AI robot training has reached a Bengaluru barbershop. Mohammed Ashim, a 22-year-old barber, straps a head-mounted camera on for five to six hours a day to record his hand movements while cutting hair. The camera records only his hands and fingers, and clients’ faces are blurred out. Rashdan Technologies collects the footage and sells it to technology companies and research organisations building “physical AI”: AI that can control robots in the real world.
“It’s good money, but honestly, the feeling of doing something this futuristic is what keeps me excited,” Ashim said. “I never imagined my razor techniques would be useful to a global tech company.”
How AI robot training data is collected
Rashdan Technologies co-founder and chief executive Shahil Baig said the company pays $5 to $10 for eight to 10 hours of recorded work. The company takes its orders from robotics companies and AI labs that need AI robot training footage, and the categories it covers include industrial tasks, agriculture, household chores and small business work such as pottery, garment-making, embroidery and candle-making. “We go wherever the task is,” Baig said.
The market exists because there is nothing like this footage on the internet. Manish Agarwal, founder of Humyn Labs, which has bases in Bangalore and San Francisco, said there is no existing archive of embodied human activity anywhere online. “No dataset of a kitchen being cleaned, or a car being assembled by hands that know exactly how much force to apply,” he said. That gap is why the raw material for AI robot training has to be captured from scratch, one filmed task at a time. His firm works with 96,000 data collectors across Latin America and Southeast Asia.
What workers earn for filming their jobs
Mohammed Arman has worked in textiles in Bangalore for 10 years, and he recently started filming his garment stitching and cutting. The camera follows his hands through motions he has repeated for a decade, and the recording turns work he was doing anyway into a second income. The work slots into his existing routine: the camera runs while he cuts and stitches, and the footage is what Rashdan pays for. “This is survival,” he said. “I have a family back home in Bihar, and every rupee I send back matters.”
Jolly Arifur, a science teacher in Guntakal in Anantapur district, has recorded her domestic work for seven months. She films scrubbing, cooking, sweeping and laundry for three to five hours a day, fitting the recording around her teaching and her young son. She is paid $2 to $3 per hour for the AI robot training footage. The hours add up: three to five hours a day of recorded chores, paid by the hour, sits alongside her teaching salary.
AI robot training and the job displacement question
The arrangement raises an awkward question: whether workers filming their jobs are feeding the AI robot training datasets that could one day do those jobs for them. Bino Paul, senior professor at the School of Management and Labour Studies at the Tata Institute of Social Sciences in Mumbai, said it is too early to tell. He warned, though: “By the time workers realise what they have participated in, things may have gone too far.” He wants governments to regulate AI.
India’s place in the AI robot training supply chain
Agarwal argued that India is currently “largely a supplier of raw material” in this industry. He said the country has to move up the value chain of AI robot training, from recording raw video to cleaning, structuring and validating training data. The raw video is only the first step: before it can train a robot, it has to be cleaned, labelled and checked, and Agarwal’s argument is that India should be doing that skilled work as well as the recording. “India can’t afford to stay a footage supplier,” he said.
For now, the cameras keep rolling. Barbers, textile workers and teachers across India record hours of hand movements each day, and the footage flows to the labs and companies buying AI robot training data. Humyn Labs alone works with 96,000 data collectors across Latin America and Southeast Asia.