Courtney Fahnhorst was looking for a way to supplement her income when she came across a LinkedIn ad seeking medical experts to help train AI models.
The mother of four was already working full time as a wound care and hyperbaric medicine specialist, but she wanted to earn some additional income to dedicate to a private practice she plans to open in St. Johns County, Florida, later this year.
“I really needed a job that was flexible but also allowed me to generate some income and capital to invest in my future,” he says. “This fits perfectly.”
For the past two years, the former emergency room doctor and wound care specialist has spent as much time as she can sharing her 15 years of medical experience with artificial intelligence software. “You’re posing a clinical situation and I’m evaluating how well the model responds to it: where it goes wrong, what misconceptions it may generate,” he says.
Although Fahnhorst was initially interested in the opportunity to be able to set her own hours, she says she has gotten a lot more out of the temporary contract position than just a paycheck.
“The salary is competitive with what I would earn in a clinical or hospital setting, but it gives me flexibility. It has also really given me back that passion for medicine that I think many of us miss,” he says.
“Getting back to the core reason behind what we do and removing the bureaucratic things that tie us down in medicine as doctors and providers – it’s really been a really fun aspect that I wasn’t expecting.”
Fahnhorst adds that training AI models for companies like his current employer, Mercor, also democratizes access to medical information.
“My patients come to Dr. Google and have done so for as long as I have been in medicine,” he says. “I think it’s a great service to teach these models correct information.”
Experts wanted
Fahnhorst is just one of 30,000 “experts” who regularly impart their professional knowledge to Mercor, which pays an average of $80 per hour and about $3 million per day in total.
Mercor itself is just one of a growing number of companies trading professional knowledge for cash, along with Alignerr, Data Annotation, Outlier, Mindrift, RemoExperts, RWS, CrowdGen and Handshake, to name a few.
“You can imagine how people would use ChatGPT or Gemini, and what they expect in terms of response,” says Heidi Hagberg, head of communications at Mercor. “For that response to be accurate and relevant, they need people with that expertise to respond.”
Mercor does not publicly name its clients. But Hagberg says the company works with “all the cutting-edge AI companies,” as well as a growing number of hypergrowth startups and Fortune 500 companies.
“We see this continuing to grow in our frontier laboratories, but spanning many different sectors of the economy,” he says. “If you are a company that wants to create a custom agent to help respond to customer service tickets, or help create market research agents, we can use our experts in our network to create those types of AI tools.”
Hagberg says that so-called “experts” typically register on the platform so they can earn flexible additional income, have the opportunity to gain AI skills and work experience, and have a direct stake in improving the software’s results.
“They work with us because they want to differentiate themselves and learn to work together with AI,” says Hagberg. “This is directly relevant to what they know and what they’re interested in. So being able to find work for $80 an hour that they can do on their own time is incredibly attractive.”
An old job on a new level
The task of teaching technology manually is not new. But the size, scale and speed of the industry have exploded in recent years along with the growth of artificial intelligence platforms.
“Data labeling in general has been around for over a decade, because it actually started with drawing boxes around stop signs for autonomous vehicles and with very basic image recognition,” Hagberg says. “As AI has evolved into the chatbots available to consumers today, it has expanded what it needs to be able to do. Millions of people are using them now, so you need to have a wide variety of responses.” [to AI prompts].”
Now, the technology is evolving to require even more advanced domain expertise, as AI agents look to complete multi-step processes using multiple platforms and services.
“We basically create mock pieces of software, like Gmail mocks, Workday mocks, or Excel mocks, and then you have people doing multi-step tasks that use those tools together,” says Handshake President Jonathan Stull.
“Let’s say you want to train a [AI] agent to help an accountant close his books. Not only do they do that in QuickBooks, but they pull invoices from Gmail, then they look at something in Slack, and then they bring it back into QuickBooks. “If you want to train an agent to do a multi-step, multi-tool job, you actually have to track all of those steps.”
The job of these AI trainers is to follow those steps to ultimately teach the AI agents to complete them themselves.
An AI employment boom?
Founded in 2014 as a platform to connect students and graduates with employers, Handshake added its AI division about 18 months ago in response to the growing demand it saw from employers on its recruiting platform, Stull says. Since then, the company has paid approximately $300 million to about 100,000 AI trainers, which it calls “fellows.”
“It’s a direct response to demand,” Stull says. He explains that the rising valuations of AI companies like OpenAI and Anthropic demonstrate the level of competitiveness in the industry, adding that models only advance when given more computing power and direct human knowledge.
As a professional development platform that connects students and recent graduates with job opportunities, Handshake takes a slightly different approach to its AI training workforce, according to Stull, often framing temporary contract roles in the context of its members’ career paths.
“Employers now want to hire more people with AI skills and fluency,” he says. “So if you can say, ‘I’m a PhD,’ or ‘I’m an undergraduate, and not only do I know economics, or music theory, or software development, but I’ve worked with a lot of foundational AI labs building and training their data,’ that’s a huge advantage.”
Despite the industry’s recent rapid growth, insiders believe there is still a long way to go before the need for AI trainers begins to decline. If he ever does.
“We believe, and I think most people in our space believe, that this industry will double next year,” Stull says. “And then double again the following year.”