The Human Cost of AI: African Workers Paid a Dollar an Hour to Train the Machines

Mercy leaned forward, took a breath, and loaded another task. A video appeared on her screen: a fatal car crash. Someone had filmed it and uploaded it to Facebook. Her job, as a Meta content moderator at an outsourced office in Nairobi, was to decide if the video violated the company’s guidelines. She was expected to process one “ticket” every 55 seconds during her 10-hour shift.

Then she recognized a face. It was her grandfather.

She ran for the exit, crying. Her supervisor followed—not to comfort her, but to remind her she’d need to finish her shift if she wanted to make her targets. She could have tomorrow off, he said, but since she was already at work, she might as well stay. New tickets kept appearing: the same crash, different angles, pictures of the dead. Her grandfather, over and over.

This isn’t an outlier. It’s the norm.

I’ve been following the AI supply chain for years, and this story from James Muldoon, Mark Graham, and Callum Cant hits harder than most. We talk about AI like it’s magic—large language models, autonomous systems, the next big breakthrough. But behind every algorithm is a human being, often in the Global South, doing the grunt work: labeling data, moderating content, training the models. And the conditions are brutal.

The Workers Behind the Curtain

The article profiles workers at three data annotation and content moderation centers run by one company across Kenya and Uganda. Content moderators manually trawl through social media posts to remove toxic material. Data annotators label images and text so algorithms can understand them. Together, they make AI possible.

Pay? Just over a dollar an hour. Shifts? 10 hours, with quotas of 500 to 1,000 tickets per day. Moderators watch suicides, torture, rape—”almost every day,” one said. “You normalise things that are just not normal.”

Another worker described it as being a “walking zombie.” You can’t zone out, because you have to watch every second of every video to find the highest level of violation. Violence and incitement rank higher than bullying, so you have to watch until the end, just in case it gets worse. It always gets worse.

The psychological toll is staggering. Workers reported collapsing at their desks, attempting suicide, losing spouses. One moderator who was let go said, “Most of us are damaged psychologically, some have attempted suicide… some of our spouses have left us and we can’t get them back.”

And when they break down? Management offers a 30-minute break with a “wellness counsellor”—a colleague with zero formal training. If they run from their desks without logging the right code, their productivity scores drop. Mercy was told she’d committed a policy violation by not entering “bathroom break” before fleeing after seeing her grandfather’s corpse.

This Isn’t New, But It’s Getting Worse

I’ve read similar reports from the Philippines, India, and even Eastern Europe. The pattern is always the same: big tech outsources the dirty work to countries with weak labor laws, pays pennies, and calls it “business process outsourcing.” Meta, OpenAI, Google—they all do it.

What’s different here is the scale. AI training data is insatiable. Every new model needs millions of labeled examples. Every social media platform needs constant moderation. The demand for this kind of work is exploding, and the workers are the ones paying the price.

The authors interviewed dozens of workers. One Nigerian migrant said the company policies were “more strenuous than the job itself.” Workers who witnessed beheadings were told to see the untrained “wellness counsellor.” Productivity scores were marked down if they didn’t enter the right code before having a breakdown.

The Irony

These workers are building the very systems that will eventually replace them. AI models trained on their labor will automate content moderation and data annotation. The companies are already experimenting with AI moderators. When that happens, these workers won’t even have this job anymore.

But for now, they’re the invisible backbone of the AI revolution. And they’re doing it for $1.30 an hour.

I don’t have a neat solution. Fair trade AI? Better regulation? Maybe. But the first step is acknowledging that every time we use ChatGPT or scroll through Facebook, there’s a human cost. And it’s not going away because we ignore it.