How Big AI Companies Exploit Data Workers in Kenya
Artificial intelligence (AI) has revolutionised modern life, influencing everything from the social media platforms we frequent to the chatbots we consult. While AI promises innovation and efficiency, its operations are underpinned by a hidden workforce that is often overlooked and underappreciated. In Kenya, thousands of data workers play a pivotal role in training these systems. Their contributions, however, come at a significant cost, raising questions about labour exploitation, mental health, and equity.
The Essential Role of Data Workers
AI systems rely on vast amounts of annotated data to function effectively. This process involves teaching machines to recognise patterns in images, text, or videos. While AI can distinguish between a cat and a dog based on physical features, more nuanced decisions—such as identifying violent or harmful content—require human intervention.
In Nairobi, individuals like Joan Kinyua, a single mother, work as data annotators. For years, she has tagged objects in images, from vehicles to road signs, helping to train autonomous vehicle systems designed for cities thousands of miles away, such as San Francisco. But the work doesn’t stop at harmless objects. Over time, many data workers are tasked with processing graphic and violent content, exposing them to psychological trauma.
Joan shared that, during her pregnancy, she worked up to 18 hours a day, often encountering distressing material such as violent videos. “It started affecting me to a point that I didn’t even realise,” she recounted. Stories like Joan’s illustrate the physical and emotional toll of AI-related labour, especially when workers lack adequate support.
Content Moderation and Mental Health
The escalation of AI use in moderating user-generated content on platforms like TikTok has increased the demand for data workers. Stacy (a pseudonym, chosen to protect the individual’s identity), an aspiring journalist turned content moderator, reviews up to 200 videos an hour for graphic or violent content. “Seven out of ten videos are graphic,” she revealed, describing scenes of car accident victims and people being burned alive.
Despite their critical role in ensuring online safety, content moderators are often left to manage the psychological repercussions of their work alone. Stacy noted that her employer offers a monthly 30-minute therapy session, an insufficient measure for the severity of their exposure. “They tell you to watch funny videos on TikTok to erase the bad memories,” she said. However, such superficial solutions do little to address the deeper psychological scars.
The Rise of Text Annotation: A New Frontier
The advent of generative AI chatbots like ChatGPT has introduced a new dimension to data annotation—text-based tasks. Faith (a pseudonym), another Kenyan data worker, described her role in training large language models. Initially drawn to the better pay, Faith soon realised the work required her to simulate disturbing conversations, such as describing cannibalism in hypothetical scenarios.
“I had to explain how bodies are mutilated, how human flesh is boiled or fried,” she disclosed. The ethical ambiguity and emotional toll of such tasks eventually drove Faith to quit. Like other workers, she was unaware of the broader purpose of her contributions, highlighting the lack of transparency in the AI training pipeline.
Global Supply Chains and Local Exploitation
Kenya’s position as a hub for AI data annotation exemplifies the outsourcing strategies of global tech giants. Companies like Remotasks, a subsidiary of the US-based Scale AI, have recruited Kenyan workers to train AI systems while paying them significantly lower wages than their counterparts in the West. Data workers in Kenya report earning less than $2 an hour, compared to over $20 for similar roles in the United States.
This disparity is not accidental. As Milagros Miceli, a sociologist researching digital labour, explains, “It is cheaper to pay people in Kenya than in Silicon Valley.” Outsourcing also serves to obscure the existence of these workers, keeping them invisible to consumers and regulators.
In March 2024, Remotasks abruptly ceased operations in Kenya, leaving workers like Joan without warning or explanation. “We just woke up one day and found a notification saying the job was no longer available,” she said. While the company attributed this to “operational errors,” the sudden withdrawal underscores the precarious nature of gig work.
Psychological and Social Consequences
The psychological burden borne by data workers extends beyond individual suffering, affecting their families and communities. Faith admitted that her exposure to disturbing content changed how she viewed the world and led her to restrict her children’s activities out of fear. “I can’t let my kids sleep out, even at their grandmother’s house,” she shared.
The lack of psychological support exacerbates these challenges. Many workers advocate for better access to mental health services and protections against exploitative practices. However, progress is slow, hindered by weak labour laws and corporate disinterest.
Resistance and Unionisation
Despite the challenges, data workers in Kenya are beginning to organise. Efforts like the Content Moderators Union in Nairobi aim to advocate for fair pay, better working conditions, and psychological support. Joan and her colleagues are pushing for the establishment of a formal union to amplify their voices.
“Companies need to take us seriously,” Joan emphasised. “We want better conditions, not the abolition of data work.” The union’s goals include addressing the long hours and inadequate compensation that force workers to sacrifice their well-being for survival.
Ethical Implications and Corporate Responsibility
The ethical implications of exploiting a global underclass to fuel AI development cannot be ignored. While companies like Scale AI and Teleperformance profit from these operations, they fail to provide sufficient safeguards for their workers. A spokesperson for Scale AI justified their practices, stating that training AI to prevent harmful content is “critical to the safe development of AI.” However, this rationale rings hollow in the absence of fair treatment for the workers making this possible.
Experts like Aljoscha Burchardt stress the importance of understanding context in training AI systems. Annotators’ work ensures that chatbots and content moderation tools can distinguish between appropriate and inappropriate uses of sensitive topics. Yet, the lack of transparency and ethical oversight raises questions about the true cost of these advancements.
Conclusion: A Call for Change
As AI continues to transform industries and societies, the plight of data workers in Kenya highlights a critical blind spot in the narrative of progress. These workers form the backbone of AI systems, ensuring their functionality and safety. Yet, they remain underpaid, undervalued, and exposed to significant harm.
To address these injustices, companies must prioritise fair pay, transparency, and mental health support. Governments and international bodies should establish stronger regulations to protect gig workers and hold corporations accountable. The rise of AI should not come at the expense of human dignity.
Kenyan data workers are stepping out of the shadows, demanding recognition and respect. Their fight serves as a reminder that ethical AI development requires not just technological innovation but also a commitment to social justice.
Aric Jabari is a Fellow, and the Editorial Director at the Sixteenth Council.
