Doctors and lawyers join AI's $400-an-hour training gig economy
Highly paid professionals are moonlighting as AI data labelers, teaching models to replicate their expertise for up to $400 an hour.
Business · Source: Bloomberg Law
What happened
The AI training gig used to mean drawing boxes around cars for pennies. Now it means doctors, lawyers, and engineers reviewing AI outputs for hundreds of dollars an hour. Civil rights attorneys command up to $400 an hour. Condensed matter experts earn $170. At the very top of the market, some specialists make up to $1,000 an hour. LinkedIn just named data annotator as the fourth-fastest-growing job in the US.
Platforms like Mercor sit at the center of this shift. The three-year-old startup manages 30,000 active contractors and pays out up to $4 million daily. They are currently discussing a new funding round at a reported $20 billion valuation. Meanwhile, Amazon Mechanical Turk shut down last month after 21 years. The original home of cheap data labor is dead. The new era belongs to credentialed consultants.
AI labs need this deep domain knowledge to push models past basic reasoning. Instead of just tagging photos, these experts draft contract clauses. They evaluate complex medical diagnoses. They check engineering calculations. Companies are racing to secure this bespoke, privately held data to beat their rivals. Even companies like SpaceXAI are hiring in-house AI tutors for subjects ranging from finance to comedy.
Key facts
- $400 — Hourly rate for civil rights attorneys training AI
- $1,000 — Peak hourly rate for top-tier AI specialists
- $4 million — Daily payout to contractors on the Mercor platform
- $20 billion — Reported valuation Mercor is discussing for new funding
- 4th — Rank of data annotator among fastest-growing US jobs on LinkedIn
Why it matters
Training costs are fundamentally changing for anyone building AI. A general chatbot was cheap to train using scraped internet data and low-wage annotators. A specialized medical or legal model requires constant review from credentialed experts. This drives up the labor bill exponentially. As former OpenAI engineer Will DePue noted, data is now just as critical and expensive as computing power. You cannot build a frontier model without buying human expertise by the hour.
This creates a bizarre labor dynamic. Traditionally, companies rent intellectual labor. If they stop paying, they lose access. Now, they are paying experts to permanently transfer their expertise into an AI model. The employer eventually owns a digital version of that labor. These professionals are cashing in on a lucrative gig today. But they are actively building the systems that will commoditize their own professions tomorrow.
For builders
Budget for expensive expert data
If you build vertical AI, scraped data is no longer enough. You must pay premium hourly rates for domain experts to validate your model outputs. Your labor costs will scale directly with the complexity of your niche. The AI labs pay the bill.
Build specialized annotation tools
Generalist platforms are dying while specialized networks thrive. There is a massive opportunity to build annotation tools and marketplaces specifically designed for high-end professionals. Founders who connect AI labs with niche experts will capture massive value.
Exploit the two-class labor market
The gap between $10-an-hour generalists and $400-an-hour experts is widening. Founders must optimize costs by routing basic tasks to cheap commodity labor. Reserve your expensive credentialed experts strictly for complex reasoning evaluation.
My take
I find this to be the ultimate capitalist paradox. Doctors and lawyers are happily taking $400 an hour to dig their own professional graves. They think they are just doing a lucrative side hustle. In reality, they are downloading their brains into a machine that will eventually replace them.
Original reporting: Bloomberg Law. This is my rewrite and opinion.