Data Workers’ Inquiry: recentring workers’ epistemic authority

3 min read

Behind artificial intelligence’s gleaming promise of innovation lies an uncomfortable truth: its success depends on an invisible workforce doing exhausting, often traumatic work. These data workers – predominantly in the Global South – annotate, moderate and process the vast data streams that power AI systems. They are the uncredited architects of artificial intelligence, their labour rendered invisible by design.

Recognising the need to confront this erasure, we launched the Data Workers’ Inquiry (DWI) to better understand the realities of this hidden workforce. This pioneering initiative, funded by the Distributed AI Research Institute (DAIR), the Weizenbaum Institute and TU Berlin, seeks to amplify the voices of data workers and recentre their epistemic authority. For us, it was imperative that the workers were not simply research participants or interviewees; they were community researchers co-researching with us. From shaping the inquiries to determining what forms their output would take, they took ownership and leadership over the project, deciding what mattered most to them and how it should be presented.

Our team, led by Dr Milagros Miceli, alongside Dr Adio Dinika, Krystal Kauffmann, Camilla Salim Wagner and Laurenz Sachenbacher, collaborated with 17 co-researchers from diverse geographical locations – Kenya, Venezuela, Colombia, Nigeria, Syria, Lebanon, Brazil and Germany – and a range of employment contexts. The community researchers, including content moderators and data annotators working for outsourcing companies and platforms, produced a remarkable variety of outputs: vlogs, documentaries, animations, essays, podcasts, comic strips and more. Through their creative and personal contributions, the project provides a rare and intimate glimpse into the lives of those who sustain the AI industry, exposing the profound human cost of this labour.

Their findings are both harrowing and essential. In Nairobi, data workers admit turning to dangerous drugs as a desperate measure to cope with the relentless psychological trauma of reviewing graphic violence daily, compounded by the absence of adequate mental health support. In Venezuela, Alexis Chávez shares the indignity of being paid in gift cards, forcing him into convoluted exchanges to convert them into usable currency. In her powerful report, Fasica Berhane Gebrekidan, a content moderator from Tigray, Ethiopia, recounts the heart-wrenching burden of reviewing material tied to the war ravaging her own community. She describes the agony of being exposed to graphic videos and images of atrocities, often recognising places and people from her home region. This proximity to the conflict magnifies the emotional toll, as she is forced to repeatedly view and categorise images of violence from her home region, leaving her with a haunting sense of complicity – an involuntary participant in documenting atrocities she cannot prevent. Another particularly poignant project is the Zine by Botlhokwa Ranta, which focuses on the challenges faced by mothers working in content moderation, underscoring the gendered dimensions of this labour and its impact on families and communities.

These accounts lay bare the systemic exploitation ingrained in the AI industry, revealing a web of vulnerabilities that outsourcing firms specifically exploit. Refugee workers, already marginalised by their precarious legal status and limited employment options, are prime targets for firms that offer minimal compensation, no benefits and zero job security. In Kenya, for instance, the firms targeted the recruitment of people living in the slums. These firms capitalise on desperation, knowing that these workers have few alternatives. The targeting of vulnerable people also extends to countries like Germany, where migrants often face the additional burden of contracts tied to their residency permits, leaving them at the mercy of employers who can dictate exploitative terms without fear of reprisal. The situation of migrant data workers in Germany is graphically described in Sakine Mohamadi Bozorg’s essay and the podcast produced by Meta content moderators at the outsourcing firm Telus Digital in Essen. For these workers, the stakes are high – not just their livelihoods, but their legal right to stay in the country depends on keeping their jobs, no matter how unjust the conditions.

In Lebanon, Roukaya al-Hammada’s report reveals a wider pattern of instability. Short-term contracts, designed to provide only subsistence-level income, create a precarious existence where planning for the future becomes impossible. This economic fragility traps workers in cycles of uncertainty, forcing them to live paycheck to paycheck while enduring relentless workloads. The exploitative practices are further compounded by the global fragmentation of the workforce, where workers are scattered across geographies to prevent unionisation or collective action. These accounts underscore how the AI industry, while championed as the frontier of innovation, relies on labour practices that entrench systemic inequities and perpetuate a global divide.

One particularly insidious practice is the rampant use of non-disclosure agreements (NDAs) to silence workers about the nature of their jobs. These NDAs go beyond the standard restrictions of discussing company information with the press or the public – they extend into workers’ private lives, cutting off essential avenues of support. For instance, one worker shared how her NDA prohibited her from revealing the specifics of her role even to her psychotherapist. Unable to discuss the graphic content she reviewed daily, she was forced to say, “I work for a call centre” – a vague explanation that rendered her counselling ineffective. In the context of data work, NDAs deepen the isolation of already vulnerable workers, creating an environment where exploitation thrives in the dark. By forbidding transparency, these agreements not only hide the true cost of AI labour, but also perpetuate the very conditions that lead to worker burnout, trauma and disempowerment.

The Data Workers’ Inquiry (DWI) unpicks this fragmentation and disempowerment. Through participatory action research, we foster solidarity among workers, creating bridges that connect their stories and experiences across borders. Workers who were once isolated and silenced are beginning to find common ground, sharing strategies and building alliances. What emerged from this process were not victim narratives, but bold collective testimonies from agents of change – workers reclaiming their agency and demanding accountability from an industry that has long thrived on their invisibility. In addition, encouragingly, workers are beginning to organise and communicate, challenging the industry’s grip on their labour. In Africa, data workers have formed the African Content Moderators Union, creating a collective platform to advocate for their rights and push back against exploitative practices. These efforts are not limited to one region. Transnational solidarity is growing, with organisations such as Turkopticon, which supports workers on platforms such as Amazon Mechanical Turk, engaging with the African Content Moderators Union to share insights, resources and strategies. This cross-border collaboration signals a growing movement that transcends geographic and cultural divides, laying the foundation for a global network of resistance.

Through these efforts, the narrative is shifting. Workers are no longer passive recipients of injustice, but active participants in shaping their futures. Their stories and collective actions remind us that systemic change is not just possible; it is already underway, driven by the courage and determination of those who refuse to be erased. Imagine an AI industry where the well-being of its workforce is a priority, where labour rights are upheld, and where innovation is driven by principles of justice. This is the future that the Data Workers’ Inquiry envisions, and that the workers’ blueprints call upon us to create.