For the past few months, Facebook has secretly been rolling out a new feature to US users: the ability to search photos by what’s depicted in them, rather than by captions or tags.
The idea itself isn’t new: Google Photos had this feature built in when it launched in 2015. But on Facebook, the update solves a longstanding organization problem. It means finally being able to find that picture of your friend’s dog from 2013, or the selfie your mom posted from Mount Rushmore in 2009… without 20 minutes of scrolling.
To make photos searchable, Facebook analyzes every single image that’s uploaded to the site, generating rough descriptions of each one. This data is publicly available—there’s even a Chrome extension that will show you what Facebook’s AI thinks is in each picture—and the descriptions can also be read out loud for Facebook users who are vision-impaired.
For now the image descriptions are vague, but expect them to get a lot more precise. Today’s announcement specified that the AI can identify the color and type of clothes a person is wearing, as well as famous locations and landmarks, objects, animals, and scenes (garden, beach, etc.) Facebook’s head of AI research, Yann LeCun, told reporters that the same functionality would eventually come for videos too.
Facebook has in the past championed plans to make all of its visual content searchable—especially Facebook Live. At the company’s 2016 developer conference, head of applied machine learning Joaquin Quiñonero Candela said that one day AI would watch every Live video happening around the world. If a user wanted to watch someone snowboarding in real time, they would just type “snowboarding” into Facebook’s search bar. On-demand viewing would take on a whole new meaning.
There are privacy considerations, however. Being able to search photos for specific clothing or religious place of worship, for example, could make it easy to target Facebook users based on religious belief. Photo search also extends Facebook’s knowledge of users beyond what they like and share, to what they actually do in real life. That could allow for far more specific targeting for advertisers. As with everything on Facebook, features have their cost—your data.