At a recycling colony on the edge of New Delhi, 43-year-old Sunita Rathore does the same work she has done for years: removing labels from plastic bags, sorting waste, stacking heavy sacks. One thing has changed. An iPhone is now strapped to her head, recording every movement of her hands. For each hour she films, she earns an extra Rs 150 on top of her roughly Rs 20,000 monthly wage.
Rathore does not know exactly where the footage goes. She only knows the extra money is helping her save for the education of her four children. She is part of a workforce quietly spreading across India: people paid to record their own everyday jobs, wearing cameras and motion-tracking devices in factories, warehouses, farms and homes, as Bloomberg reported this week.
An iPhone Strapped to Her Head
The scene plays out every shift. In a shoe factory near Delhi, a worker films his hands as he stitches and assembles footwear, earning an extra Rs 300 a day for the footage. In warehouses, workers record themselves packing boxes and scanning products. On farms and in textile units, camera rigs sit among the machinery, capturing the same motions thousands of times.
Robotics companies need this material more than almost anything else. Humanoid robots can perform dazzling demonstrations in videos, but making them reliably fold a towel, pack a box, weld a joint or lace a shoe requires enormous amounts of real-world training data. That is where India offers an unusual advantage: millions of workers still perform manual tasks that were automated decades ago in richer countries, giving AI labs a vast and varied library of human movement.
Startups are now sending camera kits into the field. Some workers wear phones attached to caps or helmets. Others use specialized cameras or motion-tracking gloves. The raw recordings are cleaned, labelled and sold to robotics companies and AI research labs, feeding the models that control the next generation of machines.
A Gold Rush in Human Movement Data
The numbers explain the rush. According to the Bloomberg report, raw first-person video can fetch more than $7 an hour. Highly annotated footage, where every grip and gesture is labelled, commands $15 to $30 an hour or more. For workers earning a few hundred rupees a day, that is life-changing money, and the market is only growing as the global race to build useful robots intensifies.
The tasks being recorded are almost mundane in their variety:
- Folding and stacking clothes in textile units
- Packing and scanning products in warehouses
- Sorting, separating and baling waste in recycling colonies
- Stitching, lacing and assembling footwear in factories
- Picking, cutting and preparing produce on farms
Yet behind the extra income sits an irony the industry itself does not deny. Workers are effectively writing the instruction manuals for machines that could one day do their jobs. The more footage a robot ingests, the closer it gets to replacing the very hands that filmed it.
The Question Nobody Answers Yet
There is also a growing privacy problem. Many workers have little idea how their recordings will be used or who will ultimately buy them. Cameras mounted on heads and helmets can unintentionally capture faces, homes, personal belongings and conversations far beyond the hands they are meant to track.
India's Digital Personal Data Protection Act promises stricter obligations for companies handling such data, but key provisions are not due to take effect until 2027. Until then, the demand for human movement data races ahead of the rules, and the workers supplying it carry most of the risk.
In Bangalore, warehouse worker Yeshwanth N has already thought about what happens if the robots arrive for his job. He says he would go back to his family's land and farm it. "No robot can do that," he told Bloomberg.
The story of this new data economy is not yet a story of mass job losses. It is a quieter one, told in small exchanges: an extra 150 rupees here, a head-mounted camera there. But every hour of footage makes the question harder to ignore. The people being paid to teach the machines are, with each shift, making themselves a little more replaceable, one recorded motion at a time.
Comments