Earnings calls are usually meant to do little more than reassure investors — but in Facebook's most recent call, the day before Halloween, Mark Zuckerberg unveiled something unexpected. Facebook had formed an Artificial Intelligence group, Zuckerberg announced, and the company was acquiring a machine-translation company and hiring the best academic minds in the field. The goal, he told investors, was "to do world-class artificial intelligence research using all of the knowledge that people have shared on Facebook." He teased new products that would be more natural to interact with, and could solve problems beyond the reach of current technology. For anyone on the call, the point was clear: the future of Facebook would be powered by AI.

Now we're getting a peek at what that future might look like. Last week, the company announced it was hiring NYU professor Yann LeCun, a specialist in the field of machine recognition. His work deals with increasingly intricate forms of sorting, whether it's sorting music or algorithms to build 3D spaces out of 2D photos. At DARPA, he used that tech to model the human brain’s powers of visual perception, recognizing faces and potentially powering tech to let automated drones navigate buildings without crashing. At Facebook, which needs to manage and sort all the content posted by their 1.2 billion users, he’ll be more concerned with categorizing photos and status updates. LeCun says the team’s first task will be to optimize Facebook's image-content analysis, but it will soon move on to more sophisticated video and language analysis, towards a clearer picture of Facebook at large. "We want a much better understanding of what can be of interest to users," LeCun says. "That's the foundation for the longer term."

Machine learning is nothing new to Facebook. Self-learning systems already scan all the photos and text on the system, making crucial choices as to what ends up in your Newsfeed. LeCun's goal is to streamline that process, automating it even further. At the same time, he'll be able to use Facebook's immense quantity of content to perform new research, testing out new techniques on one of the largest platforms on the web.

If that playbook sounds familiar, it should: it's the same pitch Google gave earlier this year when it rolled out the Google Brain project, helmed by a similarly eminent professor at Stanford, Andrew Ng. Google has different problems — more data, less user interaction — but the basic parameters are the same. Modern web companies are dealing with vast quantities of data, so vast that they can only be managed by self-learning programs. That's how Google indexes the web with PageRank, and how Facebook assembles your Newsfeed. And as Google proved in with its search, a good algorithm can make a huge difference.

The result has been a kind of arms race in artificial intelligence — once one of the least practical corners of computer science. Facebook, Google, and Microsoft all have their own AI research wings, each pursuing next-generation algorithms that may not pay off until the next decade. Paul Allen has his own privately funded project, while IBM is calling for a revolutionary shift to "cognitive computing," in which computers will be entirely driven by self-learning programs. Meanwhile, very little has changed for the average user. Your Newsfeed has gotten a little smarter and image searches have become more precise, but it's an easy shift to miss if you're not looking for it. The changes are still mostly happening behind the scenes.

Still, LeCun doesn't think they'll stay behind the scenes for too much longer. "The purpose for Facebook is to facilitate communication between people, and that requires a pretty good model of what people want," LeCun says. "I have to believe that you need intelligent systems to do a good job on that." That could mean better friend recommendations, but for the speculative it could also mean something even stranger, like a next-generation chat app that tells you who to talk to or a service that autofills the location of every photo in your feed. LeCun is tight-lipped about the direct applications of the tech (he doesn’t officially start until January), but he’s not shy about talking up the potential. "It could transform Facebook in very major ways five, ten, or fifteen years from now," LeCun says. Or, as skeptics will remind you, it could be a very expensive research project. The surprising thing is that, as the machine-learning arms race heats up, even the best brains can't predict exactly where it's headed.