by Intimate images captured by robot vacuum cleaners during testing have been shared on social media, according to a report. The sensitive photos included several images of a woman sitting on the toilet with her shorts pulled down to her thighs. Another shows a child lying on the floor with his face clearly visible.
The images were among 15 screenshots taken from recordings made by special development versions of the Roomba J7 series robot vacuum cleaners in 2020, which were obtained by MIT Technology Review. These images have reportedly been shared on private groups on Facebook and the Discord chat app by workers in countries like Venezuela whose job it is to tag audio, photo and video data used to train the AI (TO THE).
Roomba maker irobot – which Amazon is acquiring – said the recordings originated from robots with hardware and software modifications that aren’t sold to the public. These devices are assigned to testers and staff who must explicitly accept their data, including video recordings, with iRobot.
However, the incident demonstrates the cracks in a system where treasure troves of data are traded between technology makers and companies that help improve their AI algorithms. This information, including photos and videos, can sometimes end up in the hands of low-wage contract workers employed in remote destinations around the world. These workers do everything from deleting harmful social media posts to transcribing audio recordings designed to enhance voice assistants.
In the case of iRobot, the company is partnering with Scale AI, a San Francisco-based startup that relies on gig workers to review and tag audio, photo and video data used to train the AI. iRobot has previously said it has shared over 2 million images with Scale AI and an unknown amount with other data annotation platforms.
Both iRobot and Scale AI said sharing screenshots on social media violated their agreements. The images in question included a mix of personal photos and everyday photos from inside the homes, including furniture, decorations and objects placed on walls and ceilings. They also contained descriptive tags like “tv”, “plant_or_flower” and “ceiling light”.
The taggers had been discussing the images on Facebook, Discord and other online groups they’d created to share advice on handling payments and tagging complicated items, according to MIT Technology Review. iRobot told the publication that the images came from its devices in countries including France, Germany, Spain, the United States and Japan.
This process of recording and labeling images is used to enhance a robot vacuum’s computer vision, which allows the devices to accurately map their surroundings using high-definition cameras and an array of laser-based sensors. The technology helps them determine the size of a room, avoid obstacles like furniture and cables, and adjust their cleaning regimen.
Computer vision is limited to only the high-end robots on the market, including the Roomba J7, a device that currently costs £459.
iRobot said it collects the vast majority of its image datasets from real homes occupied by its employees or from volunteers recruited by third parties, with the latter being offered incentives for participation. The rest of the training data comes from “phased data collection” that the company uses to build models which it then logs. Its consumer devices capture mapping and navigation information and share some usage data about the feature in the cloud, though that doesn’t include images, according to the company’s support page.
iRobot said it has terminated its relationship with the “service provider that leaked the images” and is actively investigating the matter and “taking steps to help prevent a similar leak by any service provider in the future.”
