The dream of fully automated humanoids walks on the shoulders of humans you've never seen.
In a studio apartment in central Nigeria, a medical student named Zeus straps an iPhone to his forehead, turns on a ring light, and begins recording himself folding laundry with exaggerated slowness. His movements are deliberate, clinical — designed to capture every angle of his hands for hours on end. He is paid $15 an hour to be watched while he does chores most people consider beneath them. The data he generates will train robots at companies like Tesla, Figure AI, and Agility Robotics — the same robots that will, presumably, one day make human domestic labor obsolete.
This is the contradiction at the heart of the humanoid robotics boom. Companies are promising investors a future where robots replace human workers in factories and homes. Micro1, a Palo Alto startup, has assembled a workforce of thousands across 50+ countries — India, Nigeria, Argentina — to film themselves performing the very tasks these robots will supposedly eliminate. The workers are contract laborers, paid per video, and told to keep quiet about it.
The robotics industry has embraced this approach because it works. Just as large language models learned from vast text datasets, humanoid robots need massive amounts of physical movement data to learn manipulation skills. Simulations can teach a robot to flip, but not how to grasp a coffee mug without crushing it. That requires millions of human hands performing millions of mundane actions, captured frame by frame.
But the irony cuts deeper when you examine who these workers are. Zeus earns $15 an hour — good money in Nigeria's strained economy with high unemployment, a pittance in the United States where his data trains robots that will work in American homes. This is geographic arbitrage at scale: companies harvest cheap labor from the Global South to build products that will primarily serve the Global North, all while maintaining plausible deniability about working conditions.
The workers themselves are kept invisible. All workers interviewed by MIT Technology Review used pseudonyms because they weren't authorized to discuss their work. They signed NDAs. They are, in effect, ghosts in the machine — human labels powering systems that promise to render human labor obsolete.
Zeus finds the work boring. "I'm the kind of person that requires a technical job that requires me to think," he told MIT Technology Review. But he needs the money, and so do thousands of others like him. Micro1 found its workforce by posting on LinkedIn and YouTube, and people came — desperate for income in economies with few other options.
This is not automation replacing labor. This is labor being extracted from people who can least afford to lose it, feeding systems that will return the favor by eliminating jobs in the very industries these workers might otherwise enter. The tech industry calls this progress. The workers call it survival.
The question the robotics industry would rather not answer: what happens when the robots are trained and the humans are no longer needed? Based on the pattern already established, probably the same thing that happens to every other disposable workforce — they disappear, unremarked, into the economic margins. The humanoid revolution will arrive not with fully automated dignity but with thousands of unseen hands folding sheets in dimly lit apartments, each movement captured, each gesture commodified, each worker ultimately rendered as obsolete as the tasks they perform.