The practice highlighted in the 60 Minutes segment—experts such as physicians, winemakers, and music producers spending hours training AI models through platforms like Mercor—has clear short-term gains and longer-term uncertainties.
Pros include rapid improvement of AI in specialized domains where internet-scraped data is insufficient. Models need human judgment on tasks like interpreting lab results or refining lyrics, and companies pay for that expertise.
Contractors report average rates around $85–95 per hour, with specialists reaching $200 or more, creating flexible side income while many keep full-time jobs. AI training has become one of the fastest-growing job categories on LinkedIn. Mercor CEO Brendan Foody argues this mirrors past technological shifts: automating routine work can raise overall demand for human expertise and productivity rather than simply eliminate roles. Some trainers say the process also sharpens their own skills by forcing them to articulate tacit knowledge.
Cons center on displacement risk and precarious work. The same experts who train the systems are sometimes teaching AI to perform core parts of their professions; journalists and others have described the internal conflict of improving tools that could reduce demand for their skills. Project-based gigs are unstable—work can pause or end with little notice, producing feast-or-famine income and no benefits as 1099 contractors. Reports also note monitoring software, abrupt pay changes on projects, and legal disputes over data handling. Broader evidence is mixed but not uniformly reassuring: early data show relative employment declines for younger workers in highly AI-exposed roles, and analyses such as McKinsey’s project that millions of U.S. workers may need to change occupations even if net job numbers eventually rise, with lower-wage workers facing larger skill gaps.
The immediate effect is a paid market for human expertise that makes current AI better. Whether that market remains durable, or mainly accelerates the automation of the expertise being sold, is still unresolved and depends on how quickly demand for remaining human judgment grows relative to what the models absorb.
