This week, an AI training startup called Shift said it would clean New Yorkers’ homes for free. It has plans to expand to other cities, including London, and looking around my flat, I get a complaint.
But there is a catch. There is always a catch.
When it comes to cleaning, Shift wants photos of its cleaners at work: scrubbing dishes, wiping counters, dusting tables, mopping floors. It wants everything. A video of every boring housework we’d happily release if we could – and how robotics companies are racing to teach machines to do it so they can sell us something to do it for us.
That’s harder than it sounds. Unlike chatbots, image generators, and other AI tools that have exploded in recent years, robots have to deal with the virtual world. That means understanding space, motion, force, friction, strange shapes and materials, unusual light, and everything else humans – and other living things – tend to grasp naturally. That’s why things that are usually simple for us, like folding clothes, picking up an apple, or pouring a glass of water, seem too crazy for roboticists to put together.
Teaching machines to do those things requires data. There is a lot. Text, images, and videos can be easily erased from the Internet on an industrial scale. And they were, often without compensating the people who did it. The physical world is very hard to scrub, and it’s even harder to scrub quietly without paying for it. This means that access to high-quality data is a major hurdle for companies developing wearable AI. It’s a lucrative opportunity, so companies like Shift are getting creative.
They are not alone. In India, recent reports revealed that home services platform Pronto has been using clients’ homes as a source for AI training videos for household tasks such as cooking, cleaning, and laundry. Pronto says it only records videos if customers opt-in — it’s unclear what the customers get in return, other than a copy of the videos — but the practice is still causing a wave of backlash in the market, with controversial startups insisting they’ve never recorded indoors to train AI and have no plans to do so.
Some startups are focused on trying to scale data collection. The Silicon Valley-based Human Archive, for example, hopes to partner with companies like Pronto and have gig workers record their work using non-stylish camera caps. The hats collect images from the perspective of the wearer, exactly the kind of first-person data robotics companies need to teach machines how people navigate a virtual environment. Shift, on the other hand, also taps consumers directly, and says it has paid tens of thousands of people in 15 countries to record their activities with its app.
Some companies skip the useful function altogether. Instead, workers are paid to complete the same tasks over and over again while cameras and sensors capture every movement. Such on-stage data farms are designed to turn rote physical work — folding towels, picking up cups, carrying boxes — into AI tutorials valuable enough to justify paying people to create them.
And some data is generated by robots that already exist in the world. Despite the hype, true automation is still a long way off – hence the need for all this data – but companies are keen to ship products anyway. They will use data from customers’ homes to improve the product. Many companies rely on remote workers to step in when robots inevitably take hold. They will use that data again.
Of course, the act of trading data for value is not new. Companies have been offering discounts, conveniences, and free services to get access to your data for years, from loyalty cards and cookies to dashcams, insurance apps that monitor how people drive, and that smart TV that keeps showing ads.
What’s new is the type of data companies are willing to pay. For now, that means maybe letting someone clean your home with a scented hat for free so that, eventually, the company sells you a robot to do it instead.
