“Last October, Bangladesh was rocked by one of its worst bouts of violence in years when Muslims across the country attacked hundreds of Hindu homes and temples, killing at least six people and injuring dozens of others.

“Since its founding in 2004, Facebook has been aware of the potential dangers of its platform. Its terms of service have always reserved the right to remove threatening speech, and as harmful content increasingly plagued the platform, it regularly updated its policies to better define its guidelines and enforcement approaches. For years, Facebook’s corporate owner, now known as Meta, has funneled billions of dollars into improving ‘safety and security’ measures across all of its products, including WhatsApp and Instagram.

“Facebook’s ad hoc and piecemeal approach to content moderation is failing in Bangladesh and around the world, with stark consequences.

“In Bangladesh, allegations of blasphemy, often made by ordinary citizens, are particularly potent at mobilizing the country’s Muslim majority against minority communities. Long before Facebook was developed, these hate campaigns spread through street movements and word of mouth, often to violent results. In 1999, militants brutally attacked Shamsur Rahman, a poet and secular activist, over his controversial views.

“The attack set off what has since become a dire annual trend. In 2013 and 2014, Muslim rioters attacked Hindu villages after a Hindu man was falsely accused of demeaning the Prophet Mohammed on Facebook; two years after the latter incident, rioters vandalized at least 15 Hindu temples. In 2017, a 20,000-person mob set fire to a Hindu village; an assault on a Hindu community in 2019 killed four people and injured 50 others. Earlier in 2021, a Hindu man’s car and home were vandalized after he allegedly criticized the Prophet Mohammed in a Facebook conversation.

“To combat hate speech, Facebook relies on a multipronged strategy that incorporates algorithms, user reporting, and internal content moderators. The company’s algorithms are designed to identify and remove harmful content, while its automated system reviews user-reported posts. When the system fails to draw a conclusion, the post is sent to human content moderators, many of whom face substantial language barriers and lack the bandwidth to review a high volume of content.

“One key issue is that Facebook has largely focused its moderation efforts in English-speaking Western countries, while neglecting other regions where hate speech can prove to be even more dangerous. To better curb violence in these countries, the platform must ensure that sufficient resources are devoted to them—including by bolstering staff, adding localized resources, expanding its hate speech training to more languages, and improving product designers’ knowledge of country-specific context, culture, and trends. Governments can push Facebook to enact these changes by employing key regulatory tools, such as the European Union’s Digital Services Act, that increase transparency and accountability.

“With these efforts in place, Facebook should consider developing regional or country-specific guidelines on high-risk hate speech. In Bangladesh, such policies would have helped moderators understand that provocative content about Islam is more likely to incite violence than other posts.

“Since 2018, Facebook’s ‘at-risk countries’ team has monitored and removed hateful content in countries it considers to be most vulnerable to violence, particularly during election seasons. Insufficient funding, however, has constrained the number of places that the team can focus on. Expanded resourcing would allow Facebook to broaden its attention to countries like Bangladesh, which experience high religious tensions and are especially susceptible to communal violence.

“As Facebook continues to grapple with hate speech, disinformation, and malicious content, religious violence will only continue to afflict Bangladesh. And with more than 47 million Bangladeshis now on the platform, the stakes of inaction are too high to ignore…”

Up ArrowTop