Over the past year, news reports have revealed that human reviewers are listening to recordings of people talking to their voice assistants. As it turns out, human reviewers are also looking atfootage.
Bloomberg on Thursday reported that teams in India and Romania review video snippets from Amazon’s in-home security camera to annotate them and train the camera’s artificial intelligence. The $120 camera starts recording and can send notifications to a user when it sees motion.
There’s a significant difference, though, between the Cloud Cam and voice assistant reviews. Amazon pulls from a large pool of user audio to train its Alexa voice assistant and admitted it failed to offer enough information to customers about this practice. Apple, Google, Microsoft and Facebook also said they’ve used human reviews of customers’ audio to train their AI software.
But in the case of the Cloud Cam, Amazon only reviews video clips that are voluntarily submitted by users for troubleshooting, as well as clips from employee testers, according to both Amazon and Bloomberg.
The Cloud Cam reviews, while more limited than the voice assistant controversy, could add to customer concerns that Amazon and other smart-home companies are being too lax with their sensitive data. As Amazon works to broaden Alexa’s capabilities and invite people to use it in new privacy needs much more to ease customer worries., it may have to discuss
Bloomberg reported that some reviewed Cloud Cam clips included sensitive videos, such as rare instances of people having sex. Those sensitive clips are discarded and not used to train the camera’s AI. Additionally, the publication said that despite tight security around the review of Cloud Cam clips, employees have at times shared footage with others.
“We take privacy seriously and put Cloud Cam customers in control of their video clips,” Amazon spokesman Leigh Nakanishi said in a statement Thursday. “Only customers can view their clips, and they can delete them at any time by visiting the Manage My Content and Devices page. Using the ‘feedback’ option in the Cloud Cam app, customers are able to share a specific clip with Amazon to improve the service.
“When a customer chooses to share a clip, it may get annotated and used for supervised learning to improve the accuracy of Cloud Cam’s computer vision systems,” he continued. “For example, supervised learning helps Cloud Cam better distinguish different types of motion so we can provide more accurate alerts to customers.”
On the Cloud Cam’s FAQ page, the company states: “Only you or people you have shared your account information with can view your clips, unless you choose to submit a clip to us directly for troubleshooting.”
At its annual product launch event last month, Amazon hardware chief David Limp opened his presentation with a lengthy discussion about Ring cameras that shuts off all video and audio recordings on these devices. Amazon also now lets customers opt out of Alexa human reviews, and Google and Apple are making similar changes.. Those included a new “Home Mode” for Amazon’s
Ring, the Amazon-owned maker of video doorbells, has also faced concerns about human reviewers. The Intercept and The Information in the past year reported that reviewers used clips customers shared in Ring’s app to train its motion-detection software and sometimes shared clips with each other.
At a tech conference this week, Limp said he “would have been more transparent about why and when we are using human annotating” if he was given a chance to go back in time.
While some customers may find the Cloud Cam reviews concerning, others may not be so surprised by them, considering that they were submitted by users in the first place.
“Before everybody freaks out,” tech industry analyst Patrick Moorhead tweeted Thursday. “NOTE: ‘videos customers submitted for troubleshooting.’ Have we lost all sense of personal accountability and common sense?”