That AI You’re Using Was Trained By Slave Labor, Basically

"In my point of view, it is digital slavery."

Sorry Servitude

AI has a dirty secret: the people hired to train it are making literal pennies on the dollar, tantamount to slave labor.

As Wired reports, AI-labeling workers often live in places like Venezuela and the Philippines — and even in refugee camps in Lebanon and Kenya — where wages are low and people are in dire need of work. The "microtasks" they find with the gig apps that employ them, however, often end up paying even less than those countries' paltry minimum wages.

Oskarina Fuentes, a Venezuelan national who eventually moved to Colombia to escape poverty, told Wired that working for the Australian data services company Appen, she nets $280 on a good month while working 18-hour days from her bed, which is just shy of her adoptive country's $285 monthly minimum wage.

Fuentes described waking up as early as two in the morning in attempts to get first pick of Appen's tasks, which involve tagging algorithm training data for companies like Microsoft, Google, and Amazon. The tasks pay out between two and 50 cents each, and in a typical hour-and-a-half of work, she'd receive about one American dollar — though when work is slow, she'll sometimes make only make a dollar or two per day.

Crowded Market

Mutmain, an 18-year-old from Pakistan, told the magazine that he used a family member's ID to start working on Appen when he was only 15. Today, he works outrageous hours: 8am until 6pm, and then again from 2am until 6am — and still often only doesn't make more than $50 per month.

"I need to stick to these platforms at all times," the teen told Wired, "so that I don't lose work."

Because micro-gig companies like Appen and its competitors Clickworker and Scale AI only pay for actual hours worked, Mutmain said that tasks can end up taking multiple hours when accounting for research while only netting him a dollar or two.

"One needs to work five or six hours to complete what effectively amounts to an hour of real-time work, all to earn $2," he said. "In my point of view, it is digital slavery."

An Appen spokesperson told Wired that the company is taking steps to reduce its gig workers' difficulty finding tasks, but it nevertheless must find a "careful balance" between the rapid completion of tasks its clients want and the consistency of work its laborers need.

Though this is far from the first time we've heard about the horrors of AI data-training gigs, it's nonetheless shameful that these well-funded companies are outsourcing their labor for such incredibly low wages.

When the gig economy meets uber-cheap labor markets, the result can sound a lot like the sorry state of American prison labor.

More on AI and labor: LinkedIn Laying Off 700 as Microsoft Pivots to AI