Groups Similar Search Look up By Text Browse About

Facebook’s AI makes little headway in the fight against hate speech


Facebook today published its annual transparency report, and for the first time included the number of items removed in each category that violated its content standards. While the company seems to be very proficient at removing nudity and terrorist propaganda, its lagging behind when it comes to hate speech. Of the six categories mentioned in the report, the number of hate speech posts Facebooks algorithms caught before users reported them was the lowest: For hate speech, our technology still doesnt work that well and so it needs to be checked by our review teams. We removed 2.5 million pieces of hate speech in Q1 2018 — 38 percent of which was flagged by our technology. Are you doing business in Amsterdam in May? Compare that percentage with the number of posts proactively purged for violent content (86 percent), nudity and sexual content (96 percent), and spam (nearly 100 percent). But thats not to say the relatively low number is due to an defect from Facebook. The problem with trying to proactively scour Facebook for hate speech is that the companys AI can only understand so much at the moment. How do you get an AI to understand the nuances of offensive and derogatory language when many humans struggle with the concept? Guy Rosen, Facebooks Vice President of Product Management, pointed out the difficulties of determining context: Its partly that technology like artificial intelligence, while promising, is still years away from being effective for most bad content because context is so important. For example, artificial intelligence isnt good enough yet to determine whether someone is pushing hate or describing something that happened to them so they can raise awareness of the issue. If a Facebook user makes a post speaking about their experience being called a slur in public, using the word in order to make a greater impact, does their post constitute hate speech? Even we were all to agree that it doesnt, how does one get an AI to understand the nuance? And what about words which are offensive in some language, but not another? Or homographs? Or, or, or — the caveats go on and on. When its being asked to read that kind of subtlety, it shouldnt be a surprise Facebooks AI has only thus far had a success rate of 38 percent. Facebook is attempting to keep false positives to a minimum by having each case reviewed by moderators. The company addressed the issue during its F8 conference: Understanding the context of speech often requires human eyes – is something hateful, or is it being shared to condemn hate speech or raise awareness about it? … Our teams then review the content so whats OK stays up, for example someone describing hate they encountered to raise awareness of the problem. Mark Zuckerberg waxed poetic during his Congressional testimony about Facebooks plans to use AI to wipe hate speech off its platform: I am optimistic that over a five-to-10-year period we will have AI tools that can get into some of the linguistic nuances of different types of content to be more accurate. With that estimate, itd be absurd to expect the technology to be as accurate as Zuckerberg hopes itd be now. Well have to check Facebooks transparency report in the next couple of years to see how the companys progressing. The Next Webs 2018 conference is almost here, and itll be . Find out all about our tracks here.

Facebook’s new transparency report now includes data on takedowns of ‘bad’ content, including hate speech


Facebook this morning released its latest Transparency report, where the social network shares information on government requests for user data, noting that these requests had increased globally by around 4 percent compared to the first half of 2017, though U.S. government-initiated requests stayed roughly the same. In addition, the company added a new report to accompany the usual Transparency report, focused on detailing how and why Facebook takes action on enforcing its Community Standards, specifically in the areas of graphic violence, adult nudity and sexual activity, terrorist propaganda, hate speech, spam and fake accounts. In terms of government requests for user data, the global increase led to 82,341 requests in the second half of 2017, up from 78,890 during the first half of the year.   U.S. requests stayed roughly the same at 32,742; though 62 percent included a non-disclosure clause that prohibited Facebook from alerting the user – thats up from 57 percent in the earlier part of the year, and up from 50 percent from the report before that. This points to use of the NDA becoming far more common among law enforcement agencies. The number of pieces of content Facebook restricted based on local laws declined during the second half of the year, going from 28,036 to 14,294. But this is not surprising – the last report had an unusual spike in these sort of requests due to a school shooting in Mexico, which led to the government asking for content to be removed. There were also 46 46 disruptions of Facebook services in 12 countries in the second half of 2017, compared to 52 disruptions in nine countries in the first half. And Facebook and Instagram took down 2,776,665 pieces of content based on 373,934 copyright reports, 222,226 pieces of content based on 61,172 trademark reports and 459,176 pieces of content based on 28,680 counterfeit reports. However, the more interesting data this time around comes from a new report Facebook is appending to its Transparency report, called the Community Standards Enforcement Report which focuses on the actions of Facebooks review team. This is the first time Facebook has released its numbers related to its enforcement efforts, and follows its recent publication of its internal guidelines three weeks ago. In 25 pages, Facebook in April explained how it moderates content on its platform, specifically around areas like graphic violence, adult nudity and sexual activity, terrorist propaganda, hate speech, spam and fake accounts. These are areas where Facebook is often criticized when it screws up – like when it took down the newsworthy Napalm Girl historical photo because it contained child nudity, before realizing the mistake and restoring it. It has also been more recently criticized for contributing to Myanmar violence, as extremists hate speech-filled posts incited violence. This is something Facebook also today addressed through an update for Messenger, which now allows users to report conversations that violate community standards. Todays Community Standards report details the number of takedowns across the various categories it enforces. Facebook says that spam and fake account takedowns are the largest category, with 837 million pieces of spam removed in Q1 – almost all proactively removed before users reported it. Facebook also disabled 583 million fake accounts, the majority within minutes of registration. During this time, around 3-4 percent of Facebook accounts on the site were fake. The company is likely hoping the scale of these metrics makes it seem like its doing a great job, when in reality, it didnt take that many Russian accounts to throw Facebooks entire operation into disarray, leading to CEO Mark Zuckerberg testifying before a Congress thats now considering regulations. In addition, Facebook says it took down the following in Q1 2018: You may notice that one of those areas is lagging in terms of enforcement and automation. Facebook, in fact, admits that its system for identifying hate speech still doesnt work that well, so it needs to be checked by review teams. … we have a lot of work still to do to prevent abuse, writes Guy Rosen, VP of Product Management, on the Facebook blog. Its partly that technology like artificial intelligence, while promising, is still years away from being effective for most bad content because context is so important. In other words, A.I. can be useful at automatically flagging things like nudity and violence, but policing hate speech requires more nuance than the machines can yet handle. The problem is that people may be discussing sensitive topics, but theyre doing it to share news, or in a respectful manner, or even describing something that happened to them. Its not always a threat or hate speech, but a system only parsing words without understanding the full discussion doesnt know this. To get an A.I. system up to par in this area, it requires a ton of training data. And Facebook says it doesnt have that for some of the less widely-used languages. (This is also a likely response to the Myanmar situation, where the company belatedly – after six civil society organizations,  criticized Mr. Zuckerberg in a letter – said it had hired dozens of human moderators. Critics say thats not enough – in Germany, for example, which has strict laws around hate speech – Facebook hired about 1,200 moderators, The NYT said.) It seems the obvious solution is staffing up moderation teams everywhere, until A.I. technology can do as good of a job as it can on other aspects of content policy enforcement. This costs money, but its also clearly critical when people are dying as a result of Facebooks lacking ability to enforce its own policies. Facebook claims its hiring as a result, but doesnt share the details of how many, where or when. …were investing heavily in more people and better technology to make Facebook safer for everyone wrote Rosen. But Facebooks main focus, it seems, is on improving technology. Facebook is investing heavily in more people to review content that is flagged. But as Guy Rosen explained two weeks ago, new technology like machine learning, computer vision and artificial intelligence helps us find more bad content, more quickly – far more quickly, and at a far greater scale, than people ever can, said Alex Schultz, Vice President of Analytics, in a related post on Facebooks methodology. He touts A.I. in particular as being a tool that could get content off Facebook before its even reported. But A.I. isnt ready to police all hate speech yet, so Facebook needs a stop gap solution – even if it costs.