When the Cheapest Resource is Human Labour
There's a lot of talk lately about the environmental costs of AI and they're very real, but it feels like the human cost is getting ignored—and it may be even more significant.
It's not news that big tech has long exploited cheap, third world (a phrase I dislike, because it raises the question "third to what?" but it will do for now) labour to advance its cause. Content Moderation is usually a euphemism for something being reviewed by a remote workeg being paid pennies for each post they review, with little regard for the psychological impacsts of living life at the lowest depths that humanity will go to.
AI is no different, though the conversation seems to now be focused on more on the environmental costs—water and energy use—than the human one. The reality is, as long as the people being used are out of sight and offshore most people in wealthy countries don't care.
They should.
One of the more interesting things people might want to learn is that those fancy word strings generated at will aren't actually intelligent. These intelligences were loaded with sentences written by people.
In December 2022, days after ChatGPT’s release, Winnie discovered a new type of project under the category “transcription.” It wasn’t really transcription. All of the projects were asking her to write prompts and example answers for new chatbots from companies now jostling to compete with ChatGPT. There was a project called Flamingo Generation, which gave her a topic and asked her to write “creative” prompts with a minimum of fifty words and responses that resembled “common internet content” like emails, blog posts, news articles, Twitter threads, and haikus.
==========
Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI (Karen Hao) Your Highlight on page 239 | Location 3662-3664 | Added on Sunday, 23 August 2026 18:32:15
The Winnie in this example is a Venezuelan citizen who, in the wake of that country's economic collapse, singed up for this kind of remote work. It's piecework with extremely low rates and zero security but big tech has long exploited this kind of desperation and this was no exception. ChatGPT was built on the backs of slave labour, for all practical considerations. Colonization has never stopped: it just evolved.
Don't believe it?
Less than a year later, she would learn the truth. In March 2024, Scale would block Kenya wholesale as a country from Remotasks, just like it did with Venezuela. For Scale, it was part of its housecleaning—a regular reevaluation of whether workers from different countries were really serving the business.
An entire country of workers banned, mainly becuase a cheaper well was found. Maybe something to think about while you're writing that latest prompt.
