Before Facebook shut down the fast-growing “stop theft” Facebook group on Thursday, the forum featured calls on members to prepare weapons before President Donald Trump loses the bid to stay in the White House.
When Reuters and other news organizations disabled the organization following reports, Facebook cited the forum’s efforts to legitimize the election process and “would worry about the violent appeals of certain members.”
On the eve of the Facebook Groups election, such remarks are not uncommon. This is a key driver of participation in the world’s largest social network, but it is not always treated the same.
At the request of Reuters, the digital intelligence company CounterAction conducted a survey of the Facebook group in the United States between September and October, and found that thousands of politically inclined public organizations have thousands of opinions. Violent.
Variants of twenty phrases that may be related to the violent appeal, such as “lock and step up the load” and “we need a civil war”, and appeared in about 41,000 instances of election results in two public Facebook groups in the United States. Situation for a month.
According to CounterAction, other phrases such as “shoot them” and “kill everyone” were used at least 7,345 times and 1,415 times in public groups, respectively. “Hang him” appeared 8,132 times. “It’s time to read the Gunners, guys,” read a comment.
A Facebook spokesperson said that Facebook is reviewing CounterAction’s findings. Reuters shared the findings with the company and will take action to implement the policy of “reducing real-world harm and civil unrest including groups”.
The company declined to disclose whether the examples shared by Reuters violated its rules, or refused to disclose the route of bribery in order to determine the term “inciting or contributing to serious violence,” which is a reason for deletion under its policy.
Prosecutors have linked several sabotaged militia conspiracies this year to Facebook groups, including plans to attack protesters in the “Black Life Incident” in Las Vegas and plans to kidnap the governor of Michigan.
To address concerns, Facebook announced a series of policy reforms from the summer to curb “military social movements” including the American militia, the Boogaloo network and the QAnon conspiracy movement.
It said that since August, 14,200 groups have been deleted based on these changes.
Zuckerberg said that as pressure on the company intensifies before the election, Facebook will suspend its recommendations to political groups and new groups, although this measure has not prevented the “stop stealing” groups from swelling to more than 24 hours in less than 24 hours. 365,000 members.
Since CEO Mark Zuckerberg listed Groups as a strategic priority in 2017, Facebook has been actively promoting the development of Groups, saying it will encourage more “meaningful connections.” , And included the business in the Super Bowl advertisement this year.
Last month, even though the civil rights organization warned that the product had become a hotbed of extremism and misinformation, it still stepped up its propaganda in news feeds and search engine results.
Anyone on Facebook can view, search and join public groups. Online forums also provide private options to hide the existence of posts or forums, even if there are thousands of members in the online forums.
Facebook stated that it relies heavily on artificial intelligence to monitor forums, especially private groups, which rarely issue user reports of users reporting bad behavior, because their members tend to be like-minded to flag those who may instigate violence against human content reviewers. Post.
Although the use of violent language does not always equate to an actionable threat, Matthew Hindman, a machine learning and media scholar at George Washington University, reviewed the results and said that Facebook’s artificial intelligence should be able to choose General terminology for review.
He said: “If you still find thousands of’problem solving’ and’catch the rope’ cases, it means that you have encountered a systemic problem. Modern machine learning systems will never miss this situation.”
© Thomson Reuters 2020