India Is Training Its Own Replacements: When Workers' Hands Become Robot Data
In India's textile factories, workers wear small cameras while they work. The footage feeds directly into datasets to train humanoid robots. We are building our own replacements — with our own hands, our own gestures, our own data.
In the textile factories of northern India, something subtle and deeply unsettling is happening in plain sight, and almost nobody is talking about it enough. Workers — men and women earning less than five euros a day — have started showing up wearing small cameras strapped to their heads. They wear them while sewing, folding, and cutting fabric for hours on end. They are not streaming on social media. They are not documenting their lives for fun. Those videos, recorded quietly beneath the noise of machinery and the suffocating heat of factory floors, feed directly into the training datasets of humanoid robots that some of the world's largest technology companies are developing right now.
This is not science fiction. Figure One, Tesla with its Optimus robot, and billion-dollar-backed startups like Physical Intelligence are collecting gestural data from real human workers to teach robots the fine movements of manufacturing. According to an analysis published by MIT Technology Review in March 2025, the collection of behavioral data from low-wage workers in South Asia and Southeast Asia has become one of the most active — and least discussed — frontiers in the development of what experts call embodied AI. This is not about teaching a machine to recognize an image. It is about transferring into an artificial system the muscle memory of someone who has been doing that job for twenty years: the exact pressure on fabric, the angle of a wrist threading a needle, the small instinctive adjustments that no manual could ever capture.
The paradox is brutal in its simplicity. Those who earn the least, who have the least bargaining power, who cannot afford to say no, are contributing — with their own bodies, with their own movements — to building the technology that could make them obsolete. It is not artificial intelligence replacing us in some abstract, distant way. It is us, with our own hands, supplying the most valuable raw material of all: embodied knowledge, the kind you cannot find in any book. The World Economic Forum, in its Future of Jobs Report 2025 published in January of this year, estimates that by 2030 more than 85 million global manufacturing jobs could be automated, with a disproportionate impact on low- and middle-income countries where cheap labor has been the only competitive advantage.
The question that remains open — and that no institution has yet answered convincingly — is what happens next. What will these people do when factories no longer need them? The debate around universal basic income returns cyclically to conferences and academic papers, but concrete policy remains scarce and fragmented. Finland ran pilot schemes, Kenya and Namibia conducted limited trials. No one has yet found a scalable answer for billions of people.
There is something profoundly ambivalent about all of this. On one hand, the idea that human beings are reduced to living databases — paid next to nothing to train the machine that will replace them — is one of the most dystopian images that technological capitalism has produced so far. On the other hand, if those exhausting, repetitive, physically devastating jobs genuinely disappear one day, something inside us would want that to be a liberation. The problem is that liberation and unemployment, without an adequate safety net, are the same thing with different names. And the time to build that net is running out far faster than governments seem willing to admit.
Sources: MIT Technology Review, March 2025; World Economic Forum, Future of Jobs Report 2025, January 2025; Physical Intelligence (pi.ai); Tesla Optimus project documentation.