Facebook announced today that it’s acquired London-based artificial intelligence firm Bloomsbury AI, which specializes in natural language processing. According to a post from the official Facebook Academics page, the company says Bloomsbury’s “expertise will strengthen Facebook’s efforts in natural language processing research, and help us further understand natural language and its applications.” In other words, as Facebook continues to hand more platform moderation duties over to algorithms, the underlying AI technology behind those algorithms still has a long way to go. TechCrunch originally reported on the acquisition yesterday.
In the AI community, Facebook is one of the biggest players, with positions at the company’s Facebook AI Research, or FAIR, organization commanding large salaries and projects that span the more cutting-edge sectors of deep learning, natural language processing, and other popular sub-fields. For Facebook, the grand goal is to have AI that is so adept at understanding images, videos, and text that it can effectively moderate the entire social network’s platform, including Facebook, Instagram, and its many other properties.
Right now, scores of actual human beings, often contractors overseas, are tasked with overseeing Facebook content. Those teams inspect flagged and reported material and make hard decisions about what constitutes hate speech or fake news, whether a post violates Facebook’s rules, and when something may be a piece of art and not, say, pornography. As Facebook beefs up its AI capabilities, however, more of that work will be performed by algorithms. But first, Facebook needs its software to better understand language, the intent beyond that language, and other very tricky problems still out of reach for most modern AI.
Its not immediately clear where Bloomsbury will fit into the picture here, though TechCrunch notes that Bloomsbury AI co-founder Sebastian Riedel also helped create a company called Factmata that was designed to help weed out fake news. Facebook does have a FAIR unit at its London offices, and the announcement post says that Bloomsbury’s work has thus far focused on “machine reading and understanding unstructured documents in natural language in order to answer any question.” So it’s easy to see how that could come in handy when it comes to parsing the massive amount of user-uploaded content on the social network every day. Facebook is also working on an AI-assisted home speaker with a display, in the vein of the Echo Show, codenamed Portal, and its possible Bloomsbury’s team may work on that product to help improve its ability to understand spoken commands and return answers with natural-sounding speech.