I’m not already an ordinary person. I am a channel of reality, a medium of messages. I hold a knife in my hand and slice an organic cucumber, hunching over so that the iPhone strapped to my forehead can capture all 10 of my fingers. I throw the slices into the salad bowl and finish recording. Somewhere a little robot is a little smarter.
This was my life for an entire week last month, as I collected data from the comfort of my apartment, teaching humanoids how to scrub dishes, fold laundry, and pour drinks, among other diminutive tasks. If robots are ever going to live with us and lend a hand around the house, they need to develop fine motor skills. I proudly did my chores (I don’t usually contribute to massive datasets when I put down my strips). And I was also elated that I managed to earn some money.
First-person videos, shot with a camera attached to the head or chest, are becoming more necessary as more companies try to build bots and improve their artificial intelligence models. Even though the Internet is full of videos you can dig through, extremely specific clips — like thousands of close-ups of hands pouring water into a glass without spilling — can be crucial in tuning your machines to excel at real-world tasks. This style of recording, called egocentric data in the industry, is so popular that some investors estimate leading companies will purchase hundreds of millions of hours of labor from external suppliers over the next few years.
“I want every person in the world to record themselves washing dishes,” says Avi Patel, the 22-year-old founder of data collection platform Kled. “This will create a robot that will mean you never have to wash dishes again.” Egocentric data collection is already on the rise in countries like India, where the majority of people are self-employed Average $125 per monthand these first-person video performances can offer similar rates.
As interest grows, more data collection companies are planning to expand into the United States, such as the standalone app DoorDash Tasks launched earlier this year. Soon many gig workers in the USA can start delivering reality to make ends meet, as well as typical takeaway food at room temperature.
Fortunately, I already had a head holder for my smartphone from testing the DoorDash Tasks app. I already had the impression that personalized video data was the dystopian future of concerts, but I wanted to better understand this emerging industry. Since Tasks isn’t available in California, where I live, I signed up on three other platforms: Kled, Luel, and Waffle Video.
The money I made was modest. I basically trained the robots for next to nothing and didn’t make a dent in the $2,500 a month San Francisco rent I shared with my partner. But concerts had one unexpected advantage: my apartment had never been so spotless.
Kled’s breakthrough moment came when Patel posted the message video on x earlier this year, showcasing a sampling of the company’s extensive video data archive. The clip was quickly viewed more than 4 million times, and data buyers began blowing up Patel’s phone. “Every major base model and every lab contacted me asking for data,” he tells me.
The robot training data is just a sliver of what Kled collects from its more than 300,000 users — mostly, the startup pays people to upload their entire camera roll as AI training data. Patel noted that early adopters are eager to engage in gig work in Malaysia, and there is a “special assignments” section to lend a hand promote uploaded videos. Users select the task they want to film from a list and then capture the content directly through the app. There is no hourly rate given for them; each is marked as low, medium or high paid, with no specific range. (The company says the update will cover rates for many, but not all, tasks in about a month.)