SCOTT DETROW, HOST:
Earlier this month, NPR's international correspondent Diaa Hadid went to a small town in southern India called Karur. She reported on how factory workers are made to train the AI robots that would replace them. She also found a thriving industry to train robots to perform domestic work.
(SOUNDBITE OF CHOPPING CARROTS)
DIAA HADID, BYLINE: Twenty-four-year-old Sneha Jegadish chops a carrot in her kitchen. It's an ordinary scene in a middle-class home in a small Indian town, except Jegadish has clipped a smartphone to a headset. It's recording her every hand movement.
(SOUNDBITE OF ARCHIVED RECORDING)
SNEHA JEGADISH: I put a flat cloth under the (ph)...
HADID: AI industry experts say there's thousands of people like Jegadish across India, who over the past year, have found work recording themselves like this. And for about two or three hours every day, Jegadish records herself folding laundry, mopping the floor, any housework she can find, basically, and she earns about $120 a month.
JEGADISH: Double folding, cloth folding. Also the kitchen (inaudible) washing. (Speaking Tamil).
HADID: There's enormous demand for this kind of footage because AI companies need millions of hours of these recordings to feed into AI robots so they can do household chores. This data can't be scraped off the internet.
Prathosh A.P. is a professor at the Indian Institute of Science in Bengaluru. Like many southern Indians, he uses initials to refer to his family name. He specializes in machine-learning systems in data-scarce environments. He says this is called physical AI.
PRATHOSH A P: Think of today's AIs as brains without a body, right? They don't have any agency to act physically because they don't have a body. Now you have to give them all motor skills.
HADID: Because they're brains, and they need to learn how to use their bodies.
(SOUNDBITE OF ARCHIVED RECORDING)
UNIDENTIFIED PERSON: Today, we're launching Neo (ph).
HADID: The demand for this kind of data has created opportunities for Indian companies like Objectways. It occupies a few buildings on the main street of Karur.
(SOUNDBITE OF CAR HORNS HONKING)
HADID: Senior Objectways employee Rajeshkumar Ramaswamy tells us they pay thousands of people to record themselves working.
RAJESHKUMAR RAMASWAMY: Philippines, then Nigeria. All over the world.
HADID: From all around the world. Ramaswamy's team also does some in-house recordings, he says, to obtain better quality data.
(SOUNDBITE OF FOOTSTEPS)
HADID: We walk up a few flights to a floor set up like a house.
Oh, my God. It's like uncanny valley, hey?
Like in a TV studio.
RAMASWAMY: This is a living room. Then that is the kitchen. Then we have three bedroom set ups over here.
HADID: In the set up kitchen, Rani N. records herself chopping onions. Like other southern Indians, she uses an initial to refer to her family name.
(SOUNDBITE OF CHOPPING ONIONS)
HADID: Then she'll record herself doing other jobs.
RANI N: Next time, carrot, beans. We change the task like wiping the table, wiping the cupboard.
HADID: Ramaswamy says at this pace, domestic robots could retail in India in the next few years. He thinks about how a robot might help his own stay-at-home wife, who's raising their baby son.
RAMASWAMY: It will be preparing food, washing the clothes. My wife will get rest, and all the things will be done by humanoid robot.
HADID: The chores will be done by a humanoid robot. Back at the home of Sneha Jegadish, the young woman who records herself doing housework, she's pretty sure AI robots doing housework are coming down the line...
JEGADISH: (Speaking Tamil).
HADID: ...And she can't wait for the day that a robot takes over this tiresome job, even if it means she loses this gig, the one she has right now.
Diaa Hadid, NPR News, Karur, India.
(SOUNDBITE OF MUSIC) Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
