A research project led by a Brown University student seeks to expose a reality often overlooked by the public: behind generative AI models lies a human supply chain made up of thousands of workers tasked with labeling data, moderating content, or evaluating model outputs. This work, essential to training and fine-tuning AI systems, is largely outsourced to contractors operating in low-wage countries, including the Philippines, Kenya, and Madagascar.
The project follows a line of journalistic and academic investigations that have already highlighted the difficult working conditions faced by these annotators, who are sometimes exposed to violent or disturbing content as part of moderation tasks, and paid at rates far below those common in Western countries. The Brown research aims to systematize this documentation by mapping the different stages of the data supply chain, from labeling firms to the major AI labs that rely on their services.
This effort comes amid growing academic and media attention to the hidden labor underpinning AI, as large language models require ever-larger volumes of human-annotated data, particularly for reinforcement learning from human feedback (RLHF). Technology companies face mounting pressure to disclose more information about their data sourcing practices and the working conditions of their subcontractors.
Beyond its documentary value, this type of research raises a structural question for the AI industry: reliance on a poorly paid and largely invisible workforce remains a persistent blind spot in discussions about technological innovation, one that arguably warrants greater transparency and regulatory scrutiny.