Hundreds of thousands of workers around the world (many of them in India) are being paid to strap cameras to their heads and record their hands as they cut hair, stitch garments or clean homes. As the BBC reports, the footage is sold to technology companies and research organisations to train physical AI training data systems: artificial intelligence designed to control robots in the real world, rather than just answer questions on a screen.
Physical AI is different from the chatbots most people are familiar with. It needs to learn how human hands grip a bottle, fold a shirt or guide a needle, actions that cannot simply be harvested from the internet, because no archive of that kind of embodied, real-world activity exists.
Who is doing the recording, and what are they paid?
Mohammed Ashim, a 22-year-old barber in Bengaluru, wears a head-mounted camera for five or six hours a day while cutting hair and trimming beards. His clients’ faces are blurred, and only his hands appear in the footage. ‘I never imagined my razor techniques would be useful to a global tech company,’ he says. Ashim’s video is collected by a local company and sold on to AI developers.
Shahil Baig, co-founder and chief executive of Rashdan Technologies, one such collection firm, says his company pays US$5 to US$10 (£3.74 to £7.47) for eight to ten hours of recorded work. ‘For a lot of these people, that’s meaningful,’ Baig says. He adds that consent and worker comfort are central to the process: faces are blurred, and no names or contact details are passed to clients.
Mohammed Arman, a textile worker in Bangalore with ten years of experience in garment stitching and cutting, recently began wearing a head-mounted camera on the same basis. ‘For me, it’s simple. This is survival. I have a family back home in Bihar, and every rupee I send back matters,’ he says. Science teacher Jolly Arifur, from Guntakal, Anantapur, films her domestic routines (cooking, sweeping, folding laundry) while working from home around her young son. She says she earns two to three US dollars per hour for three to five hours of recording a day, on top of her teaching salary.
Why do AI companies need this kind of footage?
Manish Agarwal, founder of Humyn Labs, which collects and processes human-generated data from bases in Bangalore and San Francisco, argues that computer simulations alone are not enough to teach robots how to operate in the physical world. ‘Physical tasks require high-quality signals involving limb movements, fingers, wrists, elbows, shoulders and dynamic force application,’ he says, for example, knowing how much pressure to apply so a plastic bottle does not crumble. His firm currently works with 96,000 data collectors across Latin America and Southeast Asia. The footage is filtered, labelled so the AI system knows what it is looking at, and packaged for clients.
Agarwal also cautions that the value of raw video collection may not last. ‘India can’t afford to stay a footage supplier. It needs to move up the value chain: from recording raw video to owning the pipeline that turns that video into something usable,’ he says. In August 2026, ETEntrepreneur reported that the International Federation of Robotics recorded 9,120 industrial robots installed in India during 2024, which was 7 per cent more than the year before. In March 2025, Credence Research projected that India’s AI training datasets market was on course to grow from US$61.98 million in 2023 to US$553.61 million by 2032.
Demand is also pushing collection methods further. Baig says clients are now asking for four to six cameras rigged around a worker’s head to cover an entire scene, not just the hands. Getting workers to agree to wearing what amounts to a five-camera headset is, he says, ‘a completely different, and harder’ conversation than asking them to wear a single small camera, and the pay for that tier of data collection is higher as a result.
Bino Paul, senior professor at the School of Management and Labour Studies at the Tata Institute of Social Sciences in Mumbai, says it is too early to tell whether workers are, in effect, helping to train machines that will eventually take their jobs. For some, AI could be a useful tool, but robots may automate other roles entirely. ‘By the time workers realise what they have participated in, things may have gone too far,’ he says. Paul is among those calling for government regulation of AI, arguing that ‘AI cannot be viewed merely as a technical system’ and that ethics must be upheld through human agency.
