Thousands of Indian workers are filming themselves to train the robots that could replace them

[Ederson]

Thousands of Indian workers, from plastic waste sorters in New Delhi to garland makers in Bengaluru, are being paid to wear head mounted cameras that record their every movement, generating what's called egocentric footage that robotics companies are using to teach machines how to move and act like humans in the real world

Bloomberg's feature opens on Sunita Rathore, a 43 year old woman who has spent years sorting and cleaning discarded plastic in Karan Vihar, a recycling colony on Delhi's western fringe, for about 20,000 rupees, roughly 211 dollars, a month, she now also wears an iPhone strapped to her head while she works, angled to capture every motion of her hands, the flick of a blade stripping labels off plastic bags, the stacking of sacks bound for reprocessing by grade, she leans forward constantly to keep her legs out of frame and shuts the camera off if anyone calls out to her, capturing faces is forbidden, and Rathore has no idea what the footage is actually for, or that her labour is now considered critical to the future of artificial intelligence and robotics

The pattern repeats across genuinely different contexts and pay scales, in Bengaluru, 55 year old Ponni has sat roadside for the last decade making flower garlands, and has similarly been paid to have a phone strapped to her forehead, she offered a genuinely pointed reflection on what this means for the next generation, saying the next generation who might have to do work similar to mine, they will face a problem, at AI data company Objectways, which has offices in India and the US and lists Fortune 500 multinationals among its clients, workers film themselves performing household tasks inside fake, fully furnished apartment rooms, one worker, Nagireddy Sriramyachandra, earns 250 rupees, about 2.6 dollars, for one hour of video filming herself slicing mangoes, her app alerts her with hands not detected whenever she's not recording her movements properly, and after several thousand hours of filming inside the same set the studio simply changes the wallpaper to give clients visual variety

The economic logic driving this is genuinely straightforward but the ethical questions remain largely unaddressed, developers believe that feeding first person, egocentric footage into specialised AI models helps robots learn to imitate human movement and behaviour with far more fidelity than footage recorded from an external camera, one industry figure captured the scale of ambition plainly, saying a welder in India could be managing a welder robot in Prague, illustrating how this labour, however modest the pay, is explicitly framed by the companies collecting it as building toward genuinely global remote operated or fully autonomous physical labour systems

Bloomberg frames the deeper unresolved question directly, India spent two decades as the internet's back office, now it's becoming the body double for robots, and the central question remains unanswered, whether workers should share in the long term value their own recorded movements ultimately create, policymakers haven't even begun defining the rules governing this kind of data collection, meaning workers like Rathore, Ponni and Sriramyachandra are generating training data that could underpin the exact automation systems that eventually replace the very jobs they're currently doing, with no consent framework, no compensation structure tied to that future value, and in many cases no real understanding of what their own footage is ultimately being used to build

Thousands Of India Workers Are Helping Ai Firms Train Robots To Replace Themwww.bloomberg.com

[Arty Kayla]

Ponni's line about the next generation facing a problem is genuinely the most quietly devastating quote in this whole piece, she understands exactly what she's building toward even without any technical knowledge of AI or robotics, that's a real and clear eyed recognition of her own situation

[Kai_37]

The hands not detected alert inside the Objectways studio is such a small but telling detail, it shows how much of this work has already been reduced to a mechanical, monitored, quota driven process, workers essentially performing their own labour specifically to be recorded and studied rather than to produce anything directly useful themselves

[RoughDaemon]

250 rupees for an hour of footage that could ultimately train a robot capable of eventually replacing that exact task is such a stark asymmetry between the immediate wage and the long term value being extracted, that gap is precisely the unresolved question Bloomberg raises about whether workers deserve any share of what their movements ultimately help build

[Murky Courier]

This connects directly to broader questions about data labour and data dignity that have been building for years across the AI industry, from content moderators to reinforcement learning annotators, this Indian robotics training workforce is just the newest and most physically literal version of the same underlying extraction pattern

[Owen84]

No consent framework and no compensation structure tied to future value is the real policy failure sitting underneath this whole story, this is exactly the kind of gap where regulation is lagging years behind the actual economic activity already happening at scale on the ground

[Kernel]

The forbidden faces detail is presumably meant as a privacy protection for the workers themselves, but it also conveniently keeps the actual human being anonymous and interchangeable in the resulting dataset, which somewhat undercuts any claim that this labour is being treated with the same dignity or specificity a traditional wage relationship would afford