Facebook may soon understand you as well as someone from your Close Friends list. And by Facebook, I don’t mean Mark Zuckerberg or another employee — I mean the actual website. The social network formed an eight person research group to develop an artificial intelligence system in order to better understand its users. According to the MIT Technology Review, Facebook’s team is assembling an AI using something called deep learning. AI that uses deep learning will process data using simulated neural networks. These networks will interpret and react to data using the same methods a human brain uses, so instead of superficial interpretations of information (like the kind of understanding Siri can glean from your voice commands) the AI will grasp the deeper meaning of actions on the social network.
Deep learning is an endeavor to create a system so sophisticated it can mimic the way a human mind functions in its totality — if it works, androids so smart they make IBM’s Watson look like a drunk numbskull could be a real possibility. And if it goes awry, we’ll have, um, HAL from 2001: A Space Odyssey, but let’s not think about that.
Most particulars of the project remain secret, but here’s what we know:
Tech companies are already knee deep in deep learning
The idea of an artificial intelligence smart enough to intuit which updates readers want to see sounds Skyrim scary. But Facebook’s not doing anything new. Google already plunged into this research field with its Google Brain project, an attempt to develop an AI that functions on the same level as a human brain. Google’s project is the most ambitious deep learning endeavor thus far, but Microsoft has already dipped its toes in as well, using the theory to develop a real-time English-Mandarin translator. And Facebook needs to hurry, because China’s dominant online company Baidu just established a research center devoted to the same thing in Silicon Valley — and researchers in Japan are using neural networks to control robots.
Facebook’s team has impressive credentials
Since the company is tardy to the tea party on deep leaning, Facebook poached some team members from places with more experience. Marc’Aurelio Ranzato joined the team from Google’s Brain project, and he’s an expert on deep learning. And Yaniv Taigman, the co-founder of Face.com (which Facebook owns), is on board, which may speak to plans to fold facial recognition advances into this deep learning project. Lubomir Bourdev is another team member with a background in facial detection, again suggesting that Facebook’s AI will be great at remembering faces.
Facebook will use it to fix the News Feed
News Feed continues to suck at the moment. It uses conventional machine learning to show users what it thinks we want to see, and it does a terrible job. Although Facebook continues to tweak its formula to “surface” updates, what you see on your feed often bears little resemblance to what you’d find most interesting on Facebook. Facebook’s chief technology officer Mike Schroepfer confirmed that this artificial intelligence would aim to help the News Feed improve. The company hasn’t figured out a way to make News Feed more utilitarian or interesting, which is why the AI will focus on improving the platform’s ability to pull out relevant information from the data barrage.
But the News Feed’s not the only thing that will benefit
Although fixing Newsfeed will be a major priority for Facebook’s AI, it won’t be the only goal. Schroepfer told the MIT Technology Review that the AI might be used to help people manage their photos — which makes sense, considering the facial recognition experts on the team. Part of this could be to show certain photos to certain people; i.e. your Cabo photos might not show up in your parents’ feeds, but will in your college roommate’s (we can only hope, right?). The Facebook AI will also work on a variety of apps to improve Facebook and projects meant to benefit the general public.
Facebook’s getting emotional
This past year, Facebook started asking you more than just what you were doing or thinking about in the status update box; it also wanted to know how you were feeling. The AI Facebook is developing will take this a step further, because the site wants to be able to know what the inflection and emotion behind your posts is; things like sarcasm are hard (for a computer… and some people) to read, and funny-sarcastic post that gets 20 comments and “likes” might not initially get the right News Feed treatment because Facebook just didn’t… get it.
Any by understanding your emotions and what your like better, Facebook might be able to show you the types of things you’re more likely to want to see – or not want to see. For instance, if you write “Really, we get it, your baby is cute – please post 18 more albums of them today!” Facebook would be able to understand you actually don’t want to see more baby photos.
The doors are open
Facebook says that the AI project won’t be a hush hush one; Facebook will share important data and anything interesting that surfaces in the process of this experiment. Of course, we’ll have to be rather trusting that every little finding is publicized, but at least we have a soft guarantee that Facebook isn’t building a robot brain to learn and perhaps eventually destroy us. At least not yet…