SCOTT DETROW, HOST:
That "Jetsons" future of robot butlers doing all of our household chores may be coming sooner rather than later, but in order to get there, you may need to strap a camera to your head and record yourself washing dishes or folding laundry or taking out the trash. The idea is that all these videos will help train the AI robots. Wired writer Reece Rogers recently spent a week collecting this so-called egocentric data, which data companies are paying not exactly top dollar for. We called him to talk about his experience. Hi, Reece.
REECE ROGERS: Hey. Thank you so much for having me on today.
DETROW: Let me just start with the why of this. Like, what is the point of all of these videos of people - I don't know - chopping onions or cleaning their sink or washing dishes?
ROGERS: Many people in Silicon Valley see robots as the next big frontier where breakthroughs are capable, and video data can really help robots understand the world, how it works, how they can move. And so many companies are paying users like myself and other people around the world to film themselves doing tasks at home and recording them so robots can learn from these thousands of videos and hopefully one day get better and actually be able to help out around the house.
DETROW: And I want to talk about your experience doing this. But first, help us draw the connection a little bit more about what happens after you, you know, upload a video of you washing dishes. Like, what are companies doing with this information?
ROGERS: So if I record a video of myself - say it's one minute - the camera is strapped to my forehead. It's recording both of my hands specifically, because it's how robots are interacting with the world. It can show them what works and what doesn't work. So I send this one video off to a company. Maybe they're paying $20 an hour for uploaded video. So I would only get maybe a dollar or two from that upload. And then the companies collate thousands of these videos and put hand-tracking algorithms on there and look for patterns and package this data in very clean ways and then sell this data to a larger robotic company or maybe an AI company working on their models. And then by bolstering the amount of data behind the robots, it's often able to improve their ability at specific movements, like doing the chores.
DETROW: How did it make you feel to do all of these chores and record yourself? Like, did this feel like a fulfilling job?
ROGERS: At the end of the week, I was the one feeling like the robot
(LAUGHTER)
ROGERS: I was feeling, honestly, a little bored...
DETROW: Yeah.
ROGERS: ...Recording all these repetitive tasks. There was one. I was tying my shoes probably 20 times - tying it on, tying it off, tying it on, tying it off. I would not say this felt like a fulfilling job, but I do see this as the future of gig work, and oftentimes gig work is not fun or enjoyable. This might be another big type of gig work that people are not necessarily choosing to do because they're passionate about it - no - but because they have to. There's a line in the piece where we say training a robot tonight to cook so you can put food on the table tomorrow. And it's a bleak sentence, but I think it really captures the spirit of what many, many workers might be living...
DETROW: Yeah.
ROGERS: ...Over the next few years.
DETROW: Did anybody ever come across you when you had a camera strapped to your head doing any of these chores? Did you have any, like, awkward conversations with your neighbors as you were taking the trash out?
ROGERS: No, but I scurried out the back door as fast as I could and threw that trash away and scurried back inside because I'm not going to lie. I was embarrassed. At the end of the day, training a robot to do these chores - if it can do the chores, it can do so much more than that. So it kind of dawned on me that I really am kind of - in my small part, was training a robot to replace not only maybe a house cleaner but also someone just walking around the street in the future.
DETROW: Reece Rogers, Wired's service writer, thank you so much.
ROGERS: Thank you. 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.
