Doctors and Lawyers Are Joining AI's $400-an-Hour Gig Economy

Two women open their laptops. Both American. Both moms. Both logging on for shifts as data workers, the human labor that trains artificial intelligence. Brook Hansen started doing data work seriously after she was let go from her customer service job when she was six months pregnant. As a stay-at-home mom, data work offered a way to supplement her husband’s income and make a couple of hundred dollars a week.

Anu Brixey, a radiologist, makes that same amount in an hour.

Long associated with low wages and considered the digital equivalent of blue-collar labor, data work is now attracting white-collar specialists who can command much higher sums. LinkedIn named data annotators as the fourth-fastest-growing job role in the US this year. As AI models improve, recruiters are searching for software engineers (pay: up to $120 per hour), condensed matter experts ($170 per hour) and civil rights attorneys ($400 per hour), all to “help train next-generation AI systems.” Known in the industry as “experts,” these workers share the data economy with lower-paid generalists performing less specialized tasks.

The growing divide has implications for both groups. Lower-paid generalists complain that conditions are deteriorating: Their tasks are becoming more difficult while steady work is harder to find. Meanwhile, the influx of white-collar specialists is causing concerns that AI models are absorbing the expertise of doctors, mathematicians, accountants and scientists.

Labor — whether manual or intellectual — has long been understood as something employers rent, says Danielle Li, a professor at the Massachusetts Institute of Technology who studies AI and the labor market. If a company stops paying, it loses access. That dynamic breaks down, she says, when an employer can own a version of that labor in the form of an AI model.

When Hansen started data work a decade ago, she was mostly transcribing audio or drawing boxes around people or cars in photos on Amazon’s Mechanical Turk platform. (Jeff Bezos once called it “artificial artificial intelligence.”) These microtasks demanded little concentration and paid in cents. The low wages were a source of stress, says Hansen, an organizer for Turkopticon, which fights to improve conditions for data workers. But she could dip in and out whenever the kids didn’t need her. With the rise of large language models in the 2020s, her assignments grew more complex: fact-checking documents or comparing chatbot answers, tasks that could take an hour and that demanded deep focus.

Now the platforms often want data workers who specialize in subjects like math or biology, Hansen says. Occasionally, she gets accepted for jobs linked to her degree in environmental conservation. But most days she flits between platforms, searching for work. Mechanical Turk, once the most important platform for American data workers like Hansen, shut down last month after 21 years. Meanwhile, newer platforms are thriving. Three-year-old Mercor, which recruits white-collar data workers, has been discussing new funding at a valuation of $20 billion.

The growing demand for specialized data workers reflects a broader shift in AI training. While publicly available information scraped from the internet formed the foundation of today’s AI models, companies are now racing to secure bespoke, privately held data to distinguish future iterations from their rivals’. In a recent blog post, former OpenAI engineer Will DePue argued that data is just as important for the industry as computing power. Intermediary companies are soliciting not just the skills of professionals like radiologists but also paying for private phone chat logs between friends and first-person videos of ordinary people doing chores. Companies like SpaceXAI have also hired in-house “AI tutors” specializing in subjects as varied as finance, Finnish and comedy.

As AI companies demand more specialist knowledge, recruiters have been sleuthing LinkedIn to find people with those skills. Four years ago, Brixey received a message from one of these companies seeking radiologists who could interpret chest, lung and heart scans for an AI company. By then, she’d encountered the technology at a medical center where she worked: A model designed to catch lung nodules was prone to false positives. Still, Brixey believed AI could help radiologists deal with an ever-increasing demand for scans.

That’s why, after a long week of work, she sometimes spends her weekends labeling data, highlighting areas where she can see calcium in a heart scan, for example. She signs up for at most two projects a year motivated less by the pay than by the technology’s potential. “I’m doing it for improved patient care,” she says.

Researchers who have studied data work’s evolution are skeptical of the scale of the shift to white-collar specialists. Antonio Casilli, a professor at the Polytechnic Institute of Paris, maintains that highly paid experts participate in relatively few projects compared with lower-paid generalists. Companies promising investors superintelligent machines prefer to appear to be recruiting superintelligent people, he says, even as they remain reliant on the same workforce as before. These experts “are just the cherry on top.” Most data work, he says, is still done by low-wage workers who face precarious conditions and sudden account deactivations.

Average pay at Mercor is now $110 per hour, though the company said it did not have data on what percentage of tasks pay that rate. British data-work platform Prolific also declined to say how many of its tasks are completed by its “experts,” who make up 30% of the platform’s total participants.

Fan Yang, a research fellow at Melbourne Law School who has been interviewing data workers in Australia, Canada, China and Japan, says the rise of data work is creating a new form of outsourcing. Rather than hiring people directly, businesses can buy AI tools that still rely on humans — now in the form of data workers on more ad hoc contracts — to fine-tune them. AI isn’t replacing humans so much as repositioning them in the labor market, Yang says.

But MIT’s Li thinks the outsourcing comparison understates how dramatic the changes could be. A small group of data workers could in theory replace or substantially devalue an entire profession, she says. “When you are supplying data, other people’s choices are very much impacted by yours.”

For some professionals, that possibility is already a source of anxiety. “People are worried,” says Pauley Chea, another radiologist who has been moonlighting as a data worker. He says the uncertainty is particularly pronounced among medical students and residents who still have years of training ahead.

Some professions are easier to recruit from than others. Moonlighting may be common among clinicians, but investment bankers are often bound by employment contracts and have little spare time, as users of the forum Wall Street Oasis pointed out in response to recruiters. To attract busy professionals, data-work platforms are offering increasingly generous pay, according to Ana Price, a vice president at Prolific. “Given the need for white collar — these are typically people who I would categorize as time-poor but more money-rich,” she says. “So the compensation needs to be compelling.”

Some data companies envision a future in which they are no longer so reliant on recruiting professionals. Instead, they could buy data wholesale from employers. In such a future, almost everyone working with a computer could be considered a data worker, whether they like it or not.

Ali Ansari, the founder and chief executive officer of Micro1, an AI data startup valued at $4 billion, argues that companies should think of their employees as not only producing goods and services but also generating data with its own market value. “Every hour of work can simultaneously: Run the organization, train AI models, generate additional revenue for the organization,” he wrote in January.

Some companies are already turning their employees’ daily tasks into training data. In April, Meta started installing software to track its employees’ clicks, keystrokes and mouse movements. “This is where all Meta employees can help our models get better simply by doing their daily work,” said a memo about the change.

Hansen, who has spent a decade watching data work evolve, isn’t convinced AI will ever stop needing people like her. Recently, most of her assignments have involved spending up to 20 minutes in a Zoom room talking to someone she’s never met. She assumes the conversations are being used to train AI to speak more naturally, less robotically.

Without us, Hansen says, the models can never be perfect.

To contact the author of this story: Morgan Meaker in London at mmeaker1@bloomberg.net

© 2026 Bloomberg L.P. All rights reserved. Used with permission.

Bloomberg Law provides trusted coverage of current events enhanced with legal analysis.

Log in to keep reading or access research tools and resources.