OTWopentechwire
Tech Intelligence, Openly Wired
AI

The Gig Economy Reaches Cardiologists: How AI Platforms Extract Value From High-Skilled Labour

Data work for machine learning models now spans physicians and PhD holders, yet the same pattern of declining pay and vanishing tasks persists across continents and credentials.

HP
Hana Park
Semiconductors Reporter · Seoul
Nov 22, 2026
5 min read
The Gig Economy Reaches Cardiologists: How AI Platforms Extract Value From High-Skilled Labour
Credit: James Manning / Getty Images

From Venezuela to Kansas: The Common Thread

A computer science teacher in Venezuela earned $2 on his first day labelling images after his 13-year-old son discovered the work. Meanwhile, recruiting platform Outlier advertises rates of up to $180 per hour for US-based cardiologists designing cardiac scenarios and ranking AI model outputs. The same platform offers $7.50 per hour for identical work performed in Tagalog in the Philippines, and $12.50 per hour for Swahili speakers in Tanzania.

The gap in advertised rates is vast, but six years of field interviews with data workers reveal a shared experience: platforms build workforces with attractive pay, then cut rates and relocate operations. What began as outsourced data labelling for the cheapest available labour has expanded to physicians, lawyers, and doctoral graduates annotating training data, verifying outputs, and evaluating model responses. Higher credentials lift the initial offer, yet platforms retain unilateral control over payment terms and working conditions.

A Family's Arc Through the Task Economy

Rogelio, the Venezuelan teacher, joined his son on platforms including Hive and Scale AI's Remotasks after the country's economic collapse eliminated conventional employment. The boy worked on a laptop the Hugo Chávez government had distributed to schoolchildren years earlier, earning 18 cents for several hours of image annotation.

One task, internally labelled Beast Mode, required workers to identify objects in domestic settings. In December 2020, with schools closed and the entire household working, the family earned $20 in a single day. It remains their highest total. By early 2021, Beast Mode had vanished from the platform, and their weekly earnings dropped below Hive's $5 payout threshold, leaving accumulated work unpaid.

Remotasks terminated its hourly programme in Venezuela that year and launched operations in Kenya. In March 2024, the platform discontinued work in Kenya, Nigeria, and Pakistan. Scale AI cited "enhanced security protocols" and stated that three weeks' notice had been issued, though it later acknowledged that Kenyan workers received no notification "due to operational errors". Workers reported being locked out of accounts without warning.

Extractivism Beyond Natural Resources

The cycle has historical precedent. Extractivism traditionally describes systems that remove resources from regions on terms set by buyers rather than sellers, leaving little local accumulation. Venezuela's crisis stemmed partly from an oil economy that built limited alternatives; the pension one former state oil company employee received by 2021 was worth roughly $1 per month, insufficient for six eggs. She turned to Hive for income, though that proved even less stable than the collapsed pension system.

The distinction between resource extraction and platform labour lies in mobility. An oil field cannot relocate; a platform can redirect tasks to workers in another country within hours.

AI development is frequently framed as extraction of existing data: books, artwork, and other materials used without negotiation to train large language models. A second layer operates in parallel. The examples, evaluations, and corrections required to refine those models must be created by people, for fees set unilaterally by commissioning platforms. Whether the rate is $2 or $180, the structure remains consistent.

Terms That Travel Across Platforms

Outlier's worker agreement permits the company to delay or suspend payment during investigations of submitted tasks, and to withhold "some or all amounts accrued" where it identifies a violation "in the company's sole determination". Accounts may be deactivated "immediately without prior notice for any reason". The identical phrasing appeared in Remotasks terms governing work in Venezuela in 2021.

Reporting earlier this year documented laid-off lawyers, scientists, and screenwriters now populating these platforms. Of more than 30 workers interviewed across professions, nearly all described escalating demands, shrinking target completion times, and falling pay as projects progressed. One designer tagging videos for Meta at $21 per hour saw the project end for "scope changes". Days later, an identical assignment reappeared under a new project name at $16 per hour, replicating the sequence Rogelio's family experienced five years prior.

Who Stays When Rates Fall

Workers depart these platforms regularly. Those who remain are often constrained. One woman interviewed was confined to her home during the Covid-19 pandemic by a health condition that made other employment impossible. "This is my only source of income," she explained. The combination of sudden rate changes, disappearing tasks, and zero input into terms creates a population of workers with limited alternatives.

Industry awareness of these dynamics is not in question. At a 2024 meeting convened by the nonprofit Partnership on AI and the Fairwork project, which evaluates platform working conditions, representatives of AI companies and the outsourcing firms supplying them gathered. AI companies stated they did not employ data workers directly; outsourcing firms replied that their pricing was dictated by AI company expectations. Consensus emerged that pay should improve, but discussion halted when worker representation was raised. No workers were present.

Legal Challenges Across Jurisdictions

Workers have pursued redress through courts. In March 2023, a Kenyan judge temporarily blocked the dismissal of 260 Facebook content moderators, prevented Meta from transferring their work to another contractor, and barred Meta and contractor Majorel from blacklisting the workers. As litigation proceeded in Kenya, Meta relocated content moderation operations to West Africa.

American courts are handling related cases. Claims by workers on Scale AI's platforms led to a proposed settlement that received preliminary approval in San Francisco; the company denies wrongdoing. A separate class action alleges Mercor misclassified more than 30,000 lawyers, doctors, and engineers while controlling their pay and reviewing their output. Mercor's clients have included OpenAI, Anthropic, and Meta, though none are named as defendants. The case continues.

Unequal Starting Points, Identical Constraints

A cardiologist offered a three-digit hourly rate occupies a different economic position than a Venezuelan teacher earning $2 for a day's labour. What links them is structural: rates shift without consultation, tasks vanish without explanation, and neither party influences the terms. Credentials may alter the initial number, but not the underlying arrangement.

The work will continue migrating. Rules will remain set by the companies commissioning the tasks, enforced through terms that travel unchanged across platforms and borders, applied wherever the next worker logs in, whether in Nairobi, Caracas, Manila, or Dallas.

Read next
AI

Japan's Micronics Eyes Second Plant as Memory Chip Testing Demand Accelerates

Kenji Watanabe · 6 min
AI

OpenAI Grants Outside Teams Access to Test Models for Scheming Behaviour

Arjun S. Mehta · 4 min
AI

Digital Realty Plans $10 Billion Japan Expansion as AI Compute Demand Surges

Kenji Watanabe · 5 min
Spot something wrong? Email corrections@opentechwire.com. We log every correction publicly.