Many digital technologies, especially AI-driven products, require vast amounts of data to enable them to function properly. This data is collected, curated, annotated, and evaluated by a largely invisible global labor force comprised of millions of workers — sometimes referred to as “ghost workers” — who toil at highly stressful, low-wage, precarious work. Content moderators for social media platforms are often tasked with moderating unseemly or harmful content.
The Exploitation of Data Workers
Data workers are generally independent contractors of third-party subcontractors known as AI business process outsourcing (BPO) companies, or they perform work on digital labor or “crowdwork” platforms. A BPO is a company with independent contractors, who work in an office with a traditional hierarchy and managerial structure. However, on a “crowdwork” digital labor platform, work (usually in the form of “microtasks”) can be completed remotely, anywhere in the world, using a computer and internet connection. Clients who use “crowdwork” platforms post bulk tasks that need completion; workers select the tasks and are paid for each task or piece of work completed. The platforms pay the workers the price indicated by the client minus their fee.
Courtesy of CBS News' "60 Minutes
Data workers fulfill a host of essential functions, including content moderation for social media platforms; image identification; transcription and annotation; data collection and processing; audio and video transcription; and translation. This training data becomes the foundation for AI machine-learning models.
It’s estimated that there are millions of data workers globally. Many data workers on digital labor platforms, in less developed economies, earn as little as $2 per hour when paid and unpaid work is considered, according to the International Labour Organization (ILO). (On average, workers on these platforms, spent 20 minutes on unpaid activities for every hour of paid work, searching for tasks, taking unpaid qualification tests, and researching technical solutions.) Workers in more developed economies such as the U.S. and EU can be paid higher rates than those in the Global South, but even those are often below the minimum wage in many jurisdictions. “Despite performing valuable work for many highly successful companies, compensation from crowdwork is often lower than minimum wages, workers must manage unpredictable income streams, and they work without the standard labour protections of an employment relationship,” says the ILO.
Timnit Gebru, former co-lead of the AI Ethics team at Google — who was fired by Google in 2020 for raising issues of discrimination in the workplace — and co-authors describe a vast and unregulated labor market of data workers. It extends “from Venezuela, where workers label data for the image recognition systems in self-driving vehicles, to Bulgaria, where Syrian refugees fuel facial recognition systems with selfies labeled according to race, gender, and age categories.” Tasks are often outsourced to precarious workers in countries like India, Kenya, the Philippines, or Mexico. “Workers often do not speak English but are provided instructions in English, and face termination or banning from crowdwork platforms if they do not fully understand the rules,” they write. (After leaving Google, Gebru founded the Distributed AI Research Institute (DAIR), a nonprofit research organization that explores “what the world can look like when we design and deploy technology that centers the needs of our communities.”)
In 2023, Time reported that contractors in Kenya were paid as low as US$1.32 per hour and were hired to label toxic content such as “textual descriptions of sexual abuse, hate speech, and violence” to help OpenAI develop automated filters that prevent the public from seeing those disturbing outputs. The Wall Street Journal reported that the contracted workers were paid an average between $1.46 and $3.74 an hour, a fraction of the $12.50 per hour paid to the U.S. outsourcing data training company, Sama, that hired them. “Several of the Kenya workers said they have grappled with mental illness and that their relationships and families have suffered. Some struggle to continue to work,” the Journal said.
Facebook and its owner, Meta Platforms Inc. (NASDAQ: META) are embroiled in ongoing litigation in Kenya. 184 content moderators in Kenya have sued Meta and two contractors — Sama and Majorel — saying the workers’ union organizing efforts had cost them their jobs.
In May 2024, as Kenya’s President William Ruto met in Washington, D.C., with President Joe Biden, 97 data labelers, content moderators, and AI workers in Kenya, wrote an open letter to Biden, which said: “US Big Tech companies are systemically abusing and exploiting African workers. In Kenya, these US companies are undermining the local labor laws, the country’s justice system and violating international labor standards. Our working conditions amount to modern-day slavery.” Among other requests, the workers asked Biden to ensure that “US Big Tech companies can be held accountable in the US courts for their unlawful operations abroad, in particular for their human rights and labor violations.”
“Despite performing valuable work for many highly successful companies, compensation from crowdwork is often lower than minimum wages, workers must manage unpredictable income streams, and they work without the standard labour protections of an employment relationship.”
One of the largest crowdsourcing platforms is Amazon Mechanical Turk (MTurk), a website owned and operated by Amazon. MTurk is a digital platform where people can complete small, simple tasks online for payment. These “microtasks,” often called “HITs” (Human Intelligence Tasks), can include tasks like categorizing images, transcribing text, or answering surveys. A 2016 investigation of Mechanical Turk by the Pew Research Center found most workers on MTurk live in the U.S., although people from around the world are able to use the site. The Center’s survey found that more than half (52%) of MTurk workers reported making an hourly rate of $4.99 or less. Worker advocacy on behalf of MTurk workers is led by Turkopticon, a nonprofit organization that seeks to “organize mutual aid, resources, and advocacy to improve conditions for all people using Amazon’s Mechanical Turk platform while striving to make this work a good job for all.”
In September 2023, four U.S. Senators and four members of Congress, led by Senator Edward J. Markey and Representative Pramila Jayapal, wrote to the CEOs of Alphabet (NASDAQ: GOOG), Amazon.com (NASDAQ: AMZN), IBM (NYSE: IBM), Meta (NASDAQ: META), Microsoft (NASDAQ: MSFT) and other AI companies to express “deep concerns and questions about the working conditions of those who perform your companies’ ‘ghost work’ — unseen but critical tasks such as data labeling” that is critical to AI development. “Tech companies have a responsibility to ensure safe and healthy working conditions, fairly compensated work, and protection from unjust disciplinary proceedings,” the letter said. “Tech companies also must be more transparent about the role data workers play in their AI, so that consumers can make informed choices about the products they use. Unfortunately, many companies have sidestepped these duties, and that must change.”
The legislators asked each CEO to provide answers to a number of detailed questions, including whether their company currently makes publicly available information about the role data workers play in developing AI technology; what steps each company takes to pay workers a “living wage”; whether data workers’ working conditions jeopardize AI model accuracy and introduce bias; and what steps each company takes “to mitigate those risks.”
“Tech companies also must be more transparent about the role data workers play in their AI, so that consumers can make informed choices about the products they use. Unfortunately, many companies have sidestepped these duties, and that must change.”
The plight of data workers and their importance to the tech industry has also been noted by the Partnership on AI, whose membership includes Google, Amazon, IBM, Meta, and Microsoft. In 2021 the Partnership published a report that cited “a growing body of research” that “reveals the precarious working conditions” confronted by data workers, including:
Inconsistent and unpredictable compensation for their work
Unfairly rejected and therefore unpaid labeling tasks
Long, ad-hoc working hours
A lack of means to contest or get an explanation for the decisions affecting their take-home pay and ratings
A lack of transparency around data enrichment labor-sourcing practices in
the AI industry.
“As the complexity of AI systems continues to increase, so too will the demand for data enrichment work. It is important that workers at the heart of this growth are respected, supported, and fairly compensated for their contributions,” the Partnership on AI says. “For the AI industry to grow sustainably, creating the infrastructure to transform data enrichment work into decent jobs is imperative.”