Data Work and the Extractivist Cycle of Platforms

• In Event: Networks of Empire: Borders, Racialization, and Social Reproduction through Algorithmic Systems

Abstract

Cecilia, a student from Venezuela, encountered significant financial challenges when the COVID-19 pandemic lockdowns affected her hometown of Coro. Her family members either went on leave or lost their jobs, necessitating alternative sources of income. With the country experiencing a severe financial crisis since 2014, many individuals turned to online labor platforms. These platforms offered opportunities to earn income in U.S. dollars, thereby bridging critical financial gaps.

Recent advancements in machine learning and substantial investments in the field—from autonomous vehicles to large language models—have driven the artificial intelligence (AI) industry to increasingly rely on remote workers. These workers log into platforms to perform various tasks, such as inputting or labeling data. For this work, they earn a few dollars weekly, which is significantly higher than the local minimum wage of one U.S. dollar per month in 2021 but still inadequate to provide a decent living.

The platform Cecilia joined offered an hourly payment program—uncommon in the gig economy—along with bonuses for more complex tasks like labeling 3D images. However, after the Christmas holiday, the platform discontinued this payment program, citing issues with workers logging unworked time. Consequently, bonuses were significantly reduced, causing a sharp decline in the family's income from the platform. With the easing of pandemic restrictions, an increase in oil revenue, and a slight improvement in the country's crisis, Cecilia eventually sought alternative opportunities.

Cecilia's experience mirrors a broader trend among data work platforms. Analyzing web traffic from 14 platforms between 2017 and 2020 reveals cyclical patterns. Major data work platforms in Venezuela experienced fluctuating traffic. The Philippines peaked in traffic in 2018 before declining, followed by Venezuela peaking in 2021 and then decreasing, aligning with the accounts of dozens of workers interviewed for this research. More recently, Kenya emerged as a significant source of traffic until it too declined this past year.

These traffic patterns reflect platform economics, where digital infrastructures like Facebook and Uber rise by consolidating markets and generating network effects. As more people join a platform, it becomes difficult for individuals to leave since their social and professional circles are also there. Data production platforms function similarly, explaining their adoption. However, once a critical mass of workers was reached, bonuses were removed to increase profits. This pattern recurred in countries like the Philippines, Venezuela, and Kenya, where sufficient infrastructure for online work existed, coupled with economic crises driving people to work for minimal pay.

This research situates these patterns within the broader context of extractivism in Latin America, where foreign entities, including companies, exploit labor to extract resources for profit. In the realm of AI, data has emerged as a new resource extracted, and empirical data illustrates an “extractivist cycle” in the way platforms operate. The front end of technologies like ChatGPT ignores this work and the situation of workers. However, by examining the experiences of workers like Cecilia, this research elucidates that extractivism as a system is present in the digital realm and is a backbone of contemporary developments in AI.

Author

• Julian Posada, Yale University