It’s a story of data-hungry machines, underpaid workers and dodgy contracts, who get spied on by purchased devices.
A woman in the bathroom in a lavender T-shirt and lowered underwear, a child on her stomach in the middle of a hallway, feet, legs, empty rooms, rooms with dogs running around in solitude. All images have the same framing, as if taken by someone lying on the ground looking down. However, the photos were not taken by a human but by a vacuum cleaner.
To be precise from the Roomba J7 from iRobot, the world’s largest supplier of robot vacuum cleaners bought by Amazon for $1.7 billion. Now, it may seem absurd that a robot vacuum turns into an ubiquitous spy, but it all becomes clear by unraveling a story of data-hungry machines, underpaid workers in windowless rooms, and incomprehensible contracts built on a misleading semantics. . Let’s go step by step.
It all starts with 15 images
January 2022, MIT Technology Review Since received 15 screenshots taken by a vacuum cleaner and then posted to closed social media groups. Comment is it possible? In reality, the journey of the stolen snaps isn’t even too long. The vacuums photograph everything, then send the images to Scale Ai, a start-up where (human) workers around the world tag data, audio and photos.
Data is the first step, it is used to make robots more efficient, and able to develop machine learning to go far beyond simple aspiration. However, they don’t do everything on their own, human intervention is always necessary to achieve business goals, and that’s where Scale Ai workers come in. “There’s always a bunch of humans sitting somewhere, usually in a windowless room, just doing a bunch of point-and-click, ‘Yeah, that’s an object or that’s not an object,'” explained Matt. Beane, an assistant professor at the University of California, Santa Barbara who studies the human labor behind robotics.
And the 15 images sent to MIT are only a tiny part of the ecosystem of data provided. In fact, IRobot said it shared over 2 million images with Scale AI alone. And then there’s another unqualified pool of screenshots submitted to other data annotation platforms. James Baussmann, a spokesperson for iRobot, also said in an email that the company had “taken every precaution to ensure that personal data is taken securely and in accordance with applicable law”, but screenshots are output from the protected circuit.
Data-intensive machines
From the start, iRobot has been all about computer vision. Its first automatic device, the Roomba 980, debuted in 2015. It was the first to map a home, adjust its cleaning strategy based on room size, and identify obstacles to avoid. But the artificial vision of robot vacuum cleaners at a price that is not the sale price. To work well, it must be trained on a large and dispersed data set capable of adapting to any home, despite differences in the perimeter or distribution of objects in a room.
The problem is that heavy data can be invasive. “They have powerful hardware, powerful sensors,” Dennis Giese, a PhD student at Northeastern University who studies security vulnerabilities in Internet of Things devices, told MIT Technology Review. “And they can run around your house and you don’t have a way to control them, especially devices with advanced cameras and artificial intelligence, like iRobot’s Roomba J7 series. So to collect all the data able to power hungry machines, you have to spy inside houses.
The “labellers” come into play
The other big protagonist in this story is Scale AI, a start-up that deals with data annotation, a young and growing sector that is expected to reach a market value of $13.3 billion in 2030. The need to feed artificial intelligence with information has created a new profession, that of data labeler. The average profile is still the same as for all emerging trades of the 19th century. Low wages and employment in developing countries.
Indeed, Scale Ai, the market leader, has recruited thousands of workers from less wealthy countries. In 2020, he then ran Project IO and showed his army these ascending images sought by hidden eyes inside homes. The taggers in their Facebook or Discord groups started discussing the IO project and some screenshots were revealed. The company was quick to point out that this is a privacy breach and that employees have signed confidentiality agreements on the material. But how do you control thousands of remote workers all over the world. It lacks an adequate security system, and it is precisely lacking within a company that sees everything in everyone’s homes.
The usual trick for the perfect scam
The latest player in the vacuum cleaner scandal is unwitting buyers. The tricks are always the same, unclear regulations, difficult instructions, minimal semantic differences which then translate into the possibility or not of being spied on at home. For example, the distinction between sharing and selling data, or between privacy and security. To support this system, there is also weak and too easily circumvented legislation, and for this reason, people unknowingly give consent to robots to photograph them while they are sitting on the toilet.