Wilson is part of a massive global network of freelancers helping train AI models for companies like Outlier AI and Handshake AI.
Many of them earn only about 20 CAD/hour (equivalent to $14.6 USD), in unstable “gig work” conditions – lacking fixed hours or benefits.
Some more specialized jobs, like correcting scientific data, can pay 40 CAD/hour, but the workload is inconsistent.
Experts call this “fine-tuning” – the stage of refining the model by evaluating AI responses and retraining the system via “reinforcement learning from human feedback” (RLHF).
When ChatGPT or Claude “sound human,” it’s people like Wilson who trained them to be more “natural.”
Outlier AI has over 250,000 collaborators in 50 countries, 81% of whom hold university degrees, according to Scale AI (the parent company).
However, the market is changing: demand for generalist labor is decreasing, replaced by personnel with specialized knowledge and advanced degrees, as AI becomes more complex.
Some new models, like DeepSeek (China), have partially automated the fine-tuning process, making human labor more replaceable.
Yet, AI remains heavily dependent on low-wage labor in developing countries. Many workers in Kenya, Uganda, and the Philippines work up to 70 hours/week for just over 1 USD/hour, in conditions dubbed “digital sweatshops.”
Researcher James Muldoon states that millions of people are “feeding” AI through monotonous, tedious work, which forms the backbone of the global AI economy.
📌 Summary: Behind AI’s “magic” are millions of hidden AI trainers. Examples include: DataAnnotation (checking grammar, accuracy) and Outlier AI (>250,000 collaborators, 50 countries, 81% with degrees). The trend shows declining demand for general labor, replaced by specialists with advanced degrees due to AI complexity. However, AI still relies heavily on cheap labor in developing countries (e.g., Kenya, Uganda, Philippines) working 70 hours/week for >$1/hour in “digital sweatshops.”
