Mark Zuckerberg, the Facebook CEO, announced last week the platform will no longer algorithmically recommend political groups to users in an attempt to “turn down the temperature” on online divisiveness.
But experts say such policies are difficult to enforce, much less quantify, and the toxic legacy of the Groups feature and the algorithmic incentives promoting it will be difficult to erase.
“This is like putting a Band-Aid on a gaping wound,” said Jessica J González, the co-founder of anti-hate speech group Change the Terms. “It doesn’t do enough to combat the long history of abuse that’s been allowed to fester on Facebook.”
Groups – a place to create ‘meaningful social infrastructure’
Facebook launched Groups, a feature that allows people with shared interests to communicate on closed forums, in 2010, but began to make a more concerted effort to promote the feature around 2017 after the Cambridge Analytica scandal cast a shadow on the platform’s Newsfeed.
In a long blogpost in 2017 February called Building Global Community, Zuckerberg argued there was “a real opportunity” through groups to create “meaningful social infrastructure in our lives”.
He added: “More than one billion people are active members of Facebook groups, but most don’t seek out groups on their own – friends send invites or Facebook suggests them. If we can improve our suggestions and help connect one billion people with meaningful communities, that can strengthen our social fabric.”
After growing its group suggestions and advertising the feature extensively – including during a 60-second spot in the 2020 Super Bowl – Facebook did see a rise in use. In February 2017 there were 100 million people on the platform who were in groups they considered “meaningful”. Today, that number is up to more than 600 million.
That fast rise, however, came with little oversight and proved messy. In shifting its focus to Groups, Facebook began to rely more heavily on unpaid moderators to police hate speech on the platform. Groups proved a more private place to speak, for conspiracy theories to proliferate, and for some users to organize real-life violence – all with little oversight from outside experts or moderators.
Facebook in 2020 introduced a number of new rules to “keep Facebook groups safe”, including new consequences for individuals who violate rules and increased responsibility given to admins of groups to keep users in line. The company says it has hired 35,000 people to address safety on Facebook, including engineers, moderators and subject matter experts, and invested in AI technology to spot posts that violate it guidelines.
“We apply the same rules to Groups that we apply to every other form of content across the platform,” a Facebook company spokesperson said. “When we find Groups breaking our rules we take action – from reducing their reach to removing them from recommendations, to taking them down entirely. Over the years we have invested in new tools and AI to find and remove harmful content and developed new policies to combat threats and abuse.”
Researchers have long complained that little is shared publicly regarding how, exactly, Facebook algorithms work, what is being shared privately on the platform, and what information Facebook collects on users. The increased popularity of Groups made it even more difficult to keep track of activity on the platform.
“It is a black box,” said González regarding Facebook policy on Groups. “This is why many of us have been calling for years for greater transparency about their content moderation and enforcement standards. ”
Meanwhile, the platform’s algorithmic recommendations sucked users further down the rabbit hole. Little is known about exactly how Facebook algorithms work, but it is clear the platform recommends users join similar groups to ones they are already in based on keywords and shared interests. Facebook’s own researchers found that “64% of all extremist group joins are due to our recommendation tools”, an internal report in 2016 found.
“Facebook has let white supremacists organize and conspiracy theorists organize all over its platform and has failed to contain that problem,” González said. “In fact it has significantly contributed to the spread of that problem through its recommendation system.”
‘We need to do something to stop these conversations’
Facebook’s own research showed that algorithmic recommendations of groups may have contributed to the rise of violence and extremism.
On Sunday, the Wall Street Journal reported that internal documents showed executives were aware of risks posed by groups and were warned repeatedly by researchers to address them. In one presentation in 2020 August, researchers said roughly “70% of the top 100 most active US Civic Groups are considered non-recommendable for issues such as hate, misinfo, bullying and harassment”.
“We need to do something to stop these conversations from happening and growing as quickly as they do,” the researchers wrote, according to the Wall Street Journal, and suggested taking measures to slow the growth of Groups until more could be done to address the issues.
Several months later, Facebook halted algorithmic recommendations for political groups ahead of the US elections – a move that has been extended indefinitely with the policy announced last week. The change seemed to be motivated by the 6 January insurrection, which the FBI found had been tied to organizing on Facebook.
In response to the story in the Wall Street Journal, Facebook’s vice-president of integrity, who oversees content moderation policies on the platform, said the problems were indicative of emerging threats rather than inability to address long-term problems. “If you’d have looked at Groups several years ago, you might not have seen the same set of behaviors,” he said.
But researchers say the use of Groups to organize and radicalize users is an old problem. Facebook groups had been tied to a number of harmful incidents and movements long before January’s violence.
“Political groups on Facebook have always advantaged the fringe, and the outsiders,” said Joan Donovan, a lead researcher at Data and Society who studies the rise of hate speech on Facebook. “It’s really about reinforcement – the algorithm learns what you’ve clicked on and what you like and it tries to reinforce those behaviors. The groups become centers of coordination.”
Facebook was criticized for its inability to police terror groups such as the Islamic State and al-Qaida using it as early as 2016. It was used extensively in organizing of the Unite the Right Rally in Charlottesville in 2019, where white nationalists and neo-Nazis violently marched. Militarized groups including Proud Boys, Boogaloo Bois and militia groups all organized, promoted and grew their ranks on Facebook. In 2020 officials arrested men who had planned a violent kidnapping of the Michigan governor, Gretchen Whitmer, on Facebook. A 17-year-old in Illinois shot three people, killing two, in a protest organized on Facebook.
These same algorithms have allowed the anti-vaccine movement to thrive on Facebook, with hundreds of groups amassing hundreds of thousands of members over the years. A Guardian report in 2019 found the majority of search results for the term “vaccination” were anti-vaccine, led by two misinformation groups, “Stop Mandatory Vaccination” and “Vaccination Re-education Discussion Forum” with more than 140,000 members each. These groups were ultimately tied to harassment campaigns against doctors who support vaccines.
In September 2020, Facebook stopped health groups from being algorithmically recommended to put a stop to such misinformation issues. It also has added other rules to stop the spread of misinformation, including banning users from creating a new group if an existing group they had administrated is banned.
The origin of the QAnon movement has been traced to a post on a message board in 2017. By the time Facebook banned content related to the movement in 2020, a Guardian report had exposed that Facebook groups dedicated to the dangerous conspiracy theory QAnon were spreading on the platform at a rapid pace, with thousands of groups and millions of members.
‘The calm before the storm’
Zuckerberg has said in 2020 the company had removed more than 1m groups in the past year, but experts say the action coupled with the new policy on group recommendations are falling short.
The platform promised to stop recommending political groups to users ahead of the elections in November and then victoriously claimed to have halved political group recommendations. But a report from the Markup showed that 12 groups among the top 100 groups recommended to users in its Citizen Browser project, which tracks links and group recommendations served to a nationwide panel of Facebook users, were political in nature.
Indeed, the Stop the Steal groups that emerged to cast doubt on the results of the election and ultimately led to the 6 January violent insurrection amassed hundreds of thousands of followers – all while Facebook’s algorithmic recommendations of political groups were paused. Many researchers also worry that legitimate organizing groups will be swept up in Facebook’s actions against partisan political groups and extremism.
“I don’t have a whole lot of confidence that they’re going to be able to actually sort out what a political group is or isn’t,” said Heidi Beirich of the Southern Poverty Law Center Facebook’s Real Oversight Board, a group of academics and watchdogs criticizing Facebook’s content moderation policies.
“They have allowed QAnon, militias, and other groups proliferate so long, remnants of these movements remain all over the platform,” she added. “I don’t think this is something they are going to be able to sort out overnight.”
“It doesn’t actually take a mass movement, or a massive a sea of bodies, to do the kind of work on the internet that allows for small groups to have an outsized impact on the public conversation,” added Donovan. “This is the calm before the storm.”