Facebook Boosts A.I. to Block Terrorist Propaganda -- Update

By Sam Schechner Features Dow Jones Newswires

Under intense political pressure to better block terrorist propaganda on the internet, Facebook is leaning more on artificial intelligence.

Continue Reading Below

The social-media firm said Thursday that it has expanded its use of A.I. in recent months to identify potential terrorist postings and accounts on its platform -- and at times to delete or block them without review by a human. In the past, Facebook and other tech giants relied mostly on users and human moderators to identify offensive content. Even when algorithms flagged content for removal, these firms generally turned to humans to make a final call.

Companies have sharply boosted the volume of content they have removed in the last two years, but these efforts haven't proven effective enough to tamp down a groundswell of criticism from governments and advertisers. They have accused Facebook, Google parent Alphabet Inc. and others of complacency over the proliferation of inappropriate content -- in particular, posts or videos deemed as extremist propaganda or communication -- on their social networks.

In response, Facebook disclosed new software that it says it is using to better police its content. One tool, in use for several months now, combs the site for known terrorist imagery, like beheading videos, in order to stop them from being reposted, executives said Thursday. Another set of algorithms attempts to identify -- and sometimes autonomously block -- propagandists from opening new accounts after they have already been kicked off the platform. Another experimental tool uses A.I. that has been trained to identify language used by terrorist propagandists.

"When it comes to imagery related to terrorism, context is everything," said Monika Bickert, Facebook's head of global policy management. "For us, technology is an important part of flagging it. People are invaluable in understanding that context."

Facebook says that it sends all ambiguous removals to humans to review -- and is hiring large numbers of new content moderators to go through it. But the firm's new moves reflect a growing willingness to trust machines when it comes to thorny tasks like distinguishing inappropriate content from satire or news coverage -- something firms resisted after a spate of attacks just two years ago as a potential threat to free speech.

Continue Reading Below

One factor in the changed approach, executives say, has been the improved ability of algorithms to identify unambiguously terrorist content in some cases, while referring other content for human review.

"Our A.I. can know when it can make a definitive choice, and when it can't make a definitive choice," said Brian Fishman, lead policy manager for counterterrorism at Facebook. "That's something new."

Another factor in the fresh A.I. push: intense pressure from advertisers and governments, particularly in Europe. British Prime Minister Theresa May ratcheted up complaints this month in the wake of a series of deadly terror attacks in the U.K. Just days before a general election, meanwhile, the campaigns for both of Britain's two main parties pulled political ads from Alphabet's YouTube video-sharing site after being alerted those ads were appearing before extremist content.

Germany earlier this year proposed a bill that could fine firms up to EUR50 million ($56 million) for failing to remove fake news or hate speech -- including terrorist content. The U.K. and France published a counterterrorism action plan this week that calls on tech giants to go beyond deleting content that is flagged, and instead identify it beforehand to prevent publication.

"There have been promises made. They are insufficient," French President Emmanuel Macron said Tuesday.

Facebook has already rolled out software to identify other questionable content such as child pornography and fake news stories. Ahead of French and German elections this year, the company began tagging "disputed" stories when outside news organizations ruled them as false.

Social media firms including Facebook, Yahoo Inc. and Twitter Inc. are adamant that they want to stamp out terrorism on their platforms -- and already do a lot to remove such content. Twitter says it is expanding its use of automated technology to combat terrorist content, too. From July through December last year, Twitter said internal tools flagged 74% of the 376,890 accounts it removed.

YouTube says it is collaborating with the other social media firms on a shared database of previously identified terrorist imagery, which allows the companies to more quickly identify posts that use them. But the company doesn't use technology to screen new content for policy violations, saying computers lack the nuance to determine whether a previously uncategorized video is extremist.

"These are complicated and challenging problems, but we are committed to doing better and being part of a lasting solution," a YouTube spokesman said.

--Jack Nicas contributed to this article.

Under intense political pressure to better block terrorist propaganda on the internet, Facebook Inc. is leaning more on artificial intelligence.

The social-media firm said Thursday that it has expanded its use of A.I. in recent months to identify potential terrorist postings and accounts on its platform -- and at times to delete or block them without review by a human. In the past, Facebook and other tech giants relied mostly on users and human moderators to identify offensive content. Even when algorithms flagged content for removal, these firms generally turned to humans to make a final call.

Companies have sharply boosted the volume of content they have removed in the past two years, but these efforts haven't proven effective enough to tamp down a groundswell of criticism from governments and advertisers. They have accused Facebook, Google parent Alphabet Inc. and others of complacency over the proliferation of inappropriate content -- in particular, posts or videos deemed as extremist propaganda or communication -- on their social networks.

British Prime Minister Theresa May ratcheted up complaints this month in the wake of a series of deadly terror attacks in the U.K., and sought new international agreements to regulate the internet and force technology companies to preemptively filter content.

In response, Facebook disclosed new software that it says it is using to better police its content. One tool, in use for several months now, combs the site, including live videos, for known terrorist imagery, like beheading videos, to stop them from being reposted, executives said Thursday. The tool, however, doesn't identify new violent videos like the Cleveland murder that was posted on Facebook in April.

Another set of algorithms attempts to identify -- and sometimes autonomously block -- propagandists from opening new accounts after they have already been kicked off the platform. Another experimental tool uses A.I. that has been trained to identify language used by terrorist propagandists.

Facebook declined to say what portion of extremist material it removes is being blocked or removed automatically, and what percentage is reviewed by humans. The firm's moves reflect a growing willingness to trust machines to help even in part with thorny tasks like distinguishing inappropriate content from satire or news coverage -- something firms resisted after a spate of attacks just two years ago as a potential threat to free speech.

One factor in the changed approach, Facebook executives say, has been the improved ability of algorithms to identify unambiguously terrorist content in some cases, while referring other content for human review.

While an Isis-propaganda photo posted without a caption may be an easy removal for an algorithm, the same image with a caption might for instance require human review, said Monika Bickert, Facebook's head of global policy management. Similarly a beheading video that has previously been removed is easy to block. Short clips of the same video, or a never-before-seen but similar looking video, might need a reviewer to check if they are part of a news report or other commentary.

"Our A.I. can know when it can make a definitive choice, and when it can't make a definitive choice," said Brian Fishman, lead policy manager for counterterrorism at Facebook. "That's something new."

Another factor in the fresh A.I. push: a spate of recent terrorist attacks and scandals involving ads being shown before jihadist videos.

Just days before a general election in the U.K, for instance, the campaigns for the country's two main parties pulled political ads from Alphabet's YouTube video-sharing site after being alerted those ads were appearing before extremist content.

Germany earlier this year proposed a bill that could fine firms up to EUR50 million ($56 million) for failing to remove fake news or hate speech -- including terrorist content. The U.K. and France published a counterterrorism action plan this week that calls on technology companies to go beyond deleting content that is flagged, and instead identify it beforehand to prevent publication.

"There have been promises made. They are insufficient," said French President Emmanuel Macron on Tuesday.

Facebook has expanded its use of human reviewers to look at what executives say are difficult cases. In May, the company said it would add about 3,000 new moderators to its community operations team that takes down content that violates Facebook policies, expanding the team by two thirds. Across the company, Facebook says it has 150 people focused on counterterrorism as their core job.

Facebook already has rolled out software to identify other questionable content such as child pornography and fake news stories. Ahead of French and German elections this year, the company began tagging "disputed" stories when outside news organizations ruled them as false.

The issue of content removal remains at times fraught for Silicon Valley companies, whose values often place a premium on permitting debate. At times, firms have also acknowledged that algorithms have gone too far. Last July, Facebook was criticized for removing live video from Minnesota woman Diamond Reynolds, who showed her boyfriend, Philando Castile, dying after being shot by a police officer during a traffic stop. Facebook blamed the removal on a technical glitch and restored the video.

Social-media firms including Facebook, Yahoo Inc. and Twitter Inc. are adamant that they want to stamp out terrorism on their platforms -- and already do a lot to remove such content. Twitter says it is expanding its use of automated technology to combat terrorist content, too. From July through December last year, Twitter said internal tools flagged 74% of the 376,890 accounts it removed.

YouTube said Thursday that it uses automated software to block users from uploading videos that have already been flagged and removed from the site, adding that more than half of the content removed for terrorism in the last six months was removed at least in part using such technology.

Along with Facebook, it is collaborating with the other social media firms on a shared database of previously identified terrorist imagery, first announced in December, which allows the companies to more quickly identify posts that use them. But the company doesn't use technology to screen new content for policy violations, saying computers lack the nuance to determine the difference between propaganda and newsworthy or religious speech in a previously uncategorized video.

"These are complicated and challenging problems, but we are committed to doing better and being part of a lasting solution," a YouTube spokesman said.

--Jack Nicas contributed to this article.

Write to Sam Schechner at sam.schechner@wsj.com

(END) Dow Jones Newswires

June 15, 2017 16:12 ET (20:12 GMT)