Workers earn extra income training AI systems that could eliminate their own jobs

Professionals Train AI Models for Extra Income—And Their Own Replacement

Tasha Kozak, a social worker in Tampa, earns as much training artificial intelligence for 20 hours per week as she does helping families in her 40-hour full-time role with Hillsborough County Public Schools 1

1

. She's one of tens of thousands of white-collar professionals working for Mercor, a San Francisco startup valued at $10 billion that recruits experts to teach AI systems how to perform their jobs. The company, which raised nearly $500 million in venture capital from investors including Peter Thiel, Jack Dorsey, and Larry Summers, operates like an Uber for AI training—connecting skilled workers with gig opportunities at major tech companies including OpenAI, Anthropic, and Meta 1

1

.

Kozak discovered Mercor through a LinkedIn job posting and was interviewed by an AI agent with a conversational style that probed her experience with targeted questions 1

1

. Her work involves writing prompts for AI models to perform specific social work tasks based on fictitious case files, then evaluating the AI's responses. In San Francisco, pediatric hospitalist Melania Poonacha logs 10 additional hours weekly asking AI to interpret lab results. In Baton Rouge, novelist Robin Palmer Blanche evaluates AI-generated creative writing after struggling to make a living since the 2023 Hollywood writers strike 1

1

. These professionals are methodically translating their expertise into training data that could eventually replace the white-collar workforce they represent.

Economists Warn of AI Job Displacement and Economic Risks

Alex Imas, a University of Chicago economist tracking AI automation and its labor market impact, initially found the prospect terrifying 2

2

. His research, which appears in top journals and his widely-read Substack "Ghosts of Electricity," explores doomsday scenarios where AI job displacement leads to mass unemployment, collapsing demand, and economic contraction. Morgan Stanley recently recommended investors follow Imas as a primary resource on AI and the economy 2

2

. The Federal Reserve Bank of New York found that 2025 ended with the highest unemployment rate for recent college graduates in years, while Goldman Sachs estimates about 7 percent of workers will be displaced by AI 4

4

.

OpenAI predicts 18 percent of jobs will soon be automated, while Anthropic's Dario Amodei told Axios that AI could "wipe out nearly half of all entry-level white-collar jobs" 4

4

. An MIT report suggested AI will achieve 80 percent success rates on most tasks by 2029 4

4

. These projections have shaken the longstanding economic orthodoxy that technology always creates more jobs than it destroys. Larry Summers and other prominent economists are now questioning whether this time might be different 3

3

.

The Future of Work: Scarcity, Human Skills, and Structural Change

After months of analysis, Imas found a reason for cautious optimism in an unexpected place: Starbucks 2

2

. The $112 billion company recently reversed years of automation under CEO Brian Niccol, bringing back handwritten notes on cups, ceramic mugs, and comfortable seating. More baristas are being hired because customers value human interaction over efficiency. Imas argues this illustrates structural change theory—as AI makes commodity production cheaper, human presence, social connection, and provenance become scarcer and therefore more economically valuable 2

2

.

The analogy to agriculture offers perspective: around 1900, 40 percent of American workers farmed; today it's under 2 percent. People didn't stop eating—they reallocated labor toward manufacturing and services as productivity rose 2

2

. Whether the same transformation can absorb workers displaced by AI remains uncertain. IBM's Watson demonstrated in 2014 that it could outperform humans at sophisticated cognitive tasks, and today's version is 240 percent faster 3

3

. Google's autonomous vehicles have driven hundreds of thousands of miles with minimal accidents. Computers now excel at screening legal documents and detecting human emotion 3

3

.

Congress Remains Silent as Political Pressure Builds

Despite the mounting evidence, policy makers have offered few concrete proposals to address the impact on the workforce. OpenAI floated a "New Deal" for workers including a 32-hour workweek and public wealth fund, while Anthropic suggested the problem might require an entirely new tax code with duties levied against AI companies 4

4

. Yet Congress has been largely silent even as 71 percent of workers in one poll say they fear displacement from this technology 4

4

.

Pro-AI PACs like Leading the Future are already pouring millions into midterm elections 4

4

. Political observers suggest the 2028 election will likely center on AI's societal shift, similar to how COVID shaped 2020. Some Democrats privately acknowledge that proposing sweeping reforms while Republicans control government seems futile, while others worry about massive campaign spending against them. There's also tension around global competition—China and other countries are advancing AI regardless of U.S. policy 4

4

. For now, professionals like Kozak continue their dual existence: helping people by day, training their digital replacements by night.