A graph of a Facebook user's social network. Image via Flickr
Make no mistake, Google's not the only web services-cum-big data company that wants to crack artificial intelligence. As the MIT Technology Review reported today, Facebook is working on a "deep learning" initiative, which means its making a serious run at AI.
The company's formed a small research group, internally called the AI team. The specifics of what they're working on are secret, but it involves experimenting with deep learning, a technology that is basically traditional machine learning on steroids, making it possible for computers to think, understand, and self-teach, far beyond what humans originally train them to do.
Deep learning replicates the layers of neurons in the neocortex—the thinking part of the human brain. It's the concept behind Siri, semantic search, IBM's Watson, new medicine discovery, natural-language recognition, and, looking to the future, self-driving cars and robots in the workforce. Right now, it's the best shot we have at making the sci-fi dream of artificial intelligence a modern-day reality.
For the last two years, Google has dominated the deep learning field—not least by hiring the leading mind in the field, Geoffrey Hinton, who pioneered its use for image recognition. But it makes sense that Facebook would chase its rival in this game. The social network has a crucial resource up its sleeve: the boatload of data its been gathering from the 1 billion users (300 million active) that share hyper-personal information about their lives on the platform each day.
Big data and machine learning go hand in hand, since the "neural net" can self-teach by drawing connections and insights from the information it has access to. You probably remember when Google revealed that its virtual brain could identify cats in YouTube videos, even though no one ever told it what a cat is.
What's even more interesting from that discovery is that the computer could figure out more obscure things, like being able to distinguish between similar varieties of skate fish—something most humans wouldn't be able to do. In other words, our brains may be too complex for a machine to fully mimic, but the datasets of companies like Google and Facebook crush the amount of information a human could ever retain.
That's what a Russian venture capital firm figured when it invested hundreds of millions dollars in Facebook two years ago because it was sure the web giant was the future of artificial intelligence. Or what Facebook was probably thinking when it invested in its cofounder Dustin Moskovitz’s startup Vicarious, which set out to replicate the human brain in order to create smart, perceiving, thinking machines.
The unsettling part of all this? Facebook's virtual brain is not only built on top of users' personal data, but could help the company infer even more about you and your life. Technology Review reported that deep learning technology could figure out the emotions in status updates or events even if they aren't explicitly shared, recognize objects in photos, and even predict how people will likely behave in the future. The implications of that level of knowledge and insight are many, and scary.
To some extent, it's happening already, just with the traditional machine learning that Facebook currently uses to improve user experience and develop new features. The news recently broke that Facebook knows traits and details about you that you never told it—like if you're gay, or a Democrat. That revelation led to unnerving games like this one, where you give the website access to your Facebook data and it tells you what your personality is. (Mine was pretty spot on, except for the shy part, psh.)
The question is, do people care if their personal data is being used to build an artificially intelligent brain that can learn even more about them? Will these powers be used for "good"—to perfect the Facebook News Feed and build new and novel features, which the company says is one potential application—or for evil?
Srinivas Narayanan, a Facebook engineer assembling the new research team, told the Technology Review that in addition to company features, it's researching deep learning for "more general research on the topic that will be made public." Facebook’s end game is still a mystery, but users beware: The data you're feeding the social giant every day may be slipping even further out of your control than you thought.