The crucial yet often overlooked tasks of data moderation and annotation are typically obscured from those who benefit from their outcomes. A new initiative aims to highlight the lived experiences of data workers globally, revealing both the challenges and opportunities of tech-related employment abroad.
Many monotonous, thankless, and psychologically taxing jobs have been outsourced to lower-income countries, where workers often accept meager wages compared to their American or European counterparts. This labor market mirrors other sectors deemed "dull, dirty, or dangerous," such as electronics recycling and shipbreaking. While data moderation and annotation work may not involve life-threatening dangers, it is far from safe, pleasant, or fulfilling.
The Data Workers’ Inquiry, a partnership between the AI ethics research group DAIR and TU Berlin, draws inspiration from Marx’s late 19th-century writings on labor conditions, delivering reports that are designed to be "collectively produced and politically actionable."
All reports are freely accessible, launched today during an online event where project leaders provided insights into their findings. The ever-growing array of AI applications relies heavily on human expertise, which is still acquired at a minimal cost to corporations, often without concern for the individuals involved. When a user reports inappropriate content, the underlying process—often involving underpaid workers in regions like Syria—remains hidden from public view. Due to the overwhelming number of reports and potentially harmful content, companies often resort to outsourcing, prioritizing cost over ethical considerations.
The reports, primarily anecdotal in nature, offer systematic observations akin to anthropological studies rather than quantitative analysis. Attempts to quantify such experiences often fail to depict the true costs involved. Statistics frequently highlighted by companies—such as higher wages in comparison to local competitors or job creation—rarely mention the mental toll on moderation workers, such as sleep disturbances or substance dependencies.
Take, for example, Fasica Berhane Gebrekidan’s report detailing the struggles of Kenyan data workers with mental health and addiction issues. (You can find the full PDF here.) Although they worked for Sama, which promotes itself as a more ethical provider, Gebrekidan's report reveals a harsh reality filled with bleak working conditions and inadequate support from local management.
An image shared by a whistleblower depicts the challenging environment faced by moderators at Samasource in Kenya. Recruited to process flagged content in local languages, these workers confront an almost constant stream of violent, graphic, and abusive material that they must handle swiftly to avoid penalties affecting their pay. Depending on workload dynamics, some workers may face just over one disturbing piece of content per minute, culminating in around 500 items to review daily. While AI is often involved, these workers serve primarily as the training data providers.
"It's absolutely soul-crushing. I’ve witnessed unimaginable horrors. I fear I’ll carry the scars of this job for life,” expressed Rahel Gebrekirkos, one of the contractors interviewed. Support personnel were described as "ill-equipped, unprofessional, and underqualified," prompting many moderators to seek solace in substances, resulting in issues like depressive thoughts and anxiety.
While similar concerns have been raised previously, it's crucial to acknowledge that these troubling dynamics persist. In addition to numerous reports, individual narratives illustrate the situation further. For instance, Yasser Yousef Alrayes, a data annotator in Syria, shares his journey towards funding his education while managing visual annotation tasks that often come with unclear expectations and frustrating client demands. He documented his experience in an impactful eight-minute short film worth watching.
Workers like Yasser tend to be hidden behind multiple layers of organization, acting as subcontractors within a convoluted structure that obscures accountability in the event of issues or legal challenges. Milagros Miceli, one of the project leaders at DAIR and TU Berlin, mentioned that while no feedback or changes have yet emerged from the reported companies, it remains early in the process. The team is committed to further investigation: “We plan to continue this work with a second group of data workers, likely from Brazil, Finland, China, and India.”
Critics may downplay these reports due to their anecdotal nature, yet anecdotes convey essential truths through firsthand accounts. Whether it’s just a few moderators in Kenya, Syria, or Venezuela who face these challenges, their voices should resonate with anyone relying on their crucial work—including virtually everyone in today’s interconnected digital landscape.
Ranking
