KARUR, India — On a bustling factory floor in the southern city of Karur, workers train AI robots by strapping on GoPros and smart glasses as they go about routine tasks. Among them is 40-year-old Shoba Kirubakaran, who packs yellow chicken-patterned aprons for the Designs By Danica brand. It is the third time she says she has been asked to wear a GoPro, which records every hand movement to build training data for machines. She says no one told her the purpose until reporters explained it. “Nobody told me anything about it,” she says.
Kirubakaran has spent a decade at the textile plant operated by Texworth, working six days a week from about 9 a.m. to 6 p.m., earning around $90 a month. The income helps support her rickshaw-driver husband and their two sons. “I only know how to do this job,” she says.
Seeking “egocentric data”
Public concern surged in April when a video of rows of factory workers wearing headsets went viral, with television anchors warning that Indian employees were training machines to replace them. India has become a key source for what developers call egocentric data, the millions of hours of first-person footage needed to teach AI systems fine motor skills for real-world tasks, thanks to the country’s large workforce, lower wages and openness to new technologies.
“Think of today’s AIs as brains without a body,” says Prathosh A.P., a professor at the Indian Institute of Science in Bengaluru. Teaching those systems to perform intricate motions requires capturing how people actually work, he says. Some workers are paid by the hour to generate data, but others, like Kirubakaran, appear to be recorded without clear consent or explanation. “Call it what it is, which is exploitation of labor,” says Astha Kapoor, co-founder of the Aapti Institute, which researches technology’s social impact.
India’s labor force numbers about 600 million people, making the long-term consequences significant. “Training workers, retraining workers, absorbing them into new occupations,” says Amit Basole, an economics professor at Azim Premji University. “Do we have the policy agility to do that? I doubt it.”
The power question
Despite fears of rapid displacement, several experts note that large-scale rollout of AI robots in India may be limited in the near term because of high costs and power demands. Electricity can be unreliable and expensive, and running advanced machines adds to that burden. “We may be able to make the robots, but they’re bloody expensive to run, right?” says Kapoor. “Humans are cheaper to run than robots.”
Power constraints are evident in Karur, where the tech firm Objectways collects egocentric data for sale to AI developers. Massive generators keep electricity flowing to its offices, which occupy multiple blocks. CEO Ravi Shankar, a Karur native who moved to Arizona years ago, agrees that deploying AI robots in India may not be economical compared with human labor. “I think it will be more expensive,” he says. “I don’t think it’s going to be cheaper.”
Shankar says Indian clients seek help designing robots for jobs that lack available workers, are geographically remote or are physically taxing. Examples include tapping rubber from trees, dusting rural solar panels and destemming chiles for processing.
Inside Objectways, employee Rajeshkumar Ramaswamy demonstrates how teams gather what he calls exocentric data, footage from a robot’s perspective. Young men wearing GoPros manipulate grippy claw-like tools to mimic robot hands while performing repetitive actions for foreign clients. In one exercise, a worker repeatedly screws lids off and on a wide array of water bottles in different sizes and materials so a machine can learn to recognize and handle variations. “The robot needs to learn what is [a] water bottle,” Ramaswamy says. The process is repeated about 100 times, then fed into an AI system to teach it the correct movements. “The robot is like a child,” he says. “We are the humans. We are teaching the robots what needs to be done.”
The company also collects data to train domestic robots for household chores such as folding towels, shelving books and unpacking bags. One prominent household model already on the market sells for about $20,000.
Working in a fake kitchen and household chores
On an upper floor, Objectways has built a mock three-bedroom home outfitted like an Ikea-style apartment. There, workers record themselves doing everyday tasks, such as chopping onions in a fake kitchen. Many clients seeking first-person recordings of domestic chores are in the United States, so Objectways also pays Americans to film themselves doing housework to capture country-specific conditions.
Ramaswamy notes that training on local norms matters. In India, stainless steel plates are common. A robot that grips a steel plate firmly might do no harm, but applying the same force to a ceramic plate, more typical in U.S. homes, could shatter it.
For broader egocentric datasets, Objectways pays hundreds across India to record their hands doing routine household work. On the edge of Karur, 24-year-old Sneha Jegadish straps a smartphone to a headset and films herself chopping a carrot. She spends a few hours a day on chores such as folding clothes, mopping and washing dishes at about $2.60 an hour, earning roughly $120 a month. She says she did not grow up doing these tasks, so her mother showed her how when she got the job. It is still tedious, she says, but she hopes one day a robot will take on the drudgery. “The work we are doing now is going to be done by robots in the future,” she says.
Back at the textile plant, 42-year-old worker Kannagi Periyasaamy folds and packs aprons with a GoPro strapped to her head. She is among the few who are paid about $2.60 an hour and says she has been told the goal. “We are training them [the robots] on how to package the aprons, how to do a quality check, how to fold the aprons,” she explains.
After three months of recording, Periyasaamy believes a robot will eventually take over her role. If that happens, she says she may try farming on her family’s small plot and raise cattle. For now, she views AI systems as her students and adds, “I’m proud I can teach it to do a good job.”
With additional reporting by Vinodh Arulappan
