As artificial intelligence (AI) continues to advance rapidly, a vast and often unseen workforce is laboring under challenging conditions to support its functionality. The coverage surrounding AI frequently emphasizes the looming threat it poses to traditional job markets, painting a picture where machines may soon usurp human roles. However, the intricate reality reveals a burgeoning global workforce, comprising millions of individuals dedicated to refining AI systems through data sorting, labeling, and categorization. This indispensable labor, often conducted by those referred to as “humans in the loop,” is fundamental to the learning processes of AI algorithms for major technological entities like Meta, OpenAI, Microsoft, and Google.
The demand for such labor has largely transitioned to countries with sizable populations willing to work for low wages, including nations like Kenya, India, the Philippines, and Venezuela. In Nairobi, Kenya, for instance, where youth unemployment hovers near 67 percent, employment in AI has become a source of optimism for many. Workers like Naftali Wambalo, armed with college degrees, engage in intensive data processing tasks—spending hours labeling innumerable images and videos to enable AI systems to perform human-like recognition tasks. Their efforts span a vast array of activities, from tagging household items to identifying various racial features in facial categories.
Despite the potential of these roles, they are often marred by grim conditions, prompting civil rights advocates to label them as “modern-day slavery.” As noted by activist Nerima Wako-Ojiwa, large American tech firms engage in exploitation under the pretext of creating employment opportunities. Employment is usually through short-term contracts lacking stability, and the work environments mirror sweatshops, with hundreds of workers confined to computers in harsh conditions. Although the Kenyan government is keen to attract foreign companies by branding itself as the “Silicon Savannah,” workers are not directly employed by these tech giants—they are contracted through American outsourcing firms like SAMA. Despite SAMA receiving higher fees for each worker they contract, the pay for laborers amounts to a mere $1.50 to $2 hourly before taxes, which is severely disproportionate to the cost charged to the companies.
Moreover, the burdens placed on these employees are immense, with many being subjected to unrealistic expectations and rapid turnaround times for complex labeling tasks. Any complaints regarding their working conditions could lead to immediate dismissal, fostering a culture of fear among workers. Holding the most detrimental impact, several employees have described their experiences training AI systems to filter out violent or disturbing content—a task that comes without adequate mental health resources or support. Repercussions of these intolerable conditions have led to a group of nearly 200 workers initiating legal action against SAMA and Meta, asserting that these companies implicate themselves in the harmful working environments while exploiting individuals from vulnerable backgrounds.
Illustrating the ambiguity in the realm of AI, multiple reports highlight the phenomenon of “AI washing,” where companies exaggerate their use of AI by relying heavily on human labor. A notable instance involves Amazon’s “Just Walk Out” technology, which is presented to customers as powered predominantly by AI, allowing them to shop effortlessly within Amazon Fresh and Amazon Go. However, scrutiny reveals that human employees in India were necessary to manually verify nearly three-quarters of transactions—contradicting the portrayal of a fully autonomous system. Amazon refuted these claims, arguing that the Indian employees were merely quality-checking the technology.
This prevalent practice of AI washing arises from various stimuli, including competition for venture capital and the desire to project an innovative front within the tech industry. Reports reveal that only 10 percent of tech startups featured AI in their business model pitches in 2022, a figure that surged to over 25 percent in 2023, with forecasts predicting that it may climb to one-third this year. Despite the rising volume of claims regarding AI utilization, industry insiders, like Sri Ayangar from OpenOcean, emphasize the stark contrast between companies that assert capabilities in AI and those that can substantively deliver AI-driven outcomes. This discrepancy raises concerns about the integrity of AI advancements and the reliance on human labor behind the striking technological claims touted by numerous companies.
