Facebook is introducing new facial recognition features that will automatically notify users when their photo is posted on the social network, the company said in a blog post Tuesday.
The new features are being rolled out in the face of growing pressure on the company from regulators in Europe, the U.S. and elsewhere who have criticized Facebook for spreading fake news, fostering hate speech, eroding civil discourse and trampling privacy rights. The facial recognition technology could help combat some abusive conduct on the network, the company said, although it may also raise additional privacy concerns.
The feature builds on technology Facebook already uses to suggest tags or labels for people in photos users post, Rob Sherman, the company’s deputy chief privacy officer, said in an interview.
He said the new feature give users more control by informing them when their photo has been posted. They can review the post and then tag themselves, choose to leave themselves untagged or, if they are not comfortable with it, contact with the user who posted the photo to ask them to remove it, or file a complaint with Facebook, he said.
A user must be part of the permitted audience for the page posting the photo in order to receive the notification.
An additional feature will also inform users if anyone across the entire social network tries to post a profile picture containing them, Joaquin Candela, Facebook’s director of applied machine learning, said in the blog. “We’re doing this to prevent people from impersonating others on Facebook,” he said.
Fake accounts have been implicated in the spread of false information on Facebook as well as in some scams perpetrated on the network.
Both features will be turned on-or-off via a single toggle in Facebook’s settings, Candela said.
The new features debuting today will be available everywhere except Europe and Canada, where privacy regulators have previously raised objections to Facebook’s auto photo tagging feature, Sherman said.
Nipun Mather, a product manager in the company’s applied machine learning group, said in an interview ahead of Tuesday’s announcement that the facial recognition features build on advances the company’s artificial intelligence researchers have made in computer vision.
While now able to identify most users in head-on photos, the technology won’t recognize people whose faces are obscured, in shadow or at unusual angles. He also said it would struggle to differentiate identical twins.
“If people have a hard time recognizing someone, then in those situations, computer vision will struggle too,” he said. He said if Facebook’s system did not have high confidence in its identification of someone in a photo, it would leave them untagged.
The company also said it was adding facial recognition to its “automatic alt-text tool,” which allows visually impaired people to hear an audio description of what is in an image on Facebook. Introduced two years ago, the technology recognizes broad object categories like “trees” or “river” and will now be able to read out the names of people in the photos too, provided they Facebook users.
But the system is still not sophisticated enough to provide a full description of the action in a scene, Mather said.