Products You May Like
Meta’s collaboration with Ray-Ban has been on the market for just under a year now, and to mark the occasion, Meta has confirmed what we probably should have assumed in the first place: It’s grabbing snapshots from users’ real lives to train the glasses’ AI model. A representative for the company said Wednesday that when Ray-Ban Meta users ask their glasses to analyze something they’re looking at, Meta keeps that visual query for itself.
The discomfiting admission stems from an inquiry TechCrunch sent Meta late last month. At the time, TechCrunch staff asked Anuj Kumar, a senior director at Meta’s AI wearables division, whether Meta would train its AI models on images from the camera-equipped eyeglasses. Kumar responded that Meta was “not publicly discussing that,” and shortly after, a Meta spokesperson stepped in to confirm that Meta wouldn’t “say either way.” The company’s refusal to answer such an essential privacy question was alarming, but that was the end of that discussion.
Now Meta has elaborated on its formerly unforgiving public stance. “In locations where multimodal AI is available (currently US and Canada), images and videos shared with Meta AI may be used to improve it per our Privacy Policy,” policy communications manager Emil Vazquez said in an email to TechCrunch. Another email clarified that while photos and videos captured via Ray-Ban Meta are not used for AI training, asking the glasses a question about that media effectively hands the image or clip to Meta for model improvement.
Examples of how Meta AI can use your field of vision in real time.
Credit: Meta
There is reportedly no way to opt out of Meta’s AI training other than by refusing to use Meta AI in the first place. But AI is largely responsible for lending Ray-Ban Meta its novelty. Promotional videos for the glasses show users asking Meta AI for facts about landmarks they’re gazing at, for recipes involving the ingredients they have on the table, or to translate a sign they see in a foreign language. Without Meta AI, Ray-Ban Meta is just a camera that sits on your face.
This isn’t the first time Meta has helped itself to its users’ media. Late last year, the company’s vice president of global affairs admitted that Meta had trained Imagine, its AI-powered image generator, on more than a billion publicly visible Facebook and Instagram posts. Meta’s defense was that it hadn’t pried into any images shared exclusively with friends and family. However, the executive added that only the “vast majority” of Imagine’s training data was public. Third-party AI firms also found this summer that Meta was using custom web crawlers to scrape AI model training data from across the internet.