Frances Haugen’s complaint has opened up documents mentioning the lack of classifiers or automated detection of hate content in Hindi and Bengali on Facebook. Details about BJP IT cell and Bajrang Dal emerge.
A complaint filed by whistleblower Frances Haugens to the US Security And Exchange Commission (SEC) has mentioned internal documents citing ‘Rashtriya Swamsevak Sangh or RSS users, groups and pages’ had promoted fear and hate-mongering content.
Haugens’ counsel in a complaint on how Facebook promotes global division and ethnic violence, available on the internet, has claimed that ‘political sensitivities have prevented the social media app from categorizing or providing designation to the group. The group is apparently the above-mentioned group about whether greater monitoring was required or not.
Frances Haugens has testified in front of the Senate Commerce Committee on Wednesday this week. The complaint documents were posted online by CBS News on Monday evening, a day after the news outlet interviewed Haugen on their “60 Minutes” show on Sunday.
India placed at Tier 0
According to the internal documents cited in the complaint, India is placed at Tier 0, alongside the USA and Brazil, representing the company’s acclaimed ‘political priority. It is good and both bad for Indian users. Another internal document cited in the complaint quotes that both Tier 2 and 3 countries receive zero or no investment for proactive monitoring during their major electoral battles.
It can also be seen that the social media giant only spent 13% of the resources on other countries (India, Italy, and France). Thus, the firm spent almost 87% of its resources on the USA, in a separate document titled ‘Misinformation’. It is a bit grotesque as the USA and Canada comprise 259 million active users, combined as reported during the second quarter of 2021. And as of July 2021, India has 340 million Facebook users, according to Statista. It also revealed that by 2040, Facebook users would increase to 60% of the population in India.
It also supplements the reason behind Facebook’s palpable rate of cracking down hate-mongering content at 3%-5% and a mere 0.2% of violent inciting content.
Lack of Classifiers to spot hate content
In a complaint, an internal document titled ‘Adversarial Harmful Networks – India Case Study’, the company seemed to be visibly aware of the hate-mongering content on their platform. Due to the lack of classifiers in the local languages, reportedly in Hindi and Bengali, most of the anti-Muslim content was never reported or flagged to the system.
The automated systems or algorithms to detect hate speech or content in Facebook are called classifiers. The most prominent reason for this is the loss of profit. The cost of setting up an individual set of classifiers for local languages would outnumber their profit share from the local languages.
The document also quoted an internal source from the BJP party that has used satellite universal modulation adapter Military (SUMA) to promote pro-Hindu messaging. “A BJP IT cell worker shared coordinated messaging instructions to supporters with a copycasta campaign targeting sensitive political tags,” read the document quoting the whistleblower. “… There were a number of dehumanizing messages that compared Muslims to pigs and dogs and misinformation that the Quran asks men to rape female members of their families, ”the internal document said.
One of the documents enclosed in the complaint also confirmed that Facebook did not take action against Bajrang Dal, a Hindu affiliated group on safety and financial grounds. The report has termed them as a dangerous organization.
Hate-promoting content in other platforms and countries
The situation is grim in other countries too. Arabic hate speech detection in Instagram is 6%. The company has close to no members who can understand the Iraqi dialect to review the content.
In Afghanistan, only 0.23% of the hate content is subjected to action, as cited by one of the documents. The abysmal action rate is due to lower awareness and literacy levels, followed by the absence of precise classifiers. “Only 2% of the hate speech is caught proactively. The translation facility provided by Facebook is limited to regions with significant readers.