The modern gig economy has officially expanded into the highest ranks of professional medicine, bringing its precarious contracts, sudden rate cuts, and total lack of worker representation to the doorsteps of U.S. cardiologists.
Recruiting platforms operated by prominent artificial intelligence firms are now advertising rates of up to $180 an hour for U.S.-based physicians to design intricate cardiac scenarios, review machine learning-generated medical reports, and meticulously rank model outputs. Yet, the same jobs performed elsewhere reveal deep global disparities. On the exact same platforms, Tagalog-language data work based in the Philippines commands up to $7.50 an hour, while Swahili-language tasks in Tanzania max out at $12.50 an hour.
This tiered compensation model highlights a sprawling, borderless labor market that has quietly evolved over the past several years. Companies like Scale AI initially outsourced low-level data tasks—such as labeling everyday images or transcribing audio clips—to regions where labor was cheapest. Today, however, these platforms are casting a much wider net, recruiting high-earning professionals, including lawyers, scientists, Ph.D.s, and medical specialists, to generate, annotate, and verify complex domain-specific data. While advanced credentials and specialized expertise can command higher advertised rates, the underlying platform infrastructure remains entirely unchanged: the company maintains absolute control over compensation, task availability, and day-to-day working conditions.
This dynamic is part of a broader cycle of labor extraction that journalists and researchers have tracked for years. Observers who have spent extensive time interviewing the workforce behind the AI boom note a recurring trajectory: platforms typically begin by offering lucrative rates to rapidly build a robust, highly educated global workforce. Once the labor pool is established and dependence on the platform grows, the companies systematically cut pay, alter project scopes, and eventually move on to the next geographical market.
A striking example of this cycle played out in Venezuela, where a devastating economic collapse left a large, highly educated population stranded online and desperate for income. Following the collapse of the local economy and the disruptions of the pandemic, a former computer science teacher named Rogelio found his family without sustainable work. His young son discovered that they could earn meager amounts of money by labeling images of cars and roads on crowd-work platforms like Hive and Scale AI’s Remotasks. Using an aging laptop that the government had distributed to schoolchildren a decade prior, the boy earned a fraction of a dollar for hours of image tagging. Rogelio joined him, netting a little over two dollars on his very first day.
For a brief period, the work offered a vital lifeline. A specialized project dubbed "Beast Mode," which required workers to meticulously label household objects inside interior photographs, paid significantly more than any previous task the family had encountered. By December 2020, with local schools shuttered and the entire household pitching in, the project brought in twenty dollars in a single day—the highest single-day earning they would ever see.
By early 2021, however, "Beast Mode" vanished without explanation, and the family’s cumulative earnings plummeted below Hive’s five-dollar weekly payout threshold, effectively locking them out of their accumulated wages.
This operational pattern mirrors how many artificial intelligence and data-labeling platforms consistently function. Remotasks abruptly ended its hourly compensation program in Venezuela in 2021 and simultaneously launched operations in Kenya that same year. By March 2024, the platform abruptly discontinued its operations across Kenya, Nigeria, and Pakistan. Scale AI publicly attributed these sudden closures to "enhanced security protocols," claiming that advance notices had been dispatched weeks ahead of time. However, the company later acknowledged that Kenyan workers never received those notices due to internal operational errors. Most workers who spoke with reporters at the time reported that the platform simply locked them out of their accounts with zero warning or recourse.
Scholars and labor advocates point out that this recurring phenomenon has a precise name, one that predates the modern artificial intelligence industry: extractivism. It describes an economic system that extracts resources, value, and human labor from specific communities for accumulation elsewhere, operating strictly on terms dictated by the buyer rather than the worker. The term applies equally to what is taken and what is left behind in local economies. For instance, the economic crisis that forced Rogelio’s family into digital gig work was itself the residue of an oil-dependent national economy that failed to develop sustainable alternative industries.
One Venezuelan retiree interviewed by researchers drew a pension from the state oil company that, by 2021, had dwindled in value to roughly one dollar a month—an amount insufficient to purchase even half a dozen eggs. She turned to labeling data on Hive to bridge the impossible gap, only to find that digital platform income was even more precarious than her state pension. As labor analysts frequently note, an oil field remains physically stationary, whereas an artificial intelligence platform can instantly transmit its next batch of tasks to a completely different workforce thousands of miles away at the click of a button.
Public discourse often frames artificial intelligence as a technology built primarily on the extraction of existing cultural and intellectual assets, such as published books, journalism, and artistic works used to train foundational models. Yet, this extraction has a secondary, equally vital layer: the continuous stream of human-generated evaluations, corrections, and novel examples commissioned specifically to refine and steer those models. This material must be created by human hands and minds for a negotiated fee—whether that fee is two dollars for a struggling teacher in Venezuela or up to $180 an hour for a cardiologist in the United States.
Despite the vast differences in compensation, the contractual realities governing these workers are remarkably uniform. Outlier’s terms of service, for example, explicitly grant the company the right to delay or suspend payments while investigating a worker’s submitted tasks. The company reserves the right to withhold "some or all amounts accrued" if it determines, in its "sole determination," that a violation has occurred. Furthermore, the platform maintains the legal authority to deactivate any user account "immediately without prior notice for any reason." This exact legal language appeared word for word in the Remotasks terms of service under which Rogelio’s family labored years earlier.
The expansion of these platforms into white-collar professions has been widely documented. Laid-off corporate lawyers, research scientists, and screenwriters have increasingly populated these systems to make ends meet. Across interviews with dozens of white-collar workers, a familiar arc emerges: initial high expectations give way to increasing demands, shrinking target completion times, and falling pay rates as a project matures. One digital designer tagging video footage for Meta was earning $21 an hour when the project was abruptly canceled due to sudden "scope changes." Just days later, the exact same job reappeared under a new project title at a rate of $16 an hour—the very same cycle experienced by blue-collar and clerical data workers years prior.
While workers retain the theoretical freedom to leave these platforms, many stay because they have no viable alternatives. One Venezuelan worker interviewed by researchers was confined to her home during the coronavirus pandemic due to an underlying health condition, rendering her unable to risk losing her digital income. "This is my only source of income," she explained.
These labor conditions are well understood within the technology sector itself. During industry roundtables convened by nonprofits like the Partnership on AI and the Fairwork project—an organization that evaluates platform labor practices—representatives of major artificial intelligence firms and third-party outsourcing agencies routinely discuss these inequities. Artificial intelligence companies typically maintain that they do not directly employ the data workers, while outsourcing suppliers argue that their operational margins are entirely dictated by the tight budgets and prices expected by the tech firms. While participants generally agree that compensation should be higher, discussions concerning formal worker representation often stall, largely because the workers themselves are never invited to the table.
Despite these systemic imbalances, workers across the globe have increasingly taken collective action. In March 2023, a Kenyan court temporarily halted the mass dismissal of 260 content moderators working for a Facebook contractor, barred the company from shifting their work to a different service provider, and prohibited the firms from blacklisting the affected workers. Even as those workers fought for their rights in Kenya, however, operations were subsequently shifted by contractors to other regions in West Africa.
Legal battles are also mounting within the United States. Claims brought forward by workers on Scale AI platforms resulted in a proposed class-action settlement that received preliminary approval in a San Francisco court, though the company has consistently denied any wrongdoing. In a separate legal challenge, a class-action lawsuit alleges that the platform Mercor misclassified more than 30,000 lawyers, doctors, and engineers while unilaterally setting their compensation and evaluating their professional output. While Mercor has provided expert labor to major artificial intelligence developers including OpenAI, Anthropic, and Meta, those prominent tech firms are not named as defendants in the ongoing litigation.
A cardiologist offered a three-digit hourly rate occupies a vastly different economic reality than Rogelio and his family in Venezuela. Yet, what unites them across professions and geographies is a fundamental vulnerability: the piece rate can change without warning, the tasks can disappear overnight, and neither group holds any institutional say over the conditions of their labor.
As the industry evolves, the work will continue to move across borders, and the rules will continue to be written entirely by the companies that commission the algorithms—regardless of where the next worker logs on, whether in Kenya, Venezuela, the Philippines, or the United States.

