Your AI Runs on Underpaid Workers You Never See

Every time an AI chatbot answers you politely, someone had to teach it what a polite answer looks like. A new paper from researcher Julian Posada at arXiv argues that this teaching is done by a vast, mostly invisible workforce of data workers — people who label photos, sort text, and clean up messy information so AI systems can learn from it. Many of them live in Venezuela, Colombia and similar places, working through digital platforms for low pay and little security. Posada calls this system "platform extractivism" — the idea that AI companies extract data and labor from poorer regions, then keep the profits in places like Silicon Valley.

The paper's most surprising finding isn't about money — it's about culture. When workers label data, they're often told to follow US-style categories and norms. Those choices get baked into the AI, which then presents them to the whole world as neutral, universal knowledge. Posada argues this quietly repeats old colonial patterns, where one region's way of seeing the world becomes everyone's default. That matters if you live outside the US, or if you've ever noticed an AI giving strangely American answers about health, family, or manners.

There's a third layer too: the paper points out that this work is only possible because of unpaid labor at home — family members, community networks, and neighbors who cover childcare or share resources so someone can spend long hours at a screen. In other words, the cheap AI you enjoy rests on a chain of costs that never show up in a press release or a stock price.

The takeaway isn't that AI is fake or useless. It's that AI is a labor-intensive, deeply human industry with real people at the bottom of it. If you use AI tools at work, the paper is a reminder that "automation" often means someone else doing the boring part for very little money — and that the answers you get may quietly carry one country's assumptions.