Tina Lynn Wilson, a resident of Hamilton, Ont., has been working as a freelancer for DataAnnotation since the beginning of the year. Engaged in assessing responses from an AI model, Wilson finds the work fulfilling as it involves evaluating grammar, accuracy, and creativity. She particularly enjoys unique projects like selecting the superior piece of poetry from two options, which requires a creative judgment rather than fact-checking.
Wilson’s role is part of a vast but less recognized community of gig workers in the emerging AI industry. Companies such as Outlier AI and Handshake AI enlist these workers as “artificial intelligence trainers” to assist in training their models. While some data annotation tasks may be low-paying or exploitative in certain regions, there exists a diverse range of jobs related to training, managing, and rectifying AI systems. However, these labor practices are not commonly discussed by major tech corporations. As AI models evolve, they will demand more specialized training, potentially reducing the need for human involvement in their development.
Human expertise plays a crucial role in refining generative AI systems beyond initial pre-training stages. This fine-tuning process, reliant on human input, requires specialized knowledge and attention to detail. Individuals like Wilson, who consider this work a side hustle due to its project-based nature and variable pay rates, earn around $20 per hour for general tasks, with potential for higher earnings in specialized areas.
The market for AI training work is showing signs of transformation, with a shift towards technical training and automation of certain processes. Companies like Scale AI have laid off generalist workers in favor of individuals with advanced degrees and specialized knowledge. Specialized workers, like Eric Zhou, who freelanced part-time for Outlier AI after studying materials and nanoscience, find joy in tasks that involve assessing and correcting AI responses related to scientific subjects.
Despite the advancements in AI technology, a significant portion of the training process relies on global outsourcing to lower-wage countries, where workers often face grueling tasks in data annotation and labeling. Reports suggest that millions of workers are engaged in these activities, with accusations of exploitation in regions with lax labor laws. The industry has been criticized for operating what some describe as “digital sweatshops” in countries like the Philippines and Kenya, where workers endure long hours for minimal pay under harsh conditions.
While AI companies focus on showcasing the power and potential of their products, the human labor behind AI automation often remains unnoticed. Critics highlight the challenging and sometimes exploitative conditions faced by workers in the AI training industry, emphasizing the need for greater awareness and ethical considerations in AI development.