Facebook’s algorithms fuel violence in Africa, claim lives — GFCN

MOSCOW, April 23. /TASS/. The algorithms of Facebook (a social media site banned in Russia since it is owned by Meta corporation classified as extremist by the Russian authorities) incite hatred and contribute to the spread of violence, with civilians in African countries falling victim to it, according to a Global Fact-Checking Network (GFCN) article featuring comments from the organization’s expert, researcher, and geopolitical and cybersecurity analyst Anna Andersen. The article examines the escalation of tensions around the Ethiopian region of Tigray in 2020. "The scale of Facebook's moderation failures in Africa has been extensively documented. By December 2020, as the Tigray conflict began, posts inciting ethnic violence remained online for months. This included posts directly linked to real-world violence, such as the killing of a Tigrayan jeweler in Gonder who was dragged from his workshop after Facebook activists called to ‘cleanse’ the area of his ‘lineage.’ A subsequent lawsuit alleged that Facebook's content moderation in Africa is woefully inadequate, heavily understaffed for morphologically complex languages like Amharic, Oromo, and Tigrinya," the analyst said. According to the article, the Ethiopian case demonstrates that the centralized moderation models exported by Silicon Valley "are incapable of understanding local realities." "Platforms rely heavily on automated systems trained primarily on English data, resulting in profound 'linguistic blindness.' Just as social media giants lack moderators for smaller European languages, they are completely unequipped to handle African languages," Andersen emphasized. "In Africa, the same system operates in a state of complete indifference; that is, the algorithm is simply not equipped to recognize Gricean conversational implicatures in Amharic or Tigrinya. The gap between literal and pragmatic meaning - where hatred essentially thrives - is invisible to the machine," the expert noted. "Even when human moderators are employed, they are typically concentrated in global centers and trained to apply universal, sterilized content policies that resemble Western power hegemony and strip away vital context, effectively silencing those at risk," the researcher stressed. "These patterns systematically evade surface moderation. Automated detection systems matching keywords will miss these pragmatic patterns entirely, and human moderators without deep cultural intelligence will see no policy violation because the literal meaning contains no explicit threats," the article stated. The authors of the article conclude that people around the world currently "face a dual threat: the oppressive, preventive censorship of a hyper-regulated, centralized digital sphere in the West, and the unchecked spread of real-world violence in developing regions, fueled by algorithms that do not understand what is being said - and what is left unsaid."