TLDR: Reports from Mint and Dawn highlight the demanding and poorly compensated human labor essential for training generative AI systems. Workers often face exposure to disturbing content without adequate mental health support, leading to calls for improved conditions and fair wages in a rapidly expanding global market.
The rapid advancement of generative artificial intelligence (AI) systems, capable of tasks like writing autopsy reports or enabling self-driving cars, is underpinned by a vast and often invisible workforce engaged in gruelling, low-paid data labeling. These human workers are tasked with sorting, annotating, and processing immense volumes of digital information, frequently involving disturbing content, with little to no mental health support. This precarious labor has ignited a global movement advocating for better wages and improved working conditions, with efforts spanning from Kenya to Colombia.
Ephantus Kanyugi, a 30-year-old Kenyan national and Vice-President of the Nairobi-based Data Labelers Association (DLA), shared his experience of spending entire workdays examining images of dead bodies and crime scenes to outline wounds for AI training. He emphasized the critical lack of mental health support for workers exposed to such traumatic content. The DLA, an 800-member labor group, is set to unveil a code of conduct this month, urging major labeling platforms to enhance worker welfare.
This issue is not unique to Kenya. Oskarina Fuentes, a 35-year-old Venezuelan residing in Medellin, Colombia, works for five different data-labeling platforms, earning a meager five to 25 US cents per task. She articulated the sentiment of many in this hidden workforce: ‘We’re like ghosts. No one knows that we exist, even though we are contributing to society’s technological progress.’
The data-labeling sector has experienced explosive growth alongside the rise of generative AI. This unseen labor has been instrumental in developing technologies such as self-driving cars that recognize pedestrians, chatbots like ChatGPT that generate natural-sounding language, and automated content moderation systems that filter violent or pornographic material from social media. According to Grand View Research, the data-labeling market was valued at $3.8 billion in 2024 and is projected to exceed $17 billion within the next five years.
The demanding nature of this work has led to confrontations between workers and tech companies. In the United States, nearly 250 GlobalLogic employees, subcontractors involved in training Google’s Gemini generative AI, were terminated in September after raising concerns about pay disparities and demanding better working conditions. Andrew Lauzon, a 31-year-old former GlobalLogic worker, stated, ‘They just want docile annotators.’
Also Read:
- Generative AI Propels Sophisticated Academic Fraud in Chinese Paper Mills
- Global AI Adoption Surpasses One Billion Monthly Users, DataReportal ‘Digital 2026’ Reveals
While some companies, like Scale AI, claim to offer mental health resources and anonymous support hotlines, and assert that their pay rates meet or exceed local minimum wages, the industry continues to face scrutiny. This is not the first instance of the tech sector drawing criticism for exposing low-paid contractors to disturbing content; Meta has faced legal action from content moderators in various countries over similar allegations of poor working conditions and psychological harm.
