invisible workforce powering AI

Why the AI Revolution Still Depends on the Global South

By Jack Zhai & Shan Qi

In a crowded internet café in Nairobi, or a cramped apartment in Caracas, a young worker stares at a screen. They aren't coding the next LLM. They aren't prompting ChatGPT. They are looking at a grainy image of a stop sign, or a snippet of violent text, and clicking "Yes" or "No."

They do this thousands of times a day. For this, they might earn less than $2 an hour.

This is the dirty secret of the Artificial Intelligence revolution. We are sold a story of silicon magic—of neural networks that learn and think. But peel back the interface of ChatGPT or Gemini, and you don’t find a sentient machine. You find a global assembly line of humans in the Global South, feeding the machine the one thing it cannot generate on its own: context.

The Digital Proletariat

Silicon Valley likes to pretend that AI is "software eating the world." In reality, it’s just the old factory model updated for the 21st century. Just as Nike needed sweatshops in Vietnam to stitch sneakers, OpenAI and Scale AI need "digital sweatshops" in Kenya and the Philippines to stitch together reality.

Platforms like Remotasks and Appen have industrialized this labor. They have turned the complexity of human judgment into piecework. A worker in Venezuela might spend their day labeling dashcam footage for a self-driving car startup in San Francisco, earning cents per task to identify a pedestrian vs. a plastic bag.

This is the "Grey" economy at work. It’s not illegal, but it thrives in the margins of global inequality. For a worker in Venezuela, earning $5 a day in stablecoins (USDT) is a survival strategy against hyperinflation. For the AI company in Palo Alto, it is the ultimate arbitrage: high-value intelligence bought at third-world prices.

The Cultural Moat

Why can’t the models just learn from the internet? Because the internet is messy, sarcastic, and deeply cultural. And AI is terrible at culture.

An image recognition model trained on American data might look at a white robe and see a "dress." A worker in India knows it’s a Sari. A sentiment analysis bot might read "Great job!" as positive. A human annotator in Brazil knows that, in a specific context, it’s dripping with sarcasm.

This is why the industry needs humans. You cannot automate cultural nuance. As AI models expand globally, they crash into the "cultural moat." To launch a chatbot in Southeast Asia, you don't just need translation; you need thousands of locals to teach the model that a "thumbs up" isn't always a like—in some contexts, it’s an insult.

The Global South isn't just providing cheap hands; it’s providing the cultural data that makes these "global" models actually work. Without this invisible workforce, the "intelligence" in AI collapses into hallucination.

From Blue Collar to Click Collar

We are witnessing a shift in the geography of labor. In the Web 2.0 era, the Global South was the back office—call centers in Bangalore and manufacturing hubs in Shenzhen. In the AI era, the factory floor has moved to the cloud.

The Philippines, once the call center capital of the world, has pivoted to become a "hotspot for data labeling". But unlike the call center jobs of the 2000s, which offered a career ladder, this new work is precarious and atomized. There is no manager to complain to, only an algorithm that bans you if your accuracy score drops.

Workers in Kenya have reported creating unions to fight back, exposing the psychological toll of filtering graphic violence for American tech giants. But the leverage is limited. If Nairobi becomes too expensive or too regulated, the platforms simply move the API endpoints to Manila or Lagos.

The Invisible Cost of Intelligence

This dynamic exposes the stark reality of the "White, Grey, and Dark" economy. The "White" economy—the flashy AI product launches in San Francisco—is built entirely on the "Grey" labor of the Global South.

The AI revolution is not replacing humans. It is hiding them. It is taking the cognitive labor of the poorest people on earth, aggregating it, and selling it back to the richest as "magic."

As we marvel at the capabilities of the next GPT, we must remember: that intelligence wasn't just computed. It was harvested. And the fields are far from Silicon Valley.

Digital black markets

Myanmar’s Digital Black Markets: The New Frontier of Global Finance

By Jack Zhai & Shan Qi

On the banks of the Moei River, dividing Thailand from Myanmar, stands a city that technically shouldn’t exist. Shwe Kokko is a $15 billion sprawl of neon-lit casinos, luxury villas, and dormitory blocks rising out of the jungle. It has its own power grid...