Facebook already knows a lot about you--just look at the people in your "suggested friends" section or the carefully selected advertisements that seem to correlate pretty nicely with your recent browsing history.
But the social network wants to get a whole lot smarter. That's according to a blog post on Wednesday announcing Deep Text, an artificial intelligence system that Facebook says can understand text with near-human levels of accuracy. The goal is to go beyond merely recognizing keywords--Deep Text will actually understand what the post means.
Facebook is already testing the technology for some users. If Facebook Messenger notices you messaging a friend about needing a ride, for example, it'll pull up a Request a Ride button offering an Uber or a Lyft. But it's not as simple as detecting the word ride or taxi. A brief demonstration video in the blog post shows the Uber/Lyft option popping up when a person types "Let's take a ride there," but not when the message is "I don't need a ride."
Deep Text already operates with enough efficiency to process 10,000 posts every second, in 20 different languages, according to the blog post.
Another use case is already in play: If the AI system notices that you post about trying to pawn something, like an old bike, it'll bring up Facebook's selling tools. The social network says that Deep Text could eventually be able to recognize and remove spammy or offensive content automatically, or locate the most relevant or intelligent comments and make them more visible. If it works, this could be a way to clean up the wasteland that can be the comments section on Facebook; applied to other sites, this kind of technology could make the internet a much more tolerable place.
Facebook could also become far more searchable, providing people with results--posts by friends, news articles, photos, events--on similar topics instead of ones explicitly typed out by the user.
At the same time, though, privacy issues will abound. Messenger conversations are supposed to be private to the people involved, so the fact that a computer can extract meaning from them brings up a plethora of concerns. It's disturbing enough when Gmail pulls keywords from personal emails when deciding which ads to show you. Facebook wants to take that way further.
Still, the project began less than a year ago, so it will be a while before Deep Text achieves anything near human levels of learning. "I think this journey is still one percent complete," Facebook engineering director Hussein Mehanna told Recode, "but it's still far more revolutionary than what we had a few years ago."
If and when Facebook does get there, the company is going to have to prove the conversations won't be used for anything beyond simple suggestions--and that humans aren't reading them. If it can navigate the privacy issues and nail the accuracy of Deep Text, the Silicon Valley giant might be a step closer to creating a far more intuitive experience.