N POSADA’S ESSAY HERE>
The gig economy has expanded to include healthcare professionals. Outlier, a recruiting platform owned by Scale AI, advertises hourly rates of up to $180 for U.S. cardiologists to create intricate cardiac scenarios, evaluate AI-generated reports, and assess model outputs. For Tagalog work in the Philippines, the platform offers rates of up to $7.50 an hour, and up to $12.50 an hour for Swahili work in Tanzania.
Initially, companies like Scale AI outsourced data work to locations with the lowest labor costs. Now, they recruit physicians, lawyers, and Ph.D.s to generate and annotate data, rank model answers, and verify outputs. Higher education levels can lead to higher advertised rates, but the platform still manages payment and working conditions.
Writer Rest of World spent six years speaking with people who do this work. Many were in Venezuela, where economic collapse has left a well-educated population without options. Rogelio, a former computer science teacher, joined his 13-year-old son on labeling tasks on platforms like Hive and Scale AI’s Remotasks, making just over $2 on his first day.
Their earnings peaked in December 2020, when a task paid more than any other work they had done. By early 2021, their income fell below the platform’s $5 weekly payout threshold. This is the typical pattern: platforms begin by paying high rates to build a workforce, then reduce rates and move on. Remotasks ended its hourly program in Venezuela in 2021 and suspended operations in Kenya, Nigeria, and Pakistan the same year, citing "enhanced security protocols." In reality, Kenyan workers never received these notices due to operational errors.
This system is known as extractivism, where resources and value are taken from people and places for profit elsewhere. The crisis that put Rogelio's family online stemmed from Venezuela's oil economy, which provided little else. A state oil company pension was worth only about $1 a month in 2021, insufficient to buy half a dozen eggs.
A woman interviewed in Venezuela drew this pension and worked on Hive to supplement her income, but it was also unstable. Oil fields remain unchanged, while platforms can swiftly transfer tasks to other workers. AI is often seen as built on extracted data, such as authors' books and artists' works. However, there is another layer of extraction: data needed to improve those models. This material must be created by someone for a fee, determined by the platform.
Outlier's terms allow the company to delay or suspend payment during worker investigations and withhold "some or all amounts accrued" at its discretion. The Remotasks terms, under which Rogelio's family worked in 2021, are identical. Earlier this year, Josh Dzieza reported on lawyers, scientists, and screenwriters who have lost their jobs and now work on these platforms.
Nearly all of the 30 workers he interviewed described a similar pattern: demands increased, deadlines tightened, and pay fell as projects progressed. One designer was paid $21 an hour to tag videos for Meta, but when the project ended for "scope changes," the same job reappeared under a new name at $16 an hour. Workers can leave, but those who stay are often those who have no choice.
A woman I spoke with was confined to her home during the Covid-19 pandemic due to a health condition and could not afford to lose her job. For her, this was her only source of income. These issues are known within the industry, but when representation comes up, the room grows silent as no workers are present. Workers have still taken action.
In March 2023, a Kenyan judge temporarily halted the dismissal of 260 Facebook content moderators, preventing Meta from moving their work to another contractor and barring them from blacklisting workers. While Meta moved the work to West Africa, American courts are grappling with this issue. Workers have filed class action suits against companies like Scale AI, claiming they were misclassified and underpaid.
Written by urgent.news from Rest of World's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.