Facebook Inc. doesn’t yet have an intelligent assistant, like the iPhone’s Siri.
But the social-networking company says it’s aiming higher, in what has become one of the biggest battles raging between Silicon Valley’s behemoths: How to commercialize artificial intelligence.
The once-niche field is aimed at figuring out how computers can make decisions on a level approaching that of human intelligence. Apple Inc.’s Siri, Microsoft Corp.’s Cortana and Google Inc.’s Google Now are all early manifestations. They are voice-recognition services that act as personal assistants on devices, helping users search for information–like finding directions or rating nearby restaurants. Both “learn” from their users, adapting to accents, for instance, and learning from previous searches about users’ preferences.
Facebook thinks it can do better.
Want to publish your own articles on DistilINFO Publications?
Send us an email, we will get in touch with you.
“Siri and Cortana are very scripted,” says Yann LeCun, director of artificial-intelligence research at Facebook, in an interview. “There’s only certain things they can talk about and dialogue about. Their knowledge base is fairly limited, and their ability to dialogue is limited,” he said. “We’re laying the groundwork for how you give common sense to machines.”
Apple and Microsoft declined to comment.
Google Executive Chairman Eric Schmidt recently said the company was making progress in image and speech recognition, but admitted at a conference it was a “sore point” at the company that Siri was getting “all the credit.”
Facebook’s LeCun also sees promise in natural-language processing–machines understanding what is being said in speech in a more sophisticated way than Siri or Cortana. And he said image and video recognition is the “next frontier” at Facebook.
“It’s clear that there’s going to be a lot of progress in the way that machines can understand images and activities in video; personal interactions in video between people expressing emotions, and things like that,” he said. A raised eyebrow might mean many different things in different contexts. After a computer shifts through reams of images of people raising an eyebrow, and what happens before or after, it can start to correlate that action.
The basic theory is that the more images the computer analyzes and correlates, the more precise it becomes, statistically. The goal is to approach the same level of correlation that the human brain makes as it processes images sent from a person’s eyes.
“It’s not just about looking at your face to determine your emotions, it’s about understanding interactions between different people and figuring out if those people are friends, or angry at each other,” LeCun said.
French-born LeCun, 55 years old, is one of the world’s leading figures in artificial-intelligence research, specifically of a subset of the science called “machine learning,” or mathematical algorithms that adjust, and improve, as they receive and analyze new data.
While working at AT&T in the late 80s and 90s, Mr. LeCun developed handwriting recognition processing that was eventually used by banks to scan and verify checks. Technology on pattern recognition he developed significantly pushed the commercial applications of image and text recognition and is being used in the search and voice-recognition products and services by Google and Microsoft.
Facebook hired LeCun in late 2013, luring him from New York University, where he remains a part-time professor, shuttling between campus and Facebook’s nearby New York offices. LeCun now spends one day a week at NYU, and the rest at Facebook, where he heads the AI research lab. The lab, split between Menlo Park and New York, is currently 40-members strong, much larger than most university AI research departments–which traditionally have done the heavy lifting on AI research.
Facebook’s AI research is currently being used in image tagging, predicting which topics will trend, and face recognition. All of these services require algorithms to sift through vast amounts of data, like pictures, written messages, and video, to make calculated decisions about their content and context. Facebook has a big advantage over university campuses who have toiled for decades in the field. It can vacuum up the reams of data required to “teach” machines to make correlations.
Facebook last week said its main social network increased to 1.44 billion monthly users, up from 1.39 billion in the 2014 fourth quarter. The company added that it now has 4 billion video streams every day.
“You can work on a project that may take a few years to develop into something useful, but we all know that if it succeeds will have a big impact,” LeCun said.
Date: May 1, 2015