Indian Workers Film Their Jobs to Train Physical AI Robots
Barbers and other workers wear head cameras for pay, and the footage helps teach machines how to handle everyday tasks
By AXY Tech Desk ·
Key points
- Indian workers, including barber Mohammed Ashim in Bengaluru, are paid to film their work with head-mounted cameras that record only their hands.
- Local companies collect the footage and sell it to technology firms and research organisations training physical AI, which controls robots.
- Rashdan Technologies pays $5 to $10 (£3.74; £7.47) for eight to 10 hours of recorded work, according to co-founder and CEO Shahil Baig.
- Workers are being filmed in trades ranging from barbering to pottery, embroidery and candle-making.
- Consent, fair pay and the right to stop filming are the central questions as the practice grows.
Indian workers are being paid to film themselves on the job, and the footage is being used to train artificial intelligence. In Bengaluru, Mohammed Ashim, 22, has been cutting hair and trimming beards for four years. Lately he has been strapping a camera to his head while he works. The camera records only his hands and fingers, and the faces of his clients are blurred out.
For a man who learned his trade in a salon, the arrangement is unusual. His daily routine is now the raw material for training machines. Ashim is one of hundreds of thousands of workers across the world who are paid to record their activities, and his case shows how far the search for training data has moved from text and photographs into ordinary manual work.
Why a Barber's Hands Are Useful to Robot Makers
Most of the AI tools people use today, such as chatbots, learn largely from written text and still images. Physical AI is a different kind of system. It is a type of artificial intelligence designed to control robots that move and act in the real world, such as machines that pick up objects, handle tools or move around a room.
Teaching a machine to move is harder than teaching it to write a sentence. A robot needs to see many examples of a person doing a task with their hands, including the small adjustments that a skilled worker makes without thinking about them. Engineers often call this kind of learning from demonstration. A written manual cannot capture the angle of a wrist or the way a blade is held, so companies look for recordings of real work.
That is where a barber's routine becomes useful. A camera fixed to the head gives a close, steady view of how the hands move through a task. For a system learning from that view, a clear and consistent record of each movement can matter more than the room around it. The blurred faces suggest that the people in the chair are not the focus of the data. The hands are.
Inside the Head-Camera Routine
Ashim described the discipline the work demands. "The routine is simple, but you have to be disciplined about how you move. The operations team taught me that my hands always have to stay cleanly inside the frame, the lighting has to be sharp, and I can't tilt my head too abruptly while working."
Those rules change how a barber works. A haircut that would normally be finished by feel now has to be done within a fixed view. A quick turn of the head, which many barbers use to check a line or reach a difficult spot, can spoil a recording. Ashim has had to slow down and think about the frame as well as the client. The job is still the same haircut and beard trim, but the camera turns each movement into something that someone else will study later.
Ashim said he enjoys the work for reasons beyond the money. "It's good money, but honestly, the feeling of doing something this futuristic is what keeps me excited. I never imagined my razor techniques would be useful to a global tech company," he said. His comments show both the pull of the pay and the pride of seeing a familiar skill used in a new setting.
How Footage Moves From a Chair to an AI Lab
The path from a barber's chair to an AI development team runs through several steps. Ashim's footage is collected by a local company. That company may have hundreds of people filming their work at any one time, which makes it more like a data operation than a single employer hiring a single worker.
Once collected, the videos are sold on. The buyers are technology companies and research organisations that use the material to train physical AI. The footage does not go straight from the worker to the lab. It passes through businesses that gather, sort and process it, and each of those steps adds another set of contracts and rules about who can use the material and for what.
Rashdan Technologies is one company that works in this field. It collects and processes what it calls human generated data. The company's co-founder and chief executive, Shahil Baig, described the demand in broad terms. "Every robotics company, every frontier AI lab, has a specific kind of data they want: industrial tasks, agriculture, household chores, small business work like pottery, garment-making, embroidery, candle-making. We go wherever the task is," he said.
For a worker in Ashim's position, the chain can be hard to see. The worker films, a local company gathers the video, and the material is sold to companies that may be far away. Each link in that chain decides something about the footage, from how it is labelled to how long it is kept. The worker usually has the least view of what happens at each stage.
Rashdan's Pay Rate and the Consent Question
Baig said Rashdan pays $5 to $10 (£3.74; £7.47) for eight to 10 hours of recorded work. That is a useful figure for anyone trying to understand the economics of the work. It describes a block of filming, not an hourly wage, and it sets the basic terms for people who are paid to record tasks they already do for a living.
Baig also said that making sure workers give their consent and are comfortable with the process is important. Consent matters more in this kind of work than it might in a typical job. The camera captures the worker's hands and working habits in close detail, and it does so over many hours. The clients' faces are blurred, but the worker is the person being watched and recorded. Whether a worker feels free to say no, and whether saying no carries a cost, are the questions that will decide how fair the arrangement is in practice.
The pay also raises a simple question. If a worker's routine helps build a system that might later do similar work, who benefits over the long run? The current terms cover the hours of filming. They do not, from the information available, describe what happens to the skill once the footage has been sold. That gap is one reason the arrangement will draw attention from workers, unions and regulators in the months ahead, though no such response has been reported so far.
Why Physical AI Needs Clips of Pottery and Candle-Making
Baig's list of tasks shows how wide the search for data has become. It runs from industrial work and farming to household chores and small craft trades such as pottery, garment-making, embroidery and candle-making. Each of these jobs involves hand movements that machines still struggle to copy reliably.
That breadth suggests the demand is not limited to one trade or one country. A robot that can wash dishes, sew a seam or shape clay would need many examples of people doing those things in real conditions, with the corrections, shortcuts and quick judgements that a training manual would leave out. Recordings of working people can supply those details in a way that carefully staged demonstrations may not.
Baig's phrase "we go wherever the task is" also explains the business logic. The data is gathered where the work already happens. That means the workers doing the work become the source material, and the places where they work become the places where AI companies look for their next training set.
Whether Filmed Trades Could Spread Beyond Bengaluru
The next test is whether these arrangements become a standard way of building robot skills. If robots are going to work in homes, shops and workshops, the companies building them will need more footage of those places. That could mean more workers in more trades being asked to film their routines, and more local companies acting as the middlemen between workers and buyers.
For workers, the practical questions are concrete. They will want to know whether their consent is written into contracts, whether the pay reflects the time and skill involved, and whether they can stop filming without losing their income. They will also want to know who can see the footage, how long it is kept and whether it can be reused for purposes they did not agree to.
For the companies, the question is whether these safeguards can be made to work at scale. A local firm with hundreds of people filming at once needs clear rules, and buyers need to be able to show that the material they purchase was gathered fairly. Whether those rules are written down, enforced and tested in practice will shape how this part of the AI industry develops.
For now, the scene in Bengaluru is simple to describe. A man with a camera on his head is cutting hair, his clients are blurred on the recording, and somewhere along a chain of companies his hands are becoming part of the training data for machines that may one day move in the real world.
Questions and answers
Why are Indian workers being filmed at work?
Workers such as barber Mohammed Ashim are paid to record their hands while they do their jobs. The footage is sold to technology companies and research organisations that use it to train physical AI, which controls robots.
How much are workers paid to film their work?
Rashdan Technologies pays $5 to $10 (£3.74; £7.47) for eight to 10 hours of recorded work, according to its co-founder and CEO, Shahil Baig.
Are the workers' faces shown in the footage?
No. The camera records only the workers' hands and fingers, and their clients' faces are blurred out.
What is physical AI?
Physical AI is a type of artificial intelligence that can control robots in the real world. Training it requires examples of people performing tasks, which is why companies are buying recordings of everyday work.
Topics: physical AI, Bengaluru, Rashdan Technologies, Shahil Baig, Mohammed Ashim, AI training data, robotics, India gig work
Sources: BBC Technology
https://axynewsglobal.com/article/indian-workers-film-their-jobs-to-train-physical-ai-robots