has become operative and what its effects may be, both now and in the future. Using Taiwan's digital pandemic response as a case study, the paper explores specific ways in which material processes and arrangements have shaped digital measures, as well as the actions that rendered such measures operable, with their ensuing consequences. This analysis illustrates the importance of historical, material and technological specificities and contingencies to our understanding of how digital disease control takes a particular shape. It also demonstrates how shifting regimes of practice continually reconfigure the ways in which digital disease control functions. The paper argues that paying greater attention to the materialities of digital disease control can provide a more nuanced understanding of the complex ways in which society may be protected or harmed by its use, possibly simultaneously. It is hoped that such increased attentiveness may inform more considered and careful preparation for subsequent pandemics.
Introduction
The rapid and widespread movement of COVID-19 and its uncertain but lingering impact, have prompted the world's governments and corporations to turn to ‘digital disease control’, that is, the deployment of digital technologies and data analytics to detect, trace, monitor and contain the virus and the individuals it infects to maintain public health. Corporations were quick to respond to COVID-19's fast-moving spread and the corresponding opportunities for their technology. In May 2020, Fitbit launched its own symptom tracking and early detection study to explore which indicators would prompt its wearers to seek medical attention before their symptoms worsened. 1 Google and Apple jointly developed an exposure notification framework to assist in contact tracing. The development and promotion of these tools were taking place while the governments of countries such as Canada, Denmark, Ireland and Japan were releasing their own official contact tracing apps. 2 In an evaluative study of a ‘big data analytics’ approach to COVID-19, Taiwanese health and medicine agencies suggested that, for national governments, a digital approach to disease control can be ‘quickly constructed through an innovative technology system to support timely epidemic analysis’ (Chen et al., 2020, 7).
Clearly, digital disease control did not start with the advent of COVID-19, and will not stop there. Since its inception, it has been trialled in a number of different ways. Academic and corporate research has explored the feasibility and effectiveness of digitising ‘influenza surveillance’ (Broniatowski et al., 2013; Farrahi et al., 2014; Nsoesie and Brownstein, 2015). Social media data, search engine statistics, mobile phone tracing, wearable devices and the crawling and scraping of webpages are among some of the technologies and data sources that public health and computing science researchers have been investigating for their potential to identify, predict and contain contagious outbreaks. With the pursuit of ‘digital epidemiology’ (Salathé et al., 2013) or ‘digital disease detection’ (Vayena et al., 2015), public health research is trying to leverage these data and technologies to enable the early detection of outbreaks, ongoing epidemic assessment, continuous monitoring of behavioural and emotional changes related to pandemics and retrospective examinations of relevant events or reporting prior to outbreaks. All of this is being done in the hope of developing anticipatory and flexible systems for dealing with future pandemics.
These digital initiatives do not come without corresponding costs to society and individuals. During the COVID-19 pandemic, the promise of digital disease control has been critically examined with regard to these costs and to the effectiveness of digital measures overall (Kitchin, 2020). In an analysis of the Norwegian contact tracing app Smittestopp, Sandvik (2020) points out several factors that have prevented the app from being more widely adopted, including design, battery usage and concerns about security and privacy protection. Sandvik also reflects critically on how trust and suspicion have been affected by issues associated with the roll-out of contact tracing apps; for example, the occurrence of false positives and negatives in digital symptom detection, the personal and societal implications of diagnosis and the imaginaries and data sharing practices whose invocation relates to fear of surveillance.
The wider societal consequences of digital disease control have also been discussed. As Cinnamon (2020) cautions, discourses about the contribution to global health and the ‘public value’ of effective digital platforms and data can be used to extend the political and economic interests of the tech industry. Corresponding to this concern, how digital disease control measures treat privacy, security, surveillance and civil liberties is crucial to the protection of public health and society. Public health researchers Bengio et al. (2020, e342) made a plea in The Lancet for their colleagues in the field to consider seriously the potentially damaging consequences of digital disease control and to note that well-intended measures to ensure public health can lead to the normalisation of state surveillance, intrusion on civil liberties and an ‘impact on individual privacy that democratic societies would normally consider to be unacceptably high’. Indeed, Datta (2020) and Maalsen and Dowling (2020) have demonstrated how the accelerated appropriation of existing digital and data technologies for virus containment has extended state and corporate surveillance into the home, health and private life of individuals, both under quarantine and otherwise.
Given the urgency of the ongoing pandemic situation, responses such as those mentioned above have tended to focus on the digital and the present. As a result, insufficient attention has been paid to the processes through which specific histories, infrastructures, spaces and practices give particular shape to digital disease control and its consequences. This paper aims to broaden the discussion by demonstrating that recent theoretical developments in the analysis of materialities of data and information can provide a more nuanced understanding of how digital disease control comes about and what concerns have arisen in the process.
The contributions of this paper to the discussion are twofold. Conceptually, the paper will foreground the materialities of data and information to examine a diverse range of material arrangements and practices that render digital disease control effective. Here, its analysis draws on recent work (Dourish, 2014, 2017; Dourish and Mazmanian, 2013; Gabrys, 2016a; Mackenzie, 2017) to pay close attention to the processes of arranging data, technologies, practices, regulations and protocols that make digital disease control operable.
Empirically, the paper charts the deployment of digital disease control measures in Taiwan before and during the course of the COVID-19 pandemic, including how Taiwanese government agencies and the public have been involved in devising and implementing digital measures to contain the virus. In response to a number of successive challenges from various highly contagious viruses over recent years, Taiwan has built up its digital pandemic prevention and response capabilities. These ongoing efforts provide rich resources for a case study of how such capabilities have accumulated and expanded in response to COVID-19. The analysis draws on a range of sources that include governmental reports, documents, research publications and public announcements regarding the technical architecture of digital disease control management. In addition, as a result of the difficulty of conducting interviews during the pandemic, commentaries and reports on quarantine published by news agencies have been consulted to illuminate a range of experiences of living under digital disease control.
The paper proceeds with an overview of the materialities of information, before demonstrating how digital disease control has been put into practice in Taiwan. This is followed by a discussion of the key aspects of a materiality approach to examining the use of digital technology in disease control.
Data, digital technologies and materialities
It is important to attend to the materialities of data and information to understand how they reconfigure society. Bowker (2005) demonstrates that the specific set-up of information infrastructure can have serious consequences for the ways in which knowledge is conceived, structured and produced. To rethink the production of scientific knowledge, Gitelman (2013) examines the situated and material conditions in which data are generated and aggregated. This approach has subsequently inspired a material-technical analysis of social connections on digital platforms. For example, Helmond (2015) further unpacks the ‘work that [such] platforms do’ in framing these connections and, inevitably, extending beyond them. The technical framework of a platform configures the types of interactions that can occur on that platform, while also shaping interactions on other websites through plug-ins, embedded code and application programming interfaces (APIs). These material-technical configurations shape the specific interactions and connections that can happen and the ways that data about them can be circulated and monetised. More recently, Thomer and Wickett (2020) observe a shifting model of database construction in which ad hoc database practices are more prominent than a normative view of ‘best practices’. These changes represent a deliberate attempt to materialise the replicability and transparency of research by redesigning database construction in ways that support open science and scientific communication.
The remaking of the social and material worlds through data generation and analysis is another important aspect of the materialist approach. Gabrys (2016a) examines the concrescence of computation and worlds by analysing the practices through which digital sensors are attuned towards particular physical, biological and social processes to collect and compute data about them. These sensing and computing practices do more than just produce knowledge; instead, they actively engage in ‘in-forming’ and co-creating worlds in conjunction with the geological or biological processes being monitored. Pink et al. (2016) reject the dichotomy of digital versus material worlds. They argue that the two are not separate entities that may or may not intersect; nor do the authors employ the term ‘digital materiality’ to name new empirical objects. Instead, they construe digital materiality as ‘a process, and as emergent, not as an end product or finished object’. They are motivated by the exploration of the entanglement of data, digital technologies, humans and material objects and environments; consequently, they pay attention to ‘the making and to what emerges of these entanglements, not to a state or a quality of matter’ (Pink et al., 2016, 10–11; see also 2018).
Dourish and Mazmanian further propose a materialist approach to data, information and algorithms (Dourish, 2014, 2016, 2017; Dourish and Mazmanian, 2013), with a particular focus on specific practices to trace how they construct knowledge, organisations, worlds and certainly data-driven computing systems (Latour, 2004; Law, 2015; Mackenzie, 2006, 2017). This approach is also heavily influenced by new materialism, particularly the work of Barad and Bennett, which considers the generative capacities of non-humans to act and interact with humans or among themselves while configured in entangled and continually evolving processes. Accordingly, a materialist approach to the examination of data, information, or algorithms requires the development of a sensibility with two interrelated aspects. One is to understand how these three elements take physical, discursive and metaphorical forms and how they become part of digital assemblage through practice. The other is to explore the material manifestations, effects, operations, reconfigurations, lines of action that result from data taking particular forms and the specificities of becoming part of digital assemblage. An exploration of these areas is not limited to the study of the new, but can incorporate historical, cultural and material contingencies to understand how they shape, maintain, reconfigure and repair data and related systems (Gabrys, 2016b; Pink et al., 2018; Thomer and Wickett, 2020). With this approach, research into the materialities of data is sensitised to the historically evolving forms that information takes. This assists its practitioners to deconstruct multiple forms of information, analyse the many scales, abstractions and practices through which information is realised and disclose the competing interests, demands and accounts of present and future to which sociotechnical information systems contribute (Dourish, 2017, 201–211).
A materialist approach to digital disease control is thus drawn to analyse how digital innovation manifests in disease control practices, along with the particular ways in which materialities enable, restrict, or compel the shaping of digital disease control and public health. Such an approach explores practices that assemble data, digital technologies and various human and non-human actors to form disease control measures, while tracking the material consequences, lines of action and subsequent reconfigurations necessary for these measures to remain effective. To provide an illustrative example, the analysis below examines how a digital turn in disease control took shape and became operable in Taiwan before and during the COVID-19 pandemic.
Programming digital disease control
Using Taiwan as a case study, this section draws on Gabrys’ concept of programmability to trace and track the ontogenetic processes through which the materialities of data, infrastructures and practices combine and set in motion a wide-ranging ‘formation of events, spaces, and things’ that engineer a safe and healthy society (Gabrys, 2016a, 11–12). This analysis demonstrates how the programming of digital disease control operations has been shaped historically, while also illustrating how these operations use a more emergent, collaborative, or ad hoc form as they figure out how to ‘put the technology to work to meet […] needs’ as they arise (Dourish, 2017, 103).
As mentioned earlier, the digitising of disease control in Taiwan did not start with COVID-19 and has its own historical and material specificities. Digital capacities were gradually built up owing to intermittent but sustained epidemic outbreaks of viruses such as SARS in 2003 and the Zika virus in 2016. Since SARS, Taiwan's government has engineered real-time monitoring systems to support preparation for and decision-making during epidemics. Technical specialists and researchers at the Taiwan Center for Disease Control's (TCDC) Epidemic Intelligence Center have established a cloud-based, automatic case reporting system that aggregates data, including test results from medical centres, hospitals and national laboratories, as well as electronic medical records and National Health Insurance (NHI) claims, in order to monitor virus dissemination and predict potential outbreaks (Jian et al., 2017). Open data portals were also set up to provide automatic daily updates of cases for certain diseases and visits to emergency departments with relevant symptoms, such as respiratory distress. Furthermore, since 2003, infrared thermal imaging scanning systems have been deployed at major airports and harbours, and all passengers arriving at these ports, regardless of nationality, are required to fill in paper forms to declare personal health conditions.
When COVID-19 started to spread, these technologies were either already up and running or could be enhanced by new data streams and measures to identify symptomatic passengers and trace those who might have become infected during transit. A variety of digital operations that inherited or built on historically developed capacities for virus containment were added or modified for this purpose (see an indicative list in Table 1). To assist in case identification, recent immigration records were added to the NHI database to alert doctors to patients who presented relevant symptoms and had a history of travel to seriously affected areas (Wang et al., 2020). Subsequently, further digital measures were implemented to facilitate contact tracing and monitoring. As shown in Table 1, measures such as entry quarantine and fever detection were being implemented before COVID-19, although the health declaration form for entry quarantine became digital only during the pandemic. This form records whether incoming passengers have had fever, cough and shortness of breath, or whether they have visited high-risk countries, during the 14 days before their date of entry to Taiwan. The data acquired were used to identify and classify high-risk cases (Wang et al., 2020). Infrared thermal cameras were also deployed at airports, and later at domestic transportation hubs, to monitor travellers’ body temperatures, but without recording, linking to, or storing personal data.
Digital technology and data in disease control led or supported by Taiwanese government during COVID-19.
The big data or data-driven approach deployed globally during COVID-19 marks an important shift in the response to pandemics, as has been noted by public health officials and other commentators, all of whom have focused on the need and capacity to extract epidemiological insight from a large number of data directly or indirectly related to the pandemic. Omitted from these accounts, however, are the materialities of these data and its implications for the analyses and operations that can be performed. To better grasp how digital disease control has taken shape amid COVID-19, the remainder of this section explores contact tracing, resource distribution and quarantine measures. This exploration uncovers how critical public health demands, data infrastructure and practices coalesce in the process of making digital disease control operable.
One instance of large-scale contact tracing by Taiwan that has received considerable attention is that of the Diamond Princess, which has often been showcased by Taiwanese government officials for its use of big data analytics. For the purposes of this paper, however, the significance of this case lies in its illustration of how data infrastructure simultaneously allowed and limited the tracing of more than 3000 cruise passengers and all those who might be affected by a 1-day excursion to northern Taiwan.
The case of the Diamond Princess exemplifies the processes involved in adapting existing data infrastructure to the circumstances of a new disease. The development of a system for tracing the passengers unfolded as follows: the cruise ship docked at a northern Taiwan port on 31 January 2020. During its journey to Yokohama, Japan, there was a COVID-19 outbreak on board. This meant that tracing those who might have come into contact with the passengers at Taiwanese tourist destinations became an urgent task for the government. While the epidemiological need for this information was clear, the practical means by which it could be obtained were less so. Contact tracing apps were not in use and in fact had barely been developed; therefore, alternative methods had to be explored. A team of experts in public health, disease control and cyber security was assembled. They considered the possibilities and constraints of various systems, and their deliberations were later published in Chen et al. (2020). The team examined GPS data from shuttle buses used to transport the passengers, credit card transaction records and CCTV footage. However, the transaction data represented an indeterminate fraction of the passengers who had used credit cards on that day, and the tour bus location data were partial because they did not document passengers’ itineraries once they had left the vehicles.
Faced with these constraints, the expert team opted for passive mobile geolocation data as their primary data source (Chen et al., 2020). Compared with mobile phone GPS data, mobile geolocation data are ‘less accurate’ in a technical sense because they are generated by cell towers that pick up radio frequencies from mobile phones. However, this inferior accuracy was used to its advantage in an evaluative report published by the Taiwanese government (Jyan et al., 2020), in which it was argued that the reduced granularity of such data meant that their use was less intrusive than GPS data. By prioritising privacy over granularity (i.e. avoiding the use of triangulation or GPS data), the government steered clear of techniques intended for policing and crime prevention. However, the difficulty of determining the exact locations of contacts also made it necessary to place more people under ‘independent health management’ at home to observe whether they would develop symptoms of COVID-19.
Real-time monitoring has taken on great significance during the COVID-19 pandemic. As with other key aspects of digital disease control, it is clearly affected by the materialities of data. The way in which Taiwan set up a digital system to monitor resource distribution reveals one aspect of such materialities and demonstrates how, echoing Dourish (2017), the ongoing development of social and material infrastructures shapes possible ways of representing and accessing information, which in turn constrains and enables specific ways in which digital disease control becomes operable.
Consider, for example, the process used by Taiwan to produce data on the supply of masks and to engineer digital maps and other tools to create better public access to key resources. Since the SARS outbreak in 2003, ensuring sufficient supplies of such equipment has been a key aspect of Taiwan's pandemic preparedness. However, when the number of confirmed COVID-19 cases started to increase, panic buying and the stockpiling of surgical and N95 masks indicated that there may be insufficient information about national and regional supply levels. In response, several precautionary measures were taken: mask rationing was introduced in February 2020; manufacturers were enlisted to increase production; masks were sold only at NHI-participating clinics and pharmacies; and purchasers had to produce valid IDs that were checked against NHI records before any purchase could proceed.
It was clear that a process to confirm purchase records was required. Initially, a centralised database was set up to allow clinics, pharmacies and the public to obtain information about the number of masks held at each sale point and the individual buyer's remaining quota. However, the system quickly became overloaded, becoming irresponsive because of high volumes of traffic, which led to the data infrastructure being reconfigured in several ways. First, the government asked its contracted telecom company to expand bandwidth for the database. Then, open data on mask inventories and an API were set up, and the frequency of data updates was increased, going from hours to minutes. The engineering of the mask inventory API enabled other, improvised avenues for digital monitoring. Civic hackers and tech companies developed online maps, visualisations and other web and mobile applications to show mask supplies in different areas. The most notable of these was a map that was launched and run by the long-standing civic hacking organisation g0v, and which showed mask availability at each pharmacy or clinic. Chatbots, dedicated mobile applications, dashboards and voice assistants were also developed by individual developers and companies. 3
As is clear from the initiatives described above, the programming of digital disease control was not being implemented by the government alone; g0v and others attempted to construct alternative data collection systems or platforms to supplement government strategies. Many of these appropriated existing digital technologies and information-sharing practices, while some created new ones. For example, schools improvised data collection methods to meet legal and regulatory requirements for the tracing of travel histories, contacts and symptom development. The start of the pandemic in spring 2020 coincided with the school winter break in Taiwan, and students were expected to return to the classroom shortly afterwards. But, in order to comply with disease control protocols, and in response to parental concerns over their children's health, schools had to devise plans to ascertain whether students or their parents or close relatives (especially those sharing the same household) had visited places listed as high risk by the government. It was school administrators and teachers, rather than public health workers, who conducted the survey and deployed digital tools that were already in common use. Instead of the formal border declaration forms mentioned above, the resource of choice was group chats on instant messaging apps such as LINE (similar to WhatsApp) and closed groups on Facebook. Such channels were already widely used for school updates, announcements and information-gathering by parents and teachers. During the COVID-19 pandemic, these information exchange practices were quickly turned into a data generation mechanism to enable school administrators to compile the data requested by public health authorities. This is an apt demonstration of how specific materialities might shape data collection.
The discussion thus far has analysed the specificities of devising systems, procedures and platforms to contain the pandemic. Whether deployed by the government or improvised by members of the public, each of these measures has been shaped by past epidemics, routine information and data sharing practices, existing public health information infrastructures and commercial digital platforms. In other words, historical and material specificities played a crucial part in the fast-paced adoption and repurposing of digital technologies to meet various epidemiological requirements and address public concerns as they emerged and changed. Importantly, by making sense of these measures with a materialist approach, the above discussion shows how public health demands, digital technologies and social and material infrastructures jointly shape the ontogenetic process that configures digital disease control.
Digital disease control in action
Clearly, then, to understand how digital disease control becomes operable, it is also crucial to consider how the above programmes ‘emerge out of and participate in dynamic and ever-shifting regimes of practice’ (Dourish, 2017, 41). Through these regimes of practice, one can observe how contested or mutually constitutive relations between emergent public health demands, societal concerns, technological configurations and social relations have played out. Following the processual approach advocated by Pink et al. (2016), this section considers how discursive, imaginary, embodied, social or spatial practices are engaged in the continuous configuration of digital disease control. In particular, the analysis covers how digital systems for containing the virus have been put into action, what lines of action have been taken, and the ways in which digital disease control has subsequently been reconfigured.
Digital fences provide another useful illustration of how digital disease control is shaped by material contingencies and practices. Despite being deployed throughout Taiwan, the configuration and operation of digital fences depend on where monitoring takes place, and their effectiveness relies on the work of regional police forces, village administrators and local public health personnel. Introduced to support various at-home quarantine measures, digital fences enable authorities to ensure that people under quarantine do indeed stay at home by tracking signals from their mobile phones (which must be carried throughout their quarantine). 4 Cell tower data are monitored and used to alert local authorities (health, civil and police) of any breach. To complement digital fences, two chatbots serving very different purposes were built, one by TCDC and the other by HTC (a Taiwanese electronics company). The TCDC chatbot, which was up and running prior to COVID-19, focuses on health and risk, providing pandemic updates and answering questions from users. The HTC chatbot came into operation in April 2020 and facilitates case monitoring. It is one option provided to passengers arriving at Taiwanese ports to register their personal information, any change of location and updates on their health condition (including any development of relevant symptoms such as diarrhoea or loss of smell and taste). 5
For digital fences to work, data feeds are crucial. Cell tower rather than GPS data are used, which involves the same trade-off (privacy over granularity) as in the Diamond Princess case mentioned earlier. This technical architecture has led to more false alarms and introduced considerable uncertainty into quarantine breach alerts. There are several potential causes of a false alarm: mobile phones can run out of power and disappear from the monitoring system; reception can be patchy in some urban areas; and mobile connectivity in rural areas can be very poor. 6 Faced with these contingencies, the government's ability to dispatch on-the-ground personnel to confirm the physical location of people under quarantine has been critical to the effectiveness of digital fences.
However, location confirmation is only part of what these in-person visits achieve. The quarantine experiences of two German journalists in Taiwan during the pandemic are telling. 7 During their quarantine, the journalists were visited several times by village administrators when they forgot to recharge their phones or failed to answer telephone calls by public health staff. During these visits, the village administrators also informed the journalists of public health regulations, updated them on the pandemic situation, reassured them, answered questions and brought them daily necessities. The effect of these visits, then, was considerably broader than keeping a continuous stream of data for digital fences. Rather, they were constitutive parts of the digital materiality that fine-tunes digital disease control and maintains its effectiveness by human labour. Through their visits, village administrators attended to material, affective and social demands that were largely ignored in the technical architecture of the digital fences, but that are critical to address if digital disease control is to remain operative.
While digital fences were made workable by in-person visits, in the case of mask inventory maps, digital systems led to various unexpected lines of action that reconfigured how these systems facilitate the distribution of resources to the public. One important motivation for engineering the mask inventory API, mask maps and other applications were to render supply levels transparent and thus reduce waiting time in queues outside pharmacies. However, queues persisted because, after the roll-out of the maps, other practices emerged that could not be integrated into the API. For example, pharmacies implemented their own sale and distribution procedures, which required customers to take a ticket and return later during the day. As an unintentional side effect, this process resulted in two queues: first for tickets and then again for purchases. Ticketing was necessary for pharmacies, as they received masks in bulk from the government every morning, which their staff then had to repackage into rations for each buyer. In addition, the ticketing practice meant that elderly customers could pick up a ticket and rest somewhere else while they waited for their masks to be repackaged. Later on during the pandemic, another system was introduced whereby masks could be secured online and paid for at convenience stores, but accessing the technology to do this could be difficult for some elderly customers, for whom in-shop transactions remained the primary way to acquire masks. The importance of pharmacy access even increased for such customers, because they were also able to pick up masks for members of their family who could not do so due to work, school or other commitments. Meanwhile, because of their repeat and somewhat regular visits, these elderly customers met other people similar in age and circumstances, which turned the purchase of masks into a social occasion, rather than a mere logistical exercise. 8
In addition to the variety of adaptive measures described above, some citizens have worked voluntarily to reconfigure or recalibrate government systems or analytics for the better protection of collective health. For example, alternative data sources and analytics have been developed in response to concerns about mass gatherings and cluster infections. During the pandemic, there have been instances of tourist destinations being crowded at alarming levels, in other instances, asymptomatic individuals, unaware of being infected, have undertaken domestic travel. The government subsequently released the details of when and where people might have come into contact with such individuals, but this information was insufficient for a group of volunteer developers, who went on to build an online map so that people could compare travel histories. These developers recognised that comparing journeys can be ‘a tedious process, if there are a lot of cases, span[ning…] multiple days and multiple locations’, and that faced with this laborious process, people ‘might give up and/or determine their likelihood with some (random) heuristics’. 9 To solve this problem, the developers converted verbal descriptions of times and places released by TCDC into a data format (KML) that can be imported to Google Maps to detail itineraries. This enabled people who used Google's Timeline service to cross-reference their own location histories with those of confirmed infections.
Crowd prediction systems were also developed in response to an increasing risk of cluster infections during national holidays. People were expected to return to their ancestral homes in April 2020 for Tomb Sweeping Day, which has effectively become an occasion for family outings. Although the government released mobile and web applications to alert the public about potentially crowded tourist areas, these predictions were not accurate. According to an assessment by a university research lab, the predictions were based on average numbers of cell tower connections within a given area at peak times. 10 If connection counts started to exceed 115% of the average, the area would be labelled as ‘crowded’. However, as the lab pointed out, cell tower data cannot reliably predict if a crowd is gathering in a particular place. As with the quarantine alerts, the government prediction system was prone to false alarms for a combination of reasons. For one thing, the coverage area of each cell tower varies, depending on the population and usage in its area. For another, owing to economic considerations, a cell tower in a less populated place would cover a wider geographic area. Moreover, the government predictions did not consider the capacity of individual tourist destinations. As a result, a rural tourist destination, despite having a greater capacity and sitting in a larger area than a urban one, was more likely to be labelled as crowded because even a small increase in visitors over an extended area without crowding it might still lead to enough connection counts to trigger an alarm. Seeking to improve predictive accuracy, the lab built an alternative system that uses location data collected by Google from its users, because Google Maps data are more accurate and richer for lab analysis, and more reliably show popular visit times and durations. Subsequently, the team changed data sources and tweaked parameters relating to capacity for improving results.
In other situations, the turn towards digital disease control can invoke contradictory imaginaries and practices that are not necessarily well aligned. This became most apparent in the effect of the pandemic on delivery platforms, which experienced greater demand following the implementation of quarantine measures and social distancing rules. Here, the logic of using data to manage social contact, including whether to trace it or avoid it, was translated by platform operators into tweaks in technical design that added a ‘no contact delivery’ option to the ordering process. The delivery platforms then sought to enhance the ‘no contact’ procedure by requesting access to the quarantine database, arguing that this would better protect their workers. Due to obvious privacy concerns, this request was refused by the TCDC. 11 However, while there was no actual change in the use of quarantine data, it is worth noting how a new imaginary can be invoked to alter the technical design of delivery platforms and can act as a discursive device to justify the lack of personal protection equipment that should be provided to workers by delivery platforms. 12
It is not surprising that digital disease control has inadvertently encouraged the profiling of people and places that might be contagious. Profiling is by no means new or unique to disease control, but it has become inevitable following the adoption of data and information processing for epidemic prevention purposes. One lesson learned from the previous SARS outbreak was the importance of providing reliable information to the public during an ongoing epidemic. Thus, the TCDC has held daily press conferences to provide updated information on confirmed cases and report other related issues. The TCDC acts in accordance with the Personal Data Protection Act and the Communicable Disease Control Act, which means restricting the release of personal information to only what is necessary to contain the spread of the virus. Generic updates at press conferences and on the TCDC website contain only the following types of information: on domestic cases: case numbers, gender, channel of infection, occupation (where relevant), date of confirmation of infection, times and locations of contacts (if any) and how contacts have been treated
on imported cases: case numbers, gender, residence (region), travel histories and symptom course, date of confirmation of infection, whether contacts were made and how the contacts have been treated.
The release of the above information, however, has been sufficient for people to search for the identities of infected individuals. There are abundant instances of these people being identified and their personal details (such as where they live, work, or meet with friends and families) being ‘skillfully’ conveyed in a way that would not violate data protection regulations. 13 Furthermore, additional ‘reminders’ were circulated in closed social media groups and instant messaging channels to alert members about areas and neighbourhoods to avoid because people there had been infected or were undergoing quarantine or isolation. Effectively, these places became ‘contagious’ as a result of everyday information practices that were said to protect the community.
Discussion: Consequences of digital protection of health and society
The above analysis has taken a materialist approach that looks beyond digital and real-time technologies to the ontogenetic process in which histories, infrastructures, spaces and practices continuously configure digital disease control. This approach prompts a further rethinking of the consequences of digital disease control that will better capture how societal values, promises, changes and costs are realised in specific arrangements and practices. Several aspects that require careful attention are highlighted in this section.
First, despite the rhetoric of a radical, ‘big data’ turn, digital disease control comprises a multitude of ‘cumulative transformations’ rather than one disruptive shift. Mackenzie (2017) suggests that the development of data-intensive operations entails an accumulation of knowledges, techniques, experiences and practices in particular ways. Similarly, the mobilisation of digital data and technologies to meet epidemiological demands is a transformation more cumulative than radical. While new databases, digital technologies and data flows are being produced, they remain connected to existing infrastructures, protocols and everyday practices. In the early stages of the COVID-19 pandemic, the established automatic case monitoring system provided an infrastructure that could be used to identify people who might have contracted the virus. Digital fences introduced a way to enforce quarantine, but they were guided by existing epidemiological protocols, facilitated by frontline workers and enabled by a reconfiguration of how contact tracing and monitoring were conducted. The engineering of the mask inventory API amid the pandemic produced new, near real-time data flows for distributing critical epidemiological resources, but the data flows relied on the NHI database and had to be materialised in the daily routines of pharmacists. By attending to the accumulation of material arrangements and practices, the above accounts not only sketch a landscape of digital disease control but, more significantly, provide an ontogenetic understanding of how such control comes about. An examination of the materialities of data generation and processing reveals how important it is to look beyond the hype around digital health (Milne and Costa, 2020) and attend to the processes whereby technologies invite new promises, meet epidemiological challenges, disrupt work and life practices and lead to continuous reconfigurations of pandemic response approaches.
Second, and following directly from the above analysis, closer attention to the uneven accumulation of digital and data infrastructure reveals the specific ‘moral repertoires,’ or conceptualisations of the common good and values (Sharon, 2018), that are prioritised or deepened by digital disease control efforts. For example, Taiwan's government has presented the tracing of Diamond Princess passengers as the success story of a cost-effective epidemiological investigation. Conceived in this way, digital disease control prioritises the ‘project’ and ‘industrial’ moral repertoires that emphasise the values of innovation and efficiency, respectively. However, this prioritisation overshadows the many other kinds of practical work that are necessary to maintain digital operations in the face of uneven infrastructure for data generation and circulation. As innovative systems such as digital fences were deployed, it became clear that the efficiency of these systems was affected by the attempt to balance the granularity of data sources with privacy protection, mundane everyday practices (e.g. recharging phone batteries), quarantine locations and the splintering of network infrastructure. Furthermore, epidemiological tracing is complex and more than just informational, as the work of frontline staff makes clear. The unevenness in digital and data infrastructure and the various practices required to carry out digital disease control thus challenges the moral repertoire of efficiency subscribed to and promoted by the government. The focus on efficiency alone risks neglecting the value of different types of epidemiological work while perpetuating a version of technological solutionism that underplays the complexity of translating and incorporating existing epidemiological procedures into digital ones.
At the same time, although concerns regarding the ‘infrastructuralization of platforms’ (Plantin et al., 2018) have been raised, it is worth highlighting that the technical architecture of digital operations such as civic responses to pandemics is inevitably characterised by an intricate dynamic between competing moral repertoires. On the one hand, corporate platforms such as Google provide rich technical resources for code writing and data gathering, and often become a popular alternative to or substitute for government data sources or systems. As discussed above, the use of KML format and Google Maps certainly facilitated the cross-referencing of travel routes by allowing Google Timeline users a more convenient way to check if they had come into contact with infected individuals. Similarly, the university lab's alternative crowd prediction system used Google location data because they were richer and more fine-grained than those used by the government's prediction system. On the other hand, the favouring of private platforms can undermine efforts to provide more accessible and reliable pandemic information to as many people as possible, because access to information could become platform-dependent. Furthermore, data on various kinds of ‘non-users’ (Satchell and Dourish, 2009), such as those who resist or lag behind in technology adoption, or socially disenfranchised groups, would be absent and their experiences unaccounted for in such systems. This disjuncture would be a significant detriment to any attempt at digital disease control. Accordingly, taking these technical arrangements and practices into account, it is evident that the effort to construct more agile and collective crisis responses will involve contradictory but mutually constitutive moral repertoires. Clearly, the priorities of such moral repertoires – including collective well-being, efficiency, innovation or the accumulation of data and capital – may be at odds with one another, but together they constitute the implementation of civic digital operations. A comprehensive examination of technical arrangements and practices in both civic and government action requires careful attention to the intricate dynamics in which competing moral repertoires and societal values become entrenched and entangled.
Third, a closer look at the repurposing of everyday information practices not only demonstrates how ‘pandemic sorting’ unfolds but also highlights the contentious relationship between the publication of pandemic data and the surveillance of individuals. The use of data analytics to modulate and sort people has undoubtedly been growing in security, but it has also been seen in medicine and healthcare (Hogle, 2016; Lyon, 2014; Schüll, 2016). The experiences of Taiwan has shown how publishing data for epidemiological purposes has enabled pandemic sorting where ‘social surveillance’ techniques are adopted to categorise people and places on the basis of epidemiological information obtained by ‘ongoing eavesdropping, investigation, gossip and inquiry that constiute information gathering by people about their peers’ (Marwick, 2012, 379).
In their daily COVID-19 press updates, the TCDC removes unnecessary information that contains specific identifiers and/or would compromise privacy. Nevertheless, some people take it upon themselves to work out the personal details of infected individuals, including specific places where contact might have been made and other people and places that might be associated with these individuals. The information then shared is sufficiently vague to avoid violating the relevant regulations but clear enough to indicate specific locations (neighbourhoods, shops, buses, etc.) or groups of people (companies, occupations, ethnicities, etc.) that others should avoid. As a result, pandemic sorting enables the recategorisation of people according to how they are related to infected individuals. It creates new fearful ‘aggregates’, including people and places, or COVID-19 ‘alterities’ (Pelizza, 2020), that are profiled and tracked, while renewing fears in such a way that invites people to shame, blame and discriminate against marginal social groups (Taylor, 2016).
Complicating the matter further, there has been insufficient articulation of what would be an adequate level of surveillance to meet both epidemiological and civil liberty demands. While there are legal provisions to undertake some form of surveillance for epidemiological control, government documents do not make clear what level of personal data collection is to be allowed in exceptional situations such as the COVID-19 pandemic, or indeed what the temporal, spatial, social, or legal limits of such a state of exception might be (see an evaluation from legal perspectives in Lee, 2020). Instead, the joint effort between the Taiwanese government and corporations to develop platforms and data analytics for epidemiological control only adds to concerns about the expansion of ‘iLeviathan’ (Prainsack, 2019), where control is lost over how people's lives and bodies are translated into data, how such data are used and to whose benefit or profit the current turn towards digital disease control should accrue.
Finally, various types of non-technical action and expertise that have contributed to making digital disease control operable are not as highly valued as technical practices and skills. Technical skills, particularly coding and design, have become revered by the public for their contribution to digital measures such as mask supply maps, footprint maps, or the mask inventory API. In such an atmosphere, the value of practices that are not readily digitised can go unnoticed and unappreciated. The social and familial value of queuing up in person is one such example. More critically, while the benefits of digital methods are if anything overemphasised, the demand by some citizens and groups that personal data be collected and processed only in ways that are proportionate and appropriate to their purposes has not met with a prompt government response, despite being crucial in avoiding mission creep and surveillance capitalism (Zuboff, 2019). Civil liberties groups in Taiwan have long demanded that the government provide more details about the scope and limit of personal data collection and processing, as well as the extent to which it is justifiable according to existing legal frameworks or ethical principles. But the government clarification of the technical architecture of digital fencing and the data retention protocol, including types of data collected and their retention period, came almost a year after the initial outbreak.
The relative lack of acknowledgement of practices that are not readily digital makes it clear that they, like caring practices, can become ‘marginalized, invisibilized and neglected elements, experiences, and relations’ (Murphy, 2015, 721) simply for not prioritising normalised ways of digital living, such as accessing digital maps or mobile apps. As such, questions about government data collection and processing regimes, while critical to the protection of civil liberties in digital society, are being ignored. This indifference raises concerns about how reasonable doubts or ‘affirmative critiques’ (Braidotti, 2009) can be cast and how worthwhile failures can be accounted for.
Conclusion
This paper adopts a materialities approach to its analysis of the emergence and consequences of digital disease control, using a case study of Taiwan's response to the COVID-19 pandemic. It foregrounds the importance of accounting for locally and historically contingent accumulations of digital infrastructure that enhance pandemic response. Data sources, types of analysis, digital platforms and places of deployment have also been analysed to understand how they give particular form and function to specific disease control measures. In addition, actions, protocols and regulations that have emerged in response to digital measures, most notably the in-person and affective work of frontline workers, have been included in the discussion so as to highlight how these measures are challenged and reconfigured as they become operable. This examination of the materialities of digital disease control also reveals particular issues that can affect the future operations of digital measures, their effects on public health and wider societal consequences. These issues centre around the expansion of the logics, discourses, practices and technical arrangements of data, platforms and analytics in ways that justify an elevation of certain values, actions and opinions at the expense of others. Moral repertoires of innovation and efficiency take priority, pandemic sorting occurs, normalised digital living is privileged and reasonable doubts or affirmative critiques are silenced.
As these measures continue to take shape and their various actions and consequences reconfigure digital disease control, different practices of control, power and care may be invoked when future pandemics strike. As is shown in the analysis above, the particular sciences, bodies, people and technologies involved in protecting public health and caring for others can lead to a number of societal concerns. But there is scope to better position ourselves to handle the next pandemic if at least some of those concerns are addressed in advance. Mol's discussion on ‘how to live with/in/as a body that is both fragile and able’ (2008, 40) can be helpful in the search for a better pandemic future that benefits from digital disease control while addressing its looming dangers. What concerns Mol is the practice of leading a good life while affected by disease and how good care can be devised by paying closer attention to the entangled relations between practices, patients, bodies, professionals, states and medical and everyday knowledge.
As we turn to Mol to consider how a better pandemic future might come about, we need to foreground vulnerability, unpredictability, unevenness, partiality and the ways in which they introduce uneasy practices and arrangements into digital disease control. Orientating towards a better future in this way would require us to explore continuous processes of reconfiguring the relations and practices that involve individuals, professionals, governments and the ensuing materialities of digital disease control that can protect but also control, surveil, subdue, or shame healthy and vulnerable individuals alike. Thus conceived, the protection of health and society can attend to how programmes of digital disease control take form and how these programmes unfold in locally contingent ways. Attentiveness to these local contingencies invokes the possibility of translating good practices and arrangements to other places and situations of care while making room for reasonable doubts and accountable failures in the trialling of uneasy practices.
This way of looking for possibilities to govern future pandemics is ‘not just a matter of regulating the affairs of the state, [but] also impresses a specific shape on the relations between people’ (Mol, 2008, 29) who have different vulnerabilities, expertise, professional knowledge and practices for protecting lives and preventing virus spread. Keeping in mind that caring and doctoring practices are fluid and dynamic, people and governments can seek to explore ways to devise digital epidemiological measures, responses and arrangements for the protection of life that might be material, affective, local, contingent and continuously evolving, as well as translating these arrangements well for people and places with differentiated capabilities and vulnerabilities.
Footnotes
Acknowledgements
The author would like to thank the anonymous reviewers for their constructive comments and the colleagues at the Research Center for Epidemic Prevention of the National Yang Ming Chiao Tung University (MOST109-2327-B-010-005) for their insights and inspiration. The research for this paper is also supported by the Taiwanese Ministry of Science and Technology project AI Cities in East Asia (MOST 109-2410-H-010-001-MY3).
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Ministry of Science and Technology, Taiwan, (grant numbers MOST 109-2410-H-010-001-MY3 and MOST109-2327-B-010-005).
