The mandate for the 50-person AI team is also vintage Zuckerberg: Aim ridiculously high, and focus on where you want to go over the long term.
“One of our goals for the next five to 10 years,” Zuckerberg tells me, “is to basically get better than human level at all of the primary human senses: vision, hearing, language, general cognition. Taste and smell, we’re not that worried about,” he deadpans. “For now.”
In part, the AI effort is an attempt to prepare Facebook for an era in which devices from wristwatches to cars will be connected, and the density of incoming information which the service will have to deal with will grow exponentially. “There’s just going to be a lot more data generated about what’s happening in the world, and the conventional models and systems that we have today won’t scale,” says Jay Parikh, the company’s VP of engineering. “If there’s 10x or 20x or 50x more things happening around you in the world, then you’re going to need these really, really intelligent systems like what Yann and his team are building.”
But [Rob] Fergus and his fellow researchers have the freedom to start small rather than think immediately of the massive data problems posed by services with several hundred million users or more.
Yann LeCun [Director, Facebook AI Research] has given Facebook a lab with a strong university like feel. Rather than having to make sure their work lines up with Facebook’s product plans, researchers—many of them fellow academics—can pursue their passions while a separate group, Applied Machine Learning, is responsible for figuring out how to turn the lab’s breakthroughs into features.
“The senior research scientists, you don’t tell them what to work on,” LeCun says. “They tell you what’s interesting.”
Technologies incubated by LeCun and his team are already popping up in Facebook products such as Moments, a new app that scours your phone’s camera roll for snapshots of friends, then lets you share those photos with those people. “Most researchers do care about their stuff having practical relevance,” says Fergus, who is technically still on leave from NYU, where he worked alongside LeCun. “In academia, a great outcome is you publish a paper that people seem to like at a conference.”
LeCun’s work is directly affecting Facebook’s bottom line, in the form of better spam-prevention tools and software to verify that ads are up to company standards, a task that was once a labor-intensive manual process. “I joke that the lab has paid for itself over the next five years with work they’ve already done,” says Schroepfer.
Click here to learn about Google’s approach to AI
Click here to learn more about Mark Zuckerberg’s vision for Facebook