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 took enforcement action on 1.9 million posts related to terrorism by Al Qaeda and ISIS in the first quarter of this year, the company said, up from 1.1 million posts in the last quarter of 2018. The increased enforcement, which typically results in posts being removed and accounts being suspended or banned from Facebook, resulted from improvements in machine learning that allowed the company to find more terrorism-related photos, whether they were newly uploaded or had been on Facebook for longer. Facebook found 99.5 percent of terrorism-related posts before they were flagged by users, it said. In the previous quarter, 97 percent of posts were found by the company on its own. Facebook made the data available as part of its first ever Community Standards Enforcement Report, which documents content moderation actions taken by the company between October and March. Other findings in the report include: Graphic violence. Posts that included graphic violence represented from 0.22 percent to 0.27 percent of views, up from 0.16 to 0.19 percent in the previous quarter. The company took action on 3.4 million posts, up from 1.2 million in the previous quarter. It said violent posts appeared to have risen in conjunction with the intensifying conflict in Syria. Nudity and sex. Posts with nudity or sexual activity represented 0.07 to 0.09 percent of views, up from 0.06 to 0.08 percent in the previous quarter. The company took action on 21 million posts, about the same as the previous quarter. Hate speech. Facebook took action on 2.5 million posts for violating hate speech rules, up 56 percent from the previous quarter. Users reported 62 percent of hate speech posts before Facebook took action on them. Spam. Facebook took action on 837 million spam posts, up 15 percent from the previous quarter. The company says it detected nearly 100 percent of spam posts before users could report them. Fake accounts. Of Facebooks monthly users, 3 to 4 percent are fake accounts, the company said. It removed 583 million fake accounts in the first quarter of the year, down from 694 million in the previous quarter. The data, which the company plans to issue at least twice a year, is a move toward holding ourselves accountable, Facebook said in its report. This guide explains our methodology so the public can understand the benefits and limitations of the numbers we share, as well as how we expect these numbers to change as we refine our methodologies. Were committed to doing better, and communicating more openly about our efforts to do so, going forward. The company is still working to develop accurate metrics that describe how often hate speech is seen on the platform, said Guy Rosen, a vice president of product management, in an interview with reporters. The companys machine-learning systems have trouble identifying hate speech because computers have trouble understanding the context around speech. Theres a lot of really tricky cases, Rosen said. Is a slur being used to attack someone? Is it being used self-referentially? Or is it a completely innocuous term when its used in a different context? The final decisions on hate speech are made by human moderators, he added. Still, people post millions of unambiguously hateful posts to Facebook. In March, the United Nations said Facebook was responsible for spreading hatred of the Rohingya minority in Myanmar. Facebooks lack of moderators who speak the local language has hampered it in its effort to reduce the spread of hate speech. We definitely have to do more to make sure we pay attention to those, Rosen said, noting that the company had recently hired more moderators in the area. The enforcement report arrives a month after Facebook made its community standards public for the first time. The standards document what is and isnt allowed on Facebook, and serves as a guide for Facebooks global army of content moderators. Facebook is releasing its enforcement report at a time when the company is under increasing pressure to reduce hate speech, violence, and misinformation on its platform. Under pressure from Congress, Facebook has said it will double its safety and security team to 20,000 people this year.