Across India, workers such as Ashim are paid to film themselves while performing routine tasks, from razor shaving to garment stitching. Their footage is gathered by local firms that employ hundreds of similar contributors. The video streams are then sold to technology companies and research labs that use them to teach artificial intelligence systems how to manipulate objects in the physical world. This model links low-skill labor with the development of so-called physical AI.
The local aggregators stress consent and privacy, recording only hands and blurring any faces that appear. When clients request multi-camera rigs that surround a worker’s head, the firms negotiate higher fees because the equipment is more intrusive. Workers must keep their hands inside the frame, maintain steady lighting, and avoid sudden head movements, according to the operational guidelines they receive.
Robotics firms and frontier AI labs seek data that cover industrial processes, agriculture, household chores and small-scale crafts, according to Shahil Baig, co-founder and CEO of a data-collection startup. He notes that the companies travel to wherever the task occurs, and that the market is becoming crowded with new entrants in India.
Humyn Labs, founded by Manish Agarwal, coordinates a network of roughly ninety-six thousand data collectors spread across Latin America and Southeast Asia. The firm filters thousands of video, selects the most relevant segments, labels them and packages the data for client use. Agarwal emphasizes that no public archive exists of embodied human activity such as kitchen cleaning or car assembly, so the footage must be captured anew in varied settings to match target markets.
Academic observers warn that the surge in data harvesting could hasten automation of manual jobs. Bino Paul, a senior professor at the Tata Institute of Social Sciences, says it is too early to predict outcomes, but notes that workers may later find little recourse if robots replace their roles. He calls for a socio-technical framework that embeds ethical safeguards and human agency into AI development.
Individual earners describe the work as a modest supplement. A garment worker from Bihar explains that the extra pay for wearing a camera helps support his family, while a science teacher named Jolly Arifur reports earning two to three dollars per hour by recording domestic chores such as cooking and laundry. Both say the tasks are part of their normal routine, and the additional income provides a valuable safety net without requiring new skills.