- Tina Lynn Wilson, 45, in Hamilton, Canada, has worked as a freelancer for DataAnnotation since January 2025, specializing in checking AI responses for grammar, accuracy, and creativity – a job known as “AI training.”
- 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), with unstable employment and characteristic of “gig work” – lacking fixed hours or benefits.
- Some more specialized jobs, like scientific data calibration, can reach 40 CAD/hour, but the workload is irregular.
- Experts call this fine-tuning – the stage of refining the model by evaluating AI responses and retraining the system through “reinforcement learning from human feedback.”
- When ChatGPT or Claude “sound like a real person,” it is people like Wilson who have trained them to become more “natural.”
- Outlier AI has over 250,000 collaborators in 50 countries, 81% of whom have a university degree, according to Scale AI (the parent company).
- However, the market is changing: the demand for general labor is decreasing, replaced by personnel with specialized knowledge and high qualifications, as AI becomes increasingly complex.
- Some new models like DeepSeek (China) have partially automated the fine-tuning process, making human workers easily replaceable.
- Nevertheless, AI still heavily relies on cheap labor in developing countries. Many workers in Kenya, Uganda, and the Philippines have to work up to 70 hours/week for wages just over 1 USD/hour, in conditions referred to as “digital sweatshops.”
- Researcher James Muldoon states that millions of people are “feeding” AI with monotonous, tedious work, but this work is the backbone of the global AI economy.
📌 Behind the “magic” of AI are millions of anonymous workers who train the AI. Examples include: DataAnnotation specializes in checking AI responses for grammar, accuracy, and creativity; Outlier AI has over 250,000 collaborators in 50 countries, 81% of whom have a university degree. The trend is a decreasing demand for general labor, replaced by personnel with specialized knowledge and high qualifications, as AI becomes increasingly complex. However, AI still heavily relies on cheap labor in developing countries like Kenya, Uganda, and the Philippines, where workers work up to 70 hours/week for wages just over 1 USD/hour, in conditions referred to as “digital sweatshops.”
