Dr. Alice Chiao used to teach emergency medicine to students at Stanford University’s medical school. Now, she’s teaching artificial intelligence chatbots to think, diagnose, and prescribe like her. She’s part of a booming new economy where professional experts train AI through a process called reinforcement learning — essentially grading AI’s responses and teaching models to improve through trial and error. It’s estimated to be worth at least $17 billion, and experts can earn up to hundreds of dollars per hour teaching AI to do their own jobs.
Chiao is one of tens of thousands of experts working with Mercor, one of the companies managing reinforcement learning for major AI companies. Mercor has contracts with experts in subjects ranging from medicine, law, and finance to comedy, sports, and even wine. The company is valued at more than $10 billion and pays out over $1 million per day across thousands of experts. Companies like OpenAI, Google, and Anthropic use what Mercor’s CEO described as “large armies of people” to train their AI models. Without this human feedback, the technology doesn’t work.
How You Teach AI to Think Like You
When Chiao trains AI models, she uses real scenarios from her decades as a doctor in both primary and emergency medicine. A patient might ask whether their child should see a doctor when experiencing a cough or fever. But the system also needs to know how to respond when presented with medical jargon — what a physician might see on an intake form. The AI model sometimes provides answers Chiao wouldn’t have thought of herself. But other times, she sees responses that are misleading, alarmist, or unsafe. That’s where she intervenes.
Experts grade a model’s response using a rubric they’ve created after consulting with a team of other experts in their field. Those responses are fed back into the model, which is trained to aim for good grades. It’s trial-and-error learning, except the student is an AI system and the teacher is billing hundreds of dollars per hour. The most popular experts Mercor hires for are in software engineering, followed by finance, medicine, and law. Job posts can range from journalists to mechanics to comedians.
The 22-Year-Old Billionaire Paying You to Train Your Replacement
Brendan Foody co-founded Mercor three years ago at age 19 with friends Adarsh Hiremath and Surya Midha. The company started as a recruiting platform. When they shifted focus to AI, their rolodex of resumes was the perfect starting point for finding the experts AI companies were seeking. Foody said Mercor has grown from $1 million in revenue run rate to over $500 million in less than two years. At 22, Foody and his co-founders are likely some of the youngest tech founders to make the Forbes billionaire list since Mark Zuckerberg, who made the list at age 23.
Mercor is not alone. Last year, Meta made a $14 billion investment in Scale AI, which operates in a similar space. Other competitors like Surge AI, Handshake, and Micro1 have helped mint a new class of young, ultra-wealthy tech founders. The high valuations show investors think services like human feedback and expert testing of AI models are becoming a permanent and essential part of how AI systems are built and improved. This isn’t a temporary gig economy side hustle — it’s infrastructure.
The Uncomfortable Irony
The job posting is straightforward: “Teach AI how to do your job. Get paid well to do it.” Some people see this as training their replacement. Stable full-time careers in medicine, law, and finance get converted into gig work where professionals grade AI responses for hourly rates. The technology improves. The experts get paid. Eventually, the AI doesn’t need them anymore. That’s the concern driving anxiety about AI disruption across industries.
But Chiao doesn’t see it that way. She views her work as ensuring AI models are safe and capable enough to help doctors spend more time with patients and less time filling out forms. She sees AI as eventually assisting doctors with reading scans, filling out charts, and taking notes. “Physicians were selected because we really want to help people. We want to heal. We want to spend time talking to people — listening, engaging,” Chiao told CNN. “I don’t want to see it as AI taking over our jobs. I want to see it as AI taking over the aspects of our jobs that prevent us from being good doctors, good healers and good listeners.”
Some Things Can’t Be Taught
Not everything can be taught through this process. The more subjective the task, the more difficult it is for AI to master. Mercor tried to train one AI model to be funnier by hiring comedians from the Harvard Lampoon. They wrote rubrics to improve models and how funny they are. The problem is obvious to humans but not to machines: people have different opinions on what’s funny. What you actually need is localization of how humor varies by geography and culture. Even after thousands of hours of expert training, comedy remains stubbornly human.
Chiao said patients should use today’s AI model tools as a starting point before talking to a doctor. The technology is not a replacement for a doctor like herself with 20 years in the field. “There is a gut feeling that comes with experience, that comes with sitting with a patient, looking them in the eye, and seeing something that is beyond their history, their lab values, the words that are coming out of their mouth,” Chiao said. “So, this is where it’s really important to know that the AI is not a doctor, it’s not a human being.”
The Future of Work Looks Like This
For now, professionals who understand their fields deeply enough to grade AI responses can earn significant money doing it. The work is flexible, remote, and pays well. But it’s also the definition of training your replacement. Every response you grade makes the AI better at mimicking your expertise. Every rubric you create helps the system learn to think more like you. At some point, the AI gets good enough that it doesn’t need you anymore. That’s the entire point of the technology.
Foody sees it differently. He believes making everyone more productive through AI is how we solve bigger problems. “We need to cure cancer. We need to solve climate change,” he said. “And making everyone 10 times more productive so that they’re able to better work on those key problems is going to be a huge, huge benefit to how we make progress as a society.” Whether that vision includes the experts currently training AI to replace them remains unclear. What is clear: the AI learning industry is worth billions, it’s making young founders extraordinarily wealthy, and it’s paying professionals hundreds per hour to teach machines how to think, diagnose, and make decisions exactly like they do.