Abstract
This article draws upon a multi-sited ethnography of everyday labour in Lebanon's digital cash assistance for Syrian refugees. The datafication of humanitarian infrastructures generates technological breakdown, gaps in data and incredibly strict and cumbersome rules. In response to impediments related to biometric identification and automated poverty targeting, this article argues that humanitarian staff, refugee recipients and community members engage in ‘repair work’ – the subtle and quotidian labour that goes into addressing fragility and maintaining functionality. Inspired by feminist studies of labour, repair work is found to be invisible in being undervalued, unpaid and reproductive, which is reminiscent of labour that has historically fallen to disenfranchised people. Repair work also enables data workers to assert their autonomy and contest infrastructures that they framed as being unreasonable and unjust. In doing so, findings suggest that repair work is fundamental to the ability of data-driven aid programmes to cater to the needs of populations in crisis. This paper marks two contributions to understanding the promise and perils of ‘Technology for Good’: it introduces repair work as a novel conceptual framework to analyse labour involved in the datafication of aid, and it applies new empirical evidence to critical studies of data work.
This article is a part of special theme on Commodifying Compassion in the Digital Age. To see a full list of all articles in this special theme, please click here: https://journals.sagepub.com/page/bds/collections/commodifying_compassion
Introduction
In 2019, I conducted fieldwork at an international humanitarian technology conference to better comprehend how experts understand data and technologies that are used for emergency assistance to refugees. I was struck by the prominence of engineers and technical specialists and how their expertise was framed as essential for improving humanitarian interventions. The most memorable and seemingly well-received presentation was a keynote presentation by a self-described ‘Business Guru and Expert Innovator’. Repeatedly citing examples that sounded like they came from sci-fi novels, his lecture made a case for the many ways that artificial intelligence (AI) would make humanitarian work faster, easier and cost-effective, thereby better supporting the Sustainable Development Goals. He referred to drones that use AI to report on local ecosystems, AI to train seeing eye dogs that help blind people navigate bustling cities, and AI on a tablet to help villagers in remote areas complete basic medical diagnoses without a doctor present.
An audience member was visibly inspired by the presentation and shared this optimism regarding the future of humanitarian work, stating that, ‘If we can put people on the moon, we can solve global poverty. It's a matter of finding the right people to do the right stuff’. Others at the conference presented humanitarian labour as requiring innovation, data-driven processes and investment, which also meant attracting more technical experts. For instance, data scientists suggested that, with access to greater volumes of data, not only could they reach more people in need. They could also reduce the error rate of humanitarian technologies – technologies that people living in emergencies supposedly desire to use.
Throughout the aid sector, optimistic beliefs continue to circulate about the ‘transformative potential’ (Read et al., 2016: 1314) of ‘digital humanitarianism’, defined as the adoption of data processing, smart technologies and remote management practices to bolster aid disbursement (Johns, 2023). For instance, generative AI like ChatGPT has been framed by one of the sector's leading news outlets, The New Humanitarian, as helping to ensure that local groups are not overworked and understaffed by more efficiently generating media content and funding pledges (Roane, 2023). Prominent narratives of the datafication of aid thereby suggest the capacity for ‘innovating out of disasters’ (Mosurska et al., 2023: 13) and ‘more orchestrated workflows’ (Memon, 2022).
In contrast to popular accounts that frame leveraging data as necessary to address issues in aid labour and to fostering affordances like empowerment and financial inclusion (e.g., UNSGSA, 2019), critical technology scholars have increasingly studied how labour relations are asymmetrically transformed by datafication. For example, Femke Mulder and colleagues (2016) show how volunteer crowdsourcing challenges the notion that open data platforms support more inclusionary crisis response, as it is found to exclude crisis-afflicted populations from processing such data. Squire and Alozie (2023) argue that data-driven humanitarianism strengthens the paternalistic subordination of local groups. Ryan Burns (2015: 487) finds hierarchies of labour in Big Data, which ‘privileges’ professionalised volunteers that are foreign to areas being worked on. Critical accounts therefore suggest that datafication has made humanitarian work more inequitable, exploitative and supportive of regimes of ‘capitalisation’ (Tazzioli, 2022: 73).
While scholarship on the labour of digital humanitarianism is still in its infancy, it has a noteworthy limitation that I address in this article. Little is known about how those in the field cope with and strategically respond to operational breakdowns and obstacles that are reinforced by the datafication of aid. Using Lebanon as a case study, this article makes sense of how humanitarian ‘workers’ adapt to such challenges in practice. Lebanon evinces overlapping crises, which are occurring in a nation that delivers barely any public welfare services to its residents. This has impelled Lebanese and foreign actors to intervene with technological ‘solutions’ to fill the void left by the state (Human Rights Watch, 2022).
Drawing upon a multi-sited ethnography of everyday labour in Lebanon's digital humanitarian interventions, this article examines two data-driven infrastructures. These infrastructures are the cornerstones of the United Nation's cash-based assistance for impoverished Syrian refugees residing in the nation: digital identity (ID) validation and automated vulnerability targeting. It argues that aid workers, recipients and community members mitigate the effects of infrastructural breakdown by engaging in ‘repair work’ – the subtle and quotidian labour that goes into addressing fragility and maintaining functionality (Jackson, 2014). Critical feminist studies of labour are applied to capture how repair work is invisible in being undervalued, unpaid and unseen by the public, state, technology experts and humanitarian institutions, as well as reproductive of the daily and long-term continuance of data infrastructures that support impoverished populations’ livelihoods. This is reminiscent of labour that has historically fallen to particularly disenfranchised people. Repair work enables compliance with onerous rules, compensation for dysfunctional data infrastructures and filling gaps in data. It also enables data workers to assert their autonomy and contest infrastructures that they framed as being unreasonable and unjust. Thus, findings suggest that repair work is central to how humanitarian aid operates on the ground.
This article makes two unique contributions. First, it introduces a novel conceptual framework to extend existing understandings of the labour involved in the datafication of aid (e.g., Cheesman, 2022), bringing together discussions of ‘what technology does to humanitarian action’ (Sandvik et al., 2014: 222) with feminist critiques of labour. Second, as an analysis informed by participant observations, interviews and document analysis, this article brings new empirical evidence to bear on critical studies of data work. It does this by uncovering how repair work materialises ‘in situ’ in everyday practice (Sormani et al., 2019: 2). In doing so, this article responds to calls to shift away from focusing too narrowly on ‘official politics’ (Hilhorst, 2013: 1). Taken together, these novel insights make a larger contribution to understanding the promise and perils of ‘Tech for Good’, which refers to technologies that are employed to alleviate human suffering and tackle social problems (Powell et al., 2022).
The article is organised as follows. The next section focuses on the conceptual framework by briefly explaining repair work, how it has been used in critical data studies and its capacity to illuminate resistance practices. Critical feminist analyses are also described as being central to this article's understanding of repair work's power relations. Following, research methods are outlined by providing detail on Lebanon as a research site, the three main forms of data collection and the approach to abductive analysis. Afterwards, the article presents the empirically informed findings on repair work involved in digital ID and in automated poverty targeting. The conclusion summarises the findings and highlights how the offloading and extraction of repair work complicates imaginaries of digital humanitarian labour.
Repair work and broken-world thinking: fragility, invisibilisation, resistance
Scholars from varying disciplines have laid the foundations for studying how sociotechnical systems are continuously damaged, stabilised and modified (e.g., Graham and Thrift, 2007). Some aim to bring the unintelligible components of breakdown into sight (e.g., de Laet and Mol, 2000). In his now influential essay, Jackson (2014) coins ‘broken-world thinking’ to trouble linear notions of innovation and use breakdown, rather than innovation, as the foci of analysis. This pivot helps capture the omnipresence of iterating between rupture and repair and the ways in which rupture is generative. ‘Repair work’ refers to a broad set practices that react to breakdown by mending fissures, responding to issues and filling holes to keep systems functioning (Jackson, 2014). This includes a large ‘range of activities’, including those that are technical, social and material, as well as those on scales ranging from ‘local fixes to broad, systemic efforts’ (Henke and Sims, 2020: 3). ‘Maintenance’, a subset of repair work, occurs before failure and is more preventative and ongoing with a future-oriented focus (Carr, 2017). As Russell and Vinsel (2016) argue, studying breakdown, repair and maintenance is critical because they have ‘more impact on people's daily lives than the vast majority of technological innovations’.
This resonates with critical studies of data that interrogate the hidden dynamics of everyday data practices (Weiner et al., 2020). For instance, Sarah Pink and colleagues (2018) examine the mundane work required to create, fix and grow data given its inaccuracies, gaps and physical fractures. Such repair work not only makes data usable, extends the lifespan of data-driven devices and systems and helps users regain control. It also supports the contention that data-driven solutions are not seamless; its experiences and uses are ‘contingent’ and require ‘improvised labour’ (Tanweer et al., 2016: 13). Nevertheless, this is not to say that perceptions of breakdown are uniform. Repair work is subjective; something that is broken for one person may not be broken for another and vice versa.
While literature on the repair of technology foregrounds the work of non-elite workers (e.g., building mechanics, Henke, 1999) and marginalised groups (Burrell, 2012), this focus is extended further by approaching workers as not just those who are skilled, experts, or formally employed by institutions (e.g., protection coordinators, technical specialists). This paper conceives of repair ‘workers’ to also include refugees and community members whose jobs are not necessarily concerned with contributing to aid delivery but nonetheless provide labour that is required to keep humanitarian systems afloat. Rather than repairing data infrastructures as a requirement of formal and paid occupations, their repair work is compelled by the need to access aid, facilitate livelihoods and advocate for fairness in humanitarian systems. This approach responds to feminists’ calls for extending the meaning of labour and what counts as work (Mackenzie and Rose, 1983; Stokes and Lawhorn, 2022).
There is an important argument about who is doing the labour of repair, the conditions under which repair labour materialises and who benefits from such labour – questions that are emphasised by a large volume of feminist scholarship on work and on data (e.g., Cooky et al., 2018; D’Ignazio and Klein, 2020). Such literature suggests that marginalised groups have historically been relegated to lower-status roles (e.g., women's administrative work, migrant farm workers) and that their care and social reproductive labour are traditionally uncompensated and completed in settings that are out of sight to most people (Alda-Vidal et al., 2023; Ferguson, 2019; Mitchell et al., 2003). This aligns with how scholars characterise repair work as being distinctly mundane, quotidian and behind the scenes (Denis and Pontille, 2023). Repair work is therefore illustrative of a core question posed by feminist and third-world liberation movements, as highlighted by Star (1995: 3): ‘Cui bono? Who is doing the dishes? Where is the garbage going? What is the material basis for practice? Who owns the means of knowledge production?’
In asking these questions, this article uncovers repair work dynamics and actors that are subject to ‘invisibilisation’. To be rendered invisible refers to work and workers that are quite literally unseen and out of sight, but also that which is ignored, unprotected, taken for granted, socially and economically marginalised and absent from popular imaginaries (Crain et al., 2016). Invisible labour takes shape because it is conducted by or is characteristic of the work of socially disenfranchised groups who are presumed to have lesser credibility and expertise (Federici, 1975); is outside of normative definitions of productivity (Carr, 2017); and, is progressively devalued as it gets offloaded and passed onto various workers and institutions (Parreñas, 2012). In Lebanon, while repair work may be perceptible to labourers and their immediate communities, such labour is noticeably under-recognised by the public, state, technology experts and humanitarian institutions. By ‘rendering the collection and assemblage of data as “invisible work” rather than just “doings”’ (Canzutti and Aradau, 2024: 1), this framing enables understanding how digital humanitarian relations are formatively shaped by regimes of valuation and the mobilisation of unofficial resources and responsibilities.
Repair work highlights the agency of actors involved in reproducing humanitarian infrastructures. Residents and humanitarian staff in Lebanon actively struggle with, negotiate and workaround unjust dynamics through their repair work, asserting forms of autonomy as part of their everyday engagements with humanitarian systems. These findings connect to how scholars have described resistance as a form of repair work through its commitment to uphold a feminist ethic of care and to foment reparative practices (e.g., capital switching to offset capitalist structures, Webber et al., 2022). In the same vein, this paper highlights how resistance efforts, however small, constitute repair work by using dissent and disagreement to care for refugees in the hopes of mitigating breakdown and harm. As such, the analysis contributes to emergent literature on the relationship between repair, data and resistance (Plantin, 2021). By doing so, it highlights social justice efforts at the grassroots level. The paper's site and research methods are now explained in more depth.
Methods and site
The World Bank Group (2021: xi) has framed Lebanon as experiencing one of the ‘most severe crises episodes globally since the mid-nineteenth century’. Residents are pressed to navigate an unprecedented economic depression, government corruption and collapse, long dilapidated infrastructures (energy, water, food, health), one of the largest non-nuclear blasts ever and socio-economic pressures from hosting the largest number of refugees per capita in the world (Médecins Sans Frontières, 2021). Roughly 80% of Lebanese residents live below the poverty line, with 36% living in extreme poverty (European Commission Directorate-General for Neighbourhood and Enlargement Negotiations, 2023). Given the pervasiveness of precarity, polycrisis and political inactivity (Fakhoury, 2017), the nation is saturated in humanitarian responses – including one of the largest cash-based assistance programmes ever recorded (Chehade et al., 2020). Employing a multi-sited ethnography from September 2019 to May 2023, this research focused on two cases of humanitarian aid in the country: the Beirut Port explosion of August 2020 and the United Nations’ cash assistance programme, which delivers cash on a monthly basis to poor Syrian refugees. The latter programme is titled the Lebanon One Unified Inter-Organisational System for E-cards (LOUISE).
This research was approved by the Australian National University's Human Research Ethics Committee (protocol number 2020/496). Multiple sources of data were collected. The author completed fieldwork in Lebanon in 2019. This included participant observations of Syrian refugees and humanitarian staff while volunteering for a Lebanese non-governmental organisation, as well as observations of public settings around the nation like banking sites and local markets. The author also engaged in in-person and remote observations of humanitarian conferences, press conferences and workshops, which were primarily located in Lebanon, North America, Europe and Australia. These events were selected because they involved key actors from Lebanon's humanitarian system and attendees spent time developing their reflections to share with other practitioners. Observations enabled understanding what humanitarians were discussing, the challenges they encountered and public-facing narratives as they relate to technology.
In addition, the research project faced disruption due to the COVID-19 pandemic. As such, the author adapted by conducting forty videoconferencing interviews with participants who were directly involved in designing, delivering, researching, regulating and/or reporting on the Beirut Port explosion and/or LOUISE. Participants included country directors, operations directors, policymakers, government officials, regulators, United Nations employees, World Bank representatives, national and international non-governmental organisations, community groups, technology experts, data scientists, academics and journalists. Interviewing took place from April to October 2021, ranging from roughly one to two and a half hours long. Twenty-six interview participants resided and worked in Lebanon and the rest spanned Switzerland (n = 4), the United Kingdom (n = 2), Germany (n = 1), Italy (n = 1), Belgium (n = 1), the United Arab Emirates (n = 1), Jordan (n = 1), Kenya (n = 1), Japan (n = 1) and Australia (n = 1). Participants’ identities are anonymised by using pseudonyms. Two participants consented to note-taking only. Thirty-eight participants consented to having their interviews audio recorded, which I then transcribed myself to gain a deep familiarity with the data and to assist the generation of codes and themes (discussed below).
Participant observations and interviews were triangulated with document analysis. Four main categories of documents were collected and analysed: internally produced organisational documents (reports, program evaluations, impact assessments, country briefs, inter-agency situation updates); speech-based sources (press releases, meeting minutes, public interviews); media sources (news articles, blogs, social media); and, web sources (listservs, websites, collaborative Google Drive documents, online public databases).
After initial coding where I read each transcript, line by line, to verify the veracity of my preliminary codes, I created the first iteration of my codebook, including the inclusion and exclusion criteria for each code. Pattern coding enabled organising and grouping my codes into a more concentrated number of categories, concepts and themes. This assisted sorting and grouping my codes into themes, which included naming, defining and refining those themes.
Data was analysed abductively. Unlike ‘induction’ which assumes the researcher removes preconceptions so that theories emerge from the data and ‘deduction’ which seeks to prove or falsify a hypothesis, ‘abduction’ interprets data by actively comparing it against existing scholarship to avoid ‘the straightjacket of pre-existing concepts’ (Timmermans and Tavory, 2012: 169). To accomplish this, I first worked to comprehensively understand existing theory and literature on digital humanitarianism, critical humanitarianism and humanitarian labour. Once I achieved a baseline understanding of these literatures, I iterated between data and literature to look for instances in my data that did not fit dominant theoretical frameworks and were under-explored in scholarship. By recursively investigating novel paths from my situated research, I was able to identify how existing understandings of digital humanitarianism could ‘be extended or refined’ (Snow et al., 2003: 194), thereby supporting ‘theoretical innovation’ (Timmermans and Tavory, 2012: 169).
In line with feminist traditions, I understood my access, research decisions, data analysis and interactions with my participants to be fundamentally shaped by my positionality. I am a Canadian-Arab woman whose parents were both born in Lebanon, with my father being Lebanese and my mother being born there as a Palestinian refugee. I thus maintained both insider and outsider status at different stages of the project. The following section delves into the unique findings that were generated through abductive analysis.
Repair work and resistance in Lebanon's humanitarian data infrastructures
This section discusses different forms of repair work that negotiate and compensate for breakdown related to the datafication of Lebanon's humanitarian infrastructures. This repair work is found to be critical in mitigating the harmful effects of dysfunctional and burdensome infrastructures that leave gaps in aid provision and in data. Repair work also allows workers to assert often qualified forms of autonomy, resisting structures they perceive as limiting and unfair. It should be noted that repair work is not specific to digitised components. Repair work can compensate for breakdowns in analogue forms of aid. For example, consider literature on ‘refugee economies’ and the work that goes into mitigating failures in emergency assistance, such as requests by humanitarian institutions for locals to mill maize (Betts et al., 2017).
Repair work is found to constitute invisible labour in that it is overlooked, undervalued and uncompensated. For example, in the case of refugees and residents, their work is not treated as labour, which renders their contribution to digitised aid as not only unrecognised, but also as a freely obtainable resource. In the case of humanitarians’ backstage labour, their work may be salaried, but it is out of sight and peripheral within mainstream descriptions of what it takes to implement and sustain digitised aid.
Repair work is also reproductive because it enables the endurance of infrastructures that reproduce populations suffering from crisis (Ferguson, 2019). As reproductive labour, humanitarian systems depend on this repair work. Without it, the United Nations’ datafied infrastructures ‘would not be possible and sustainable’ (Adams, 2022: 388). By repairing interrelated data infrastructures that facilitate access to cash assistance and advocating for aid processes that are more effective and fair, repair work thereby supports the survival of refugees and of these datafied infrastructures.
Repair work for digital ID: complying with rules and compensating for malfunction
Identity data such as population registers and biometric fingerprinting have long been central to intimate forms of technologised control, tracking and categorisation (Breckenridge, 2014). Prior to digitisation, identity data in Lebanon required humanitarian personnel to verify individuals’ identity documents. Given the excessive costs and delays of analogue processes, digital ID has been adopted to support efficiency, empowerment and fraud prevention in humanitarian aid (Weitzberg et al., 2021). Advocacy for digital ID has grown so much that it is now enshrined in global compacts like the Sustainable Development Goals, with technology being framed as an important means to achieve such targets (Cheesman, 2023). Following these trajectories, the United Nations has mandated that Syrian refugees in Lebanon register their digital ID at United Nations-run validation centres if they wish to access cash assistance (Beduschi, 2019). As part of ID registration and verification processes, the United Nations collects Syrian refugees’ iris scans, fingerprints and demographic information like education levels and chronic illnesses. Refugee data is stored on a transnational database called the Refugee Assistance Information System, which is managed by the United Nations Refugee Agency (Janmyr, 2018).
It became clear in my research that the process of registering ID data and meeting its requirements are particularly cumbersome for some people. In addition, digital ID systems sometimes malfunction, especially for those with intersecting axis of marginality like gender, race, class and disability. These findings are confirmed by numerous scholarly accounts of biometric ID in welfare settings (Paragi and Altamimi, 2022) and more broadly (Buolamwini and Gebru, 2018; Kloppenburg and van der Ploeg, 2020). Third-party reports and my interviews with humanitarian workers reveal many forms of behind the scenes labour that recipients, community members and humanitarian workers employ to work around the burdens enshrined by digital ID. Such workarounds are not officially acknowledged in press releases and institutional discourse about the efficacy and affordances of digital ID (e.g., UNHCR, 2018). Nevertheless, refugee livelihoods are contingent on this labour, which is invaluable to accessing lifesaving aid. In reflecting assumptions that the United Nations will always have access to aid labour that they are not compelled to pay for, this repair work constitutes an important form of uncompensated reproductive labour.
An important rule set by the United Nations Refugee Agency and World Food Programme is that refugees who receive cash assistance must validate their identity every three months. Marla, an interview participant located in Lebanon, conducts third party monitoring of a cash assistance agency. Marla described the challenges of ID validation for refugee recipients who must travel to United Nations-run validation centres to record their biometric information, keep their demographic information up to date and confirm that they are, indeed, who they claim to be. Quarterly validation is mandatory even though some refugees live far from validation centres, struggle to reach validation sites (e.g., the elderly and those with disabilities) and face difficulties getting the arrangements to participate (e.g., those with carer responsibilities).
In a report on refugee users’ experiences with digitised cash assistance in Lebanon, refugees recounted many forms of repair work involved in ID validation. Refugees recounted walking long distances and paying for lengthy and expensive taxi rides (Ground Truth Solutions, 2021). There are also gendered burdens of ID validation labour, as mothers spoke about bringing their children on their journey or relying on community members to supervise them while they travel. These strategies reflect how some people experience breakdown when accessing aid infrastructures and that they are required to engage in repair work which makes impractical digital ID systems workable.
The United Nations Refugee Agency and World Food Programme have attempted to mitigate ID validation challenges by allowing some refugees to verify their identities at post offices (LibanPost) and money transfer locations (Cash United), which are ideally closer in proximity (Ground Truth Solutions, 2021). However, United Nations agencies do not compensate refugees for their time and money spent contributing data to digital ID infrastructures, nor do they offer protection from risks they may encounter. Instead, some refugees are forced to take time off work so that they have enough time to reach validation centres. The expense of losing income and paying for travel like petrol and bus and taxi fares comes out of refugees’ own pockets. They are not remunerated for this labour.
In addition, refugees are at increased risk of being exploited and harassed along the way (Smith, 2019). This is especially true given that transit heightens the risk of engaging with law and security officials who have been reported to detain, repatriate and violently treat refugees (Amnesty International, 2021; Human Rights Watch, 2023). These forms of informal labour and their financial and bodily costs are integral to overcoming obstacles to obtaining cash assistance and to making digital ID possible. However, such afflictions are not perceptible to the public and in public accounts that digitised cash assistance supports empowerment, livelihoods and streamlined aid delivery (e.g., UNHCR, n.d.). The invisibility of such labour and its burdens reveals how repair work by informal workers does not benefit from protection that is often afforded to more formally employed staff.
As part of the validation process, refugees receive a text message on their phone to tell them when they must validate their identity and where to go. However, many refugees struggle to comprehend such messages. Several studies have found numeracy and literacy skill levels to be ‘alarmingly poor’ amongst the nation's refugee populations (e.g., UNESCO, 2020). For instance, reports by USAID (2017a) claim that roughly 17% of Syrian refugees in Lebanon are illiterate, with women – especially women with disabilities – facing even more barriers to accessing education (USAID, 2017b). To decipher their SMS messages and ensure they are correctly understood, recipients who struggle with literacy and numeracy seek help from neighbours and family members (Smith, 2019). As repair work, mobilising community members represents mundane and informal strategies that are integral to overcoming the barriers of aid systems and to ensuring that refugees comply with validation demands. In doing so, refugees can continue to access cash assistance that they need to survive. Given such improvisations, we see how digital ID does not necessarily make work more efficient. Instead, by enabling rules that are far removed from the lived realities of refugees, digital ID has shifted what types of labour are required and who is responsible for supporting its processes.
Sometimes, biometric identity validation fails. When it does, the United Nations requires refugees to travel to registration centres again so that recipients can participate in additional forms of identity validation. Yasmeen is an interview participant with several years of experience working for a Foreign Aid Organisation in Lebanon. Her role centres around coordinating basic assistance programmes. Yasmeen said: Sometimes, with biometrics, the iris scan doesn’t go through, as some iris’ do not scan well. And sometimes there are technical glitches. This means the family has to show up again in-person to do another type of [ID] verification.
While Shoshana Magnet emphasised in 2011 that biometrics are often inaccurate and reproduce structural power asymmetries, as evidenced by Yasmeen's statement, such discrimination is still a problem in contemporary digital ID applications. Biometrics continue to disproportionately fail at capturing, and hence require additional repair work from, marginalised groups. Factors compelling marginalised groups to exert additional effort to prove their identity – and to thus compensate for failing technology – include: race (e.g., facial scans are poor at capturing Black people), age and disability (e.g., iris scans struggle to register those with cataracts), class (e.g., those who work in labour-intensive, industrial jobs have eroded fingerprints) and gender identity (e.g., transgender people display changing gender information and biological characteristics) (Magnet, 2011). As such, mandating digital ID registrations without any way to circumvent them consequently reflects and embeds discrimination, offloading repair work to oppressed groups who must compensate for the technology's inability to register differences.
Interview participant, Huda, resides in the Middle East and researches digital technologies and privacy. Huda recounted how, while conducting research in Lebanon, she witnessed biometric technology stop working at a post office (LibanPost) in Batroun. Her example illustrates how people navigate the breakdown of biometric technology and restore its functionality when validating identity: …there were some Syrian refugees getting their eyes scanned to get their [cash assistace] cards topped up. I watched the system fail multiple, multiple times and cause all these queues. The two cashiers in the LibanPost branch in Batroun were having to reboot the system to get it to work. Yeah, it was interesting. I was like, this is not as smooth and streamlined as they sell it as [laughs].
In summary, this section has reviewed the often invisible repair work that goes into making digital ID workable and feasible. Such labour contrasts popular trends to think of biometrics as leading to less work and higher quality of work. Repair work that compensates for the burdens and failures of data-driven infrastructure is less prominent in the public consciousness. When digital ID faces issues, we are inclined to believe that formally employed humanitarians are solely responsible for mending the technology and maintaining its functionality – especially those working for humanitarian agencies who choose to implement it and are uniquely positioned to make its rules. Instead, as shown here, refugees and community residents (e.g., store clerks, taxi drivers, neighbours looking after refugees’ children) labour to ensure that biometric enrolment happens, its issues are addressed, and its functionality is restored when it does not operate as intended.
These actors comprise an integral but overlooked ecosystem of reproductive work. Their work is essential to adhering to unilaterally mandated rules on identification – rules that are unreflective of the lived experiences of the people from which it requires compliance. By providing the necessary material resources and effort to meet the requirements of digital ID infrastructures, the United Nations becomes dependent on these unpaid workers to reproduce refugee livelihoods and the viability of ID infrastructures. As social reproduction, such work also reproduces public beliefs that digital ID interventions are a feasible option for aid delivery, thereby reproducing the United Nations’ own status as Lebanon's humanitarian body (Hannaford, 2020). In doing so, this division of labour reflects how the humanitarian sector can invisibilise workers who operate out of sight, which upholds globalised inequities between the poor and those who dictate aid.
Repair work for automated vulnerability targeting: contesting, updating and maintaining data-driven infrastructures
While the repair work involved in digital ID reveals the hidden and costly yet critical labour of supporting data-driven aid, in other cases, repair work also supports contesting data-driven infrastructures behind the scenes. This can be seen in the United Nations’ use of automated poverty targeting, which uses variables indicative of insecurity and vulnerability. The variables are collected by an annual, nationally representative survey of refugees, the Vulnerability Assessment of Syrian Refugees. Employing a proxy-means test, these variables are factored into a formula used to predict individuals’ expenditure and rank their ‘score’ against others. Yasmeen, who works for a Foreign Aid Organisation in Lebanon, said that computational methods are supposed to improve the capacity to target aid to the most in-need people, thereby making access to aid more equitable. In addition, by maximising data collected in the Vulnerability Assessment of Syrian Refugees, the algorithm is supposed to cut costs by removing the need for interviewing individuals and conducting household visits to assess vulnerability (Chaaban et al., 2018).
Deliberate opacities and questions of fairness in poverty targeting algorithms have been an emergent area of debate (Valdivia et al., 2022). In Lebanon, the United Nations has been critiqued for being opaque about the targeting algorithm, preventing access to critical information about how the algorithm makes decisions and the data it requires. The primary justification for keeping the formula ‘secret’ is that it will ‘prevent beneficiaries from “gaming the system”’ (Development Pathways, 2018). This is confirmed in a report by Gabrielle Smith (2019: 9), which states that United Nations agencies have justified their decision to keep the targeting formula opaque due to concerns that ‘revealing variables could lead to fraudulent claims for assistance’. Numerous participants confirmed that control of and insight into the algorithm remains in the hands of United Nations agencies, which has been corroborated in other reports. For instance, Hellberg (2018) notes that the European Commission's Department for Humanitarian Operations and the United Kingdom's Department for International Development critiqued the United Nations’ automated targeting mechanism for prohibiting other stakeholders from having input. This participatory opacity has excluded targeted recipients, as well. An impact evaluation on behalf of the Lebanon Cash Consortium states, ‘the formula is the result of statistical methods, with limited – if any – participation of targeted communities in the decision making process’ (Battistin, 2016: 5).
Several participants commented that, due to the United Nations’ decisions to target aid with an algorithm and to limit the amount of public information about it, there were low levels of understanding about how poverty targeting works among recipients and humanitarian workers. This leads to breakdown in humanitarian operations and breakdown in trust of the United Nations’ approach to poverty alleviation. For instance, Ashe is a Foreign Aid Organisation worker who helped develop the algorithm's targeting criteria and support its integration into operations. During our interview, Ashe noted that one of the biggest challenges in rolling out the algorithm was explaining how it worked to refugees. Paula is an interview participant who works in Lebanon for a European Aid Organisation on social protection. Paula said that the formula is too complex for recipients to understand why or why they were not selected, which leads to significant frustration. A report by Development Pathways (2018) noted how one beneficiary explained algorithmic targeting by saying, ‘It is luck! The computer picks names and assistance is given to those names’. Algorithmic complexity and opacity have consequently led to ‘community tensions between the perceived "winners and losers" of the targeting exercise’ (Smith, 2019: 10).
In terms of humanitarians, Ashe said that humanitarian workers asked the United Nations Refugee Agency and World Food Programme a significant number of questions about the targeting algorithm. This confusion and opacity over the algorithm amongst aid workers has been confirmed by external reports, with one finding that: Even the field staff administering the assistance in Lebanon would sometimes tell people that the programme uses random selection, or would not know what the selection criteria are, adding to the frustration of refugees… When you ask them why [targeting selects people], they say it was the computer's selection. (Development Pathways, 2018)
Repair work also came about by humanitarian actors opposing the algorithm's reliability, applicability and relevance. Ground Truth Solutions (2018: 6) published their survey of humanitarian staff and refugees’ perceptions in Lebanon, which found that 73% of refugees said targeting ‘does “not really” or “not at all” go to the most vulnerable’. Ashe told me that, in her work on developing the algorithm's formula, it was common for employees of humanitarian agencies to communicate to United Nations agencies that they were uncomfortable with implementing the algorithm. Several interview participants confirmed the prevalence of these concerns amongst the humanitarian community, but each noted different aspects of the algorithm that were vocally contested in public reports (e.g., Ground Truth Solutions, 2018), to United Nations agencies, to one another, and to me as a researcher.
Mia, an interview participant residing in Lebanon, works on monitoring a Foreign Aid Organisation, including its digitised cash assistance. Mia noted that humanitarian workers often reject the formula because the United Nations has not clarified how targeting operates and what it is based on. Such bewilderment and opacity, Mia said, are compounded by the algorithm's complexity and the United Nations’ decision to safeguard the information it releases.
Maha, an interview participant residing in Japan, manages a Foreign Aid Organisation's social assistance programmes in Lebanon. She advocated for all United Nations agencies to remove the algorithm and instead adopt what she called ‘a categorical approach, which is a life cycle vulnerability approach’. She said, ‘How do you target the poor when everyone is poor? So, it's better to target by category (children, disability, single-headed household)’.
Nada is a researcher who helped develop the algorithm used in the nation. She commented that there is extensive controversy over updating the targeting formula on a yearly basis. Updating and changing the algorithm's vulnerability criteria did not make sense, she said, when people who are extremely poor in Lebanon generally stay in that condition for a while and thus remain eligible for aid. Nada said, in her opinion, ‘it's not fair if someone was in the extremely poor category and then they’re not when formula gets updated’.
Although such opposition may seem mundane and pointless, voicing concerns about the algorithm helps destabilise the algorithm's taken-for-granted benefits as well as the United Nations’ decision to continue implementing it. Undeniably, humanitarian workers and refugees experience unique issues and burdens related to the algorithm (e.g., emotional work, added tasks) and are subject to distinct power relations. Refugees, in particular, have an enhanced risk of harm, as they are in jeopardy of being denied access to life-saving aid if the algorithm deems them ineligible. Nevertheless, their collective resistance powerfully challenges the legitimacy of data-driven aid infrastructures. As discursive forms of repair work, contestation by humanitarians and refugees thus aims to mend the algorithm's inequitable effects and sustain fairness in aid delivery.
United Nations agencies enacted a few forms of repair work to remove the algorithm's opposition, mend frustration over its opacity and remedy its miscomprehension. Because many humanitarians did not understand how the algorithm worked, Ashe said the United Nations Refugee Agency had to dedicate time and resources to develop ‘interactive training’ sessions for humanitarian staff. These trainings collected and drew upon feedback from a small group of workers, which enabled the Refugee Agency to educate humanitarian staff ‘in simple terms’ and explain its ‘added value’. Ashe said that, on top of enabling workers to better answer refugees’ questions, the Refugee Agency did this to ensure that workers would buy into an algorithmic approach so that their agencies would actively use the algorithmic method. As such, the United Nations encouraged and persuaded acceptance of their algorithmic approach to eligibility assessments.
Efforts to educate refugees involved community engagement and refugee-specific training, which Ashe said were top priorities for the United Nations. The United Nations Refugee Agency and World Food Programme also created a joint call centre where refugees call a hotline and receive information from help-desk operators about any component of digitised assistance. Yasmeen framed the call centre as necessary to increase inclusion pathways in ‘data-driven approaches’ because it enables more ‘qualitative’ and ‘subjective… determinations of who's included’. A country report by the World Food Programme (2019: 16) found that in 2019, the vast majority of calls (72%) ‘were on targeting issues (including appeals)’, which suggests that a considerable number of refugees challenged the automation of poverty targeting.
As Smith (2019: 11) writes, ‘in acknowledgement of the imperfect nature of targeting’, as of 2018, the call centre offers what is called a ‘grievance redress mechanism’. If refugees believe they were wrongfully excluded from assistance by the targeting algorithm, they can call to ask for a grievance redress mechanism, which re-assesses their vulnerability in more detail. Yasmeen framed this repair work as a ‘refugee driven approach to inclusion’, explaining that ‘families are basically profiled again using the data that [the United Nations Refugee Agency] has, so no additional data is gathered. But this re-selection process happens based on more detailed, longer profiles of families’. Data analysts look for vulnerabilities like, for example, female-headed households, family members with a disability, or risk of child labour. By compelling refugees to flag to the United Nations that their data profile counts as one of the incredibly poor, the grievance redress mechanism thus places the onus on refugees to prove that the algorithm wrongfully assessed their vulnerability. Put another way, repair work through the grievance redress mechanism displaces the responsibility of algorithmic misclassification from the United Nations onto refugees, including the responsibility to update and improve the algorithmic formula.
In order to update the targeting model yearly, researchers use the grievance redress mechanism to capture which profiles have been wrongfully excluded. Combining the profiles of people submitting grievance redress mechanisms with the most recent data from the Vulnerability Assessment of Syrian Refugees, Yasmeen said that in 2019, ‘the main household that was determined to be excluded was elderly with small household’. The United Nations thus relies on grievance redress mechanisms to maintain the algorithm and make its targeting capacity more effective. This reproductive labour is supportive of the longevity and accuracy of automated poverty targeting, providing the necessary data and accountability measures that fuel the algorithmic formula and keep it ‘accurate’ and useable.
As evidenced by the procedure for updating the formula, the algorithm requires different types of repair work. On the one hand, repair work constitutes the labour of humanitarians who help keep the algorithm up-to-date and address questions about the algorithm and algorithmic failure. This data labour has been described as ‘[a]nother downside of the PMT [proxy means testing] approach … [because] it is very costly and time-consuming’ (Development Pathways, 2018). On the other hand, refugee repair work contests the veracity of the algorithm and submits requests to be reassessed through the grievance redress mechanism. While humanitarian workers are compensated for their labour, refugees’ grievance redress labour is not. Refugees’ efforts are compelled by the need to survive rather than having to complete tasks that are a part of formal employment. Nevertheless, refugees’ grievance redress labour is integral for keeping the formula relevant and helping United Nations agencies deliver aid to broader categories of vulnerable groups in need. By recognising that the algorithm requires perpetual updating, iteration and evolution – a core feature of broken world thinking as explained by Jackson (2014) – refugees’ grievance redress labour therefore constitutes critical yet behind the scenes algorithmic repair work. This work is critical to reproducing algorithmic infrastructures and the populations who rely on the aid delivered through such infrastructures.
The work of algorithmic assessments is much more complex than simply computing variables to support efficiencies offered by predictive analytics. While algorithmic targeting may remove the labour of humanitarians conducting interviews and household visits, there is a considerable amount of backstage labour to support algorithmic systems. Refugees and various humanitarian workers complete labour that compensates for algorithmic failures, blocked access to aid and opacity, which are largely driven by data-driven decisions about how eligible recipients will be selected and who has the right to such information.
Conclusion
This article has unpacked the repair work involved in digital humanitarianism. It explained how repair work captures a particular type of mundane work – work that goes into responding to breakdown using stopgap measures – and how this can be aligned with feminist understandings of asymmetrical divisions of labour as well as resistance. By applying these insights to two data-driven infrastructures, digital ID and algorithmic poverty targeting, this article revealed many forms of humanitarian repair work. In particular, digital ID indicates the necessary work that goes into complying with rules and compensating for malfunctioning biometrics. And, algorithmic poverty targeting revealed how automation is contested as well as updated and maintained. Collectively, these forms of repair work reflect labour that is both invisible and reproductive. Such labour is often backstage, offloaded, differently valued and peripheral, but without this labour, digitised aid would be hard-pressed to operate.
In these datafied environments, technology and humanitarian firms are promising efficiency, streamlining work and cutting costs, yet their technologies are generative of material burdens, delays in accessing aid and the insecurity of recipients. Formal contracts do not account for or compensate repair work, which toils to restore, counteract and pay for the issues reified by datafication. Deficient recognition of such labour and its burdens is partly related to ‘our capitalist society’ in which ‘we tend to value work that we can see’ (D'Ignazio and Klein, 2020: 178).
By considering the often unseen experiences of frontline aid workers and residents in Lebanon, this article suggests that workloads are not always reduced or streamlined by data practices. Contrary to public imaginaries, these shifts in labour are not a lessening of work, nor do they always increase efficiency. Instead, the datafication of humanitarian infrastructures has generated different kinds of work. The disparate forms of repair work revealed in this paper are emblematic of ‘burden shuffling’, which Seim (2020) describes as one worker shifting onerous work tasks from their plate onto another's, likely to make their labour easier. In the case of digital humanitarian infrastructures, the labour of complying with biometric rules and compensating for its malfunctioning and discriminatory effects are pushed from the United Nations onto recipients and communities, especially those with intersecting axis of marginalisation. These consequences go directly against public imaginaries that data-driven aid will make humanitarian labour more efficient and ‘[build] on the relative strengths of different agencies’ (Pelly and Juillard, 2020: 15).
Not only does data-driven aid support the displacement of responsibilities and precarity. It is also generative of extractive labour relations. The asymmetrical labour dynamics recounted in this paper speak to broader forms of extraction and invisible labour politics embedded in the use and imbalances of datafication, which other scholars have claimed transfer power in the hands of actors like humanitarian institutions, states and corporations (Sandvik, 2023). As Tazzioli (2022) writes, refugees dually figure as sources from which data is extracted and they are also indirectly and imperceptibly mobilised to support humanitarian data production. By revealing how these ‘workers’ participate in the generation and resistance of data-driven humanitarian processes, the divisions of labour recounted in this article contribute to understanding the ‘infrastructures of data extraction and data circulation that sustain’ aid delivery (Tazzioli, 2022: 71).
The extraction and invisibility of repair work that supports ‘technology for good’ affects how the public understands where forms of breakdown occur, who is mobilised and made responsible for repairing humanitarian problems and the extent to which data-driven assistance meaningfully supports needs on the ground. Humanitarian workers contest, negotiate, strategise and adapt to technologically constituted breakdown to make up for its burdens and breakages (literally and implicitly). This repair work is complex and messy, cultivating considerable disagreement, diverse strategies and unique justifications.
Although repair work is integral to sustaining aid delivery in the nation, it also reveals asymmetrical labour relations by compelling added work, fostering disenfranchisement, displacing responsibilities onto marginalised groups and staff without decision-making power and contributing risks to refugees’ livelihoods. These findings on labour offer insight into perils and contested promises of Technology for Good. Rather than simply reducing workloads and enabling higher service standards, digital humanitarianism is much more complicated in everyday practice.
Footnotes
Acknowledgements
My deepest thanks to Kathryn Henne for her dedicated support, mentorship, and critical feedback. Thank you also to Maha Rafi Atal, Sofie Elbæk Henriksen, and Lisa Ann Richey for their review and suggestions.
Declaration of conflicting interests
The author declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The author disclosed receipt of the following financial support for the research, authorship and/or publication of this article: Jenna Imad Harb's research is funded by the School of Regulation and Global Governance, based at The Australian National University.
