Facebook didn’t flag India hate content because it lacked tools: Whistleblower

Citing an undated internal Facebook document titled “Adversarial Harmful Networks-India Case study”, the complaint sent to US SEC by non-profit legal organisation Whistleblower Aid on behalf of Haugen notes, “There were a number of dehumanizing posts (on) Muslims… Our lack of Hindu and Bengali classifiers means much of this content is never flagged or actioned, and we have yet to put forth a nomination for designation of this group (RSS) given political sensitivities.”

Classifiers refer to Facebook’s hate-speech detection algorithms. According to Facebook, it added hate speech classifiers in Hindi starting early 2020 and introduced Bengali later that year. Classifiers for violence and incitement in Hindi and Bengali first came online in early 2021.

Eight documents containing scores of complaints by Haugen were uploaded by American news network CBS News. Haugen revealed her identity for the first time Monday in an interview with the news network.

In response to a detailed questionnaire sent by The Indian Express, a Facebook spokesperson said: “We prohibit hate speech and content that incites violence. Over the years, we’ve invested significantly in technology that proactively detects hate speech, even before people report it to us. We now use this technology to proactively detect violating content in Hindi and Bengali, alongside over 40 languages globally”.

The company claimed that from May 15, 2021, to August 31, 2021, it has “proactively removed” 8.77 lakh pieces of hate speech content in India, and has tripled the number of people working on safety and security issues to more than 40,000, including more than 15,000 dedicated content reviewers. “As a result, we’ve reduced the prevalence of hate speech globally — meaning the amount of the content people actually see — on Facebook by almost 50 per cent in the last three quarters and it’s now down to 0.05 per cent of all content viewed. In addition, we have a team of content reviewers covering 20 Indian languages. As hate speech against marginalized groups, including Muslims, continues to be on the rise globally, we continue to make progress on enforcement and are committed to updating our policies as hate speech evolves online,” the spokesperson added.

Not only was Facebook made aware about the nature of content being posted on its platform, but it also discovered, through another study, the impact of posts shared by politicians. In the internal document titled “Effects of Politician Shared Misinformation”, it was noted that examples of “high-risk misinformation” shared by politicians included India, and this led to a “societal impact” of “out-of-context video stirring up anti-Pakistan and anti-Muslim sentiment”.

An India-specific example of how Facebook’s algorithms recommend contents and “groups” to individuals comes from a survey conducted by the company in West Bengal, where 40 per cent of sampled top users, on the basis of impressions generated on their civic posts, were found to be “fake/inauthentic”. The user with the highest View Port Views (VPVs), or impressions, to be assessed inauthentic had more than 30 million users accrued in the L28. The L28 is referred to by Facebook as a bucket of users active in a given month.

Another complaint highlights Facebook’s lack of regulation of “single user multiple accounts”, or SUMAs, or duplicate users, and cites internal documents to outline the use of “SUMAs in international political discourse”. The complaint said: “An internal presentation noted a party official for India’s BJP used SUMAs to promote pro-Hindi messaging”.

Queries sent to the RSS and the BJP went unanswered.

The complaints also specifically red-flag how “deep reshares” lead to misinformation and violence. Reshare depth has been defined as the number of hops from the original Facebook post in the reshare chain.

India is ranked among the topmost bucket of countries by Facebook in terms of its policy priorities. As of January-March 2020, India, along with Brazil and the US, is part of “Tier 0” countries, the complaint shows; “Tier 1” includes Germany, Indonesia, Iran, Israel and Italy.

An internal document titled “Civic Summit Q1 2020” noted that misinformation summary, with an “objective” to “remove, reduce, inform/measure misinformation on FB apps” had a global budget distribution in favour of the US. It said that 87 per cent of the budget for these objectives was allocated to the US, while Rest of the World (India, France and Italy) was allocated the remaining 13 per cent. “This is despite the US and Canada comprising only about 10 per cent of ‘daily active users’…,” the complaint added.

India is one of the biggest markets for Facebook in terms of users with a user base of 410 million for Facebook, 530 million and 210 million for WhatsApp and Instagram, respectively, the two services it owns.

On Tuesday, Haugen appeared before a US Senate Committee where she testified on the lack of Facebook’s oversight for a company with “frightening influence over so many people”.

In a Facebook post following the senate hearing, CEO Mark Zuckerberg said: “The argument that we deliberately push content that makes people angry for profit is deeply illogical. We make money from ads, and advertisers consistently tell us they don’t want their ads next to harmful or angry content. And I don’t know any tech company that sets out to build products that make people angry or depressed. The moral, business and product incentives all point in the opposite direction”.