KARUR, India — Three men stride across a cavernous textile factory floor of Texworth in the southern Indian city of Karur, handing GoPros and smart glasses to workers, including 40-year-old Shoba Kirubakaran. She’s packing yellow aprons decorated with chickens into plastic wrapping for the brand Designs By Danica. It’s the third time Kirubakaran says she’s
KARUR, India — Three men stride across a cavernous textile factory floor of Texworth in the southern Indian city of Karur, handing GoPros and smart glasses to workers, including 40-year-old Shoba Kirubakaran. She’s packing yellow aprons decorated with chickens into plastic wrapping for the brand Designs By Danica. It’s the third time Kirubakaran says she’s been told to don a GoPro.
It’s recording her every hand gesture, to train AI robots to take over this job — a fact that she says the company did not share with her. NPR informed her when they interviewed her on a recent day. “Nobody told me anything about it,” she says.
Kirubakaran has worked in the factory for 10 years, six days a week, from around 9 a.m. to 6 p.m. She makes $90 a month. With that money, she’s helped her husband, a rickshaw driver, raise their two sons. And she adds, “I only know how to do this job.”
Scenes like these shook the Indian public in April after a video went viral showing rows of textile factory workers wearing these headsets. “This is not a scene from a science fiction film,” warned one news anchor from outlet News9Live. “That is happening right now on factory floors across India.” Another outlet, AsiaNet News warned: “Indian workers are training AI robots to take their jobs.”
Seeking “egocentric data”
India is emerging as one key place to record this kind of footage because of its large population, relatively low wages and relative openness to new tech. And developers need millions of hours of this footage, what the industry calls egocentric data — data gathered from the perspective of a human with a camera to train AI robots.
“That can be obtained only by emulating or collecting data from real human beings that does the job,” says Prathosh A.P., a professor at the Indian Institute of Science in Bengaluru. Like many other Indians, he uses initials to refer to his family name. “Think of today’s AIs as brains without a body,” he says. “Now you have to give them all motor skills … to do all kinds of intricate motions that we do using a physical body.”
Some workers are signing up because they’re paid by the hour for the data. But it appears many others, like Kirubakaran, are simply made to record herself working without any explanation. “Call it what it is, which is exploitation of labor,” says Astha Kapoor, co-founder of the Aapti Institute, which studies the impact of technology on people.
What will the impact be for India’s workforce — the world’s second largest, with 600 million people working or seeking work? “Training workers, retraining workers, absorbing them into new occupations,” says Amit Basole, economics professor at Azim Premji University. “Do we have the policy agility to do that? I doubt it.”
The power question
But that vision of a future full of factory and domestic robots isn’t exactly around the corner, say some experts. AI robots are unlikely to be deployed at scale in India — in the near future at least — because they cost a lot of money and consume a lot of power, which is both unreliable and costly across parts of India. “We may be able to make the robots, but they’re bloody expensive to run, right?” says Kapoor of the Aapti Institute. “Humans are cheaper to run than robots.”
That power crisis is evident even in Karur, where an Indian tech company, Objectways, gathers egocentric data to sell onto AI developers. Massive generators loudly rumble to ensure seamless electricity in the buildings it powers. The company appears to be one of the town’s biggest white-collar employers, occupying at least four building blocks in the town.
The CEO, Ravi Shankar, is a Karur native who migrated to Arizona decades ago. He’s in perhaps unlikely agreement with Kapoor about the deployment of AI robots in India. “I think it will be more expensive,” he says of AI robots, in comparison to human labor. “I don’t think it’s going to be cheaper.”
He says his Indian clients want help designing AI robots for jobs that they can’t find people to do, like tapping rubber from rubber trees; jobs that are geographically isolated, like dusting off solar panels in rural areas, and jobs that are physically grueling, like destemming chiles for processing.
Inside Objectways, employee Rajeshkumar Ramaswamy shows us how they are training other robots to do repetitive industrial jobs for foreign clients, through what he calls collecting exocentric data — that’s data collected from the point of view of a robot. Instead of filming hands of actual workers, young men with GoPro cameras strapped to their heads record themselves doing jobs while holding and manipulating what look like grippy robot claws. One man records himself using the claws to screw bottle lids off and on.
There are dozens of bottles — big, small, plastic, metallic. Ramaswamy says it’s to teach robots how to identify all the different shapes and sizes of this thing called a water bottle. The “robot needs to learn what is [a] water bottle,” he says.
Ramaswamy says they’ll repeat this process of screwing bottle tops about 100 times and feed that data into an AI robot, so it can see for itself how to use its own grippy hands to do the job.
“The robot is like a child,” Ramaswamy says. “We are the humans. We are teaching the robots what needs to be done.”
The company also trains robots to do household tasks — like folding towels, putting away books and unpacking bags. It sells that data to AI companies hoping to create domestic robots — there’s already a few on the market, including one prominent model that looks like a faceless servant and retails for $20,000.
Working in a fake kitchen
Up a few flights of stairs, Ramaswamy shows us where they have humans recording themselves doing domestic jobs, on a floor kitted out to look like an Ikea-style three-bedroom house. In a fake kitchen, a woman, Rani N., records herself chopping onions.
Most clients who want first-person recordings of domestic chores are in the U.S., so Ramaswamy says they also pay Americans to record themselves doing housework — so their clients can train AI robots on American-specific data.
For instance, in India, many folks use stainless steel plates to eat. A robot trained to pick up a stainless steel plate could use a lot of pressure to grip it, and could drop it — no harm done. But if the same robot tried to pick up a ceramic plate, which is what many Americans use, the same pressure would cause it to “be separated into minute particles,” Ramaswamy says — that is, smashed to bits.
And to amass recordings of humans doing work with their hands for clients who request ecocentric data, Ramaswamy says, they pay hundreds of people across India to undertake domestic jobs, like Sneha Jegadish, a 24-year-old who lives on the edge of town.
Jegadish clipped her smart phone to a headset and began chopping a carrot, her phone recording her every hand movement.
She records herself doing housework for a few hours a day, for about $2.60 an hour, and makes around $120 a month. Jegadish folds clothes, wipes the floor, does the dishes — jobs she was never expected to do growing up, because her mother wanted her to focus on her studies. So when she got this gig a few months ago, “my mother had to show me how to do those jobs,” she smiles.
Money or not, it’s still drudgery, says Jegadish. One day, she hopes, she’ll have a robot to do her housework. “The work we are doing now is going to be done by robots in the future,” she says.
That’s the same conclusion that 42-year-old textile worker Kannagi Periyasaamy has come to.
At the textile factory, Periyasaamy was folding and packing aprons with a GoPro fastened to her head.
She’s one of the few workers who is paid about $2.60 an hour to record herself, and she’s been told why. She explains: “We are training them [the robots] on how to package the aprons, how to do a quality check, how to fold the aprons.”
Periyasaamy says she’s been doing this for the past three months, enough for her to be convinced that someday, a robot will take over her job. When that day comes, she says, “I’m thinking about taking up farming” — the family owns a scrap of land where she could raise cattle. Meanwhile, Periyasaamy thinks of AI robots as her students and says, “I’m proud I can teach it to do a good job.”
With additional reporting by Vinodh Arulappan
For stories about life in our changing world, subscribe to NPR’s Global Health newsletter.
{For more tech updates, stay tuned to our blog.|Keep following us for the latest insights.|Check back often for more exciting news!}
