Behind the scenes, Facebook is quietly--but aggressively--going after a curious list of moonshot projects: delivering Internet to the developing world via drones; entertaining people with an alternate universe via virtual reality. And then there's the artificial intelligence.
If you've clicked on a "See Translation" button on Facebook, or have seen it suggest friends for you to tag, you've already interacted with Facebook's AI. But there's much, much more powerful stuff to come. Popular Science recently dove into the company's relatively new Facebook Artificial Intelligence Research team, known as FAIR, which is working on nudging computers toward general intelligence. That means, in essence, making machines that comprehend and learn from the world around them. The goal is to make Facebook--and all its future technologies--more intuitive, and more, well, human.
Building a killer team.
The AI team at Facebook started coming together about two years ago, when, according to his LinkedIn page, Marc’Aurelio Ranzato was hired by Facebook away from Google, where he was working as a research scientist. Popular Science notes he was working on a project called Google Brain. Next came the hire that changed the tide: Yann LeCun, the inventor of modern convolutional neural networks, a technology that's key for applications such as image recognition and natural language processing. He's now the director of FAIR, and "comes from a storied tenure of artificial intelligence research," according to Popular Science. Thanks to LeCun, ATMs can read your checks.
From there, they started assembling a team of about 30 research scientists, along with 15 engineers, half in New York, half in Menlo Park--and a couple in Paris. Not that there's all that much talent to cherry-pick from. "High-level artificial intelligence research isn’t an enormous field, and many of LeCun’s pupils have gone on to seed AI startups, which would be absorbed into larger companies like Twitter," notes Popular Science.
Fine-tuning utilitarian tools.
Facebook's AI team has a number of audacious goals that are only possible in the distant future. But that doesn't mean you don't already interact with some of its more accessible projects. Language translation, for instance.
There's a separate arm within Facebook that deploys some of the software FAIR has created. It's called Language Technology, and it has developed 493 translation directions--meaning, it can translate English to Farsi, and Farsi to English, and each counts as one direction. What's mighty impressive about this, and where AI comes in is not straight-forward translations, but catching the nuance of semantics, including Internet slang. The real challenge is not in understanding idioms, but in realizing when the standard meaning doesn't apply. From Popular Science:
The AI is adaptive be nature, and can be trained on slang quickly. The Language Technology team recently learned that French soccer fans were using a new form of slang to say “wow,” and after training the neural network on that public data, it can now reliably translate that text. They're working now to grow Facebook’s lexicon by training on new data every day, but all languages are now updated monthly.
Developing a smarter digital assistant.
Facebook is pumping resources into its own version of Siri. It just launched to a few hundred Bay Area users, and is being billed as a more highly-capable version of Google Now or Microsoft Cortana. It's able to, say, book a flight and notify the airline that you'll have a lap infant with you. The assistant also can buy a birthday gift for your girlfriend, after making recommendations for a gift.
“We start capturing all of your intent for the things you want to do. Intent often leads to buying something, or to a transaction, and that’s an opportunity for us to [make money] over time,” David Marcus, vice president of messaging products at Facebook, told Wired.
If machines actually get smart enough to track users' intent over time, well, that could easily tread into invasive territory--quickly. Facebook has the ability to roll out its new technology extraordinarily quickly, which is not exactly great news for privacy advocates.
“If we have an idea that actually works, within a month it can be in front of 1.5 billion people,” LeCun said, “Lets keep our eyes focused on the horizon, where our long-term goal is, but on the way there are a lot of things that we’re going to build that are going to have applications in the short term.”
Doing good for the Internet.
While the incredible distribution power and rapid speed of deployment of new AI technology introduces a certain creep factor, there's a lot of simple and good-for-us-all stuff in the works too.
One of those projects is video identification. "Lost of video is 'lost' in the noise because of a lack of metadata, or it’s not accompanied by any descriptive text," Rob Fergus, who leads the AI research team focused on vision, told Popular Science. "AI would 'watch' the video, and be able to classify video arbitrarily."
Here's what's really neat: The technology could not only identify what's going on in the video (Facebook's software can already determine, say, what sport is being played in a video), but it also could work to stop copyrighted content or pornographic video from spreading.
The big goal? If the team can make a smarter Facebook, maybe they can make the Internet smarter too.