Congress: “I’m a strong advocate for non-content-based solutions, because those solutions will protect the most vulnerable people in the world,” making reference to the limits of content moderation as a solution.192 Non-content-based solutions include changes to Meta’s content-shaping algorithms. Haugen has specifically suggested that Meta should “get rid of the engagement-based ranking” and advocated for a return to Meta’s chronological news feed.193 The size and scale of Meta, and its presence in almost every country in the world, add to the difficulties associated with effective content moderation. These structural limitations have been compounded by the fact that Meta has, as previously noted, failed to adequately invest in resourcing its operations in the non-English speaking world, enabling algorithmically amplified “hate speech” to run riot in the absence of effective moderation. In October 2020, Meta CEO Mark Zuckerberg testified to the US Senate that: “[O]ver the last few years, we’ve gone from proactively identifying and taking down about 20% of the hate speech on the service, to now, we are proactively identifying, I think it’s about 94% of the hate speech that we ended up taking down [emphasis added], and the vast majority of that before people even have to report it to us.”194 However, internal company records contained in the Facebook Papers reveal this to be a serious mischaracterisation of the reality of Meta’s success at effectively moderating “hate speech” on the platform. An internal document entitled ‘Demoting on Integrity Signals is not enough’, which appears to have been shared internally in July 2019,195 contains the revelation that: “…according to best estimates I have seen, we only take action against approximately 2% of the hate speech on the platform”.196 These 94% and 2% figures are not necessarily contradictory as they refer to different metrics. The 94% figure refers to the amount of “hate speech” proactively identified by Meta’s content moderation algorithms, where the remainder (6%) refers to the amount of “hate speech” the company identifies through other means (such as reporting by users). This metric does not measure or account for “hate speech” not identified by Meta. The 2% figure, on the other hand, refers to the proportion of hate speech actually existing on the Facebook platform which is actioned by Meta – with the remaining 98% representing the total share of “hate speech” which is not addressed by the company. Meta has moved to publicly defend its record in relation to “hate speech” by claiming that the metrics it uses are most relevant when measuring this form of harmful content on their platform.197 However, as some critics have pointed out, “what matters to society is the amount of hate speech that is not removed from the platform” – and not the relative improvement in Meta’s content moderation algorithms at detecting “hate speech”, which Meta’s preferred metric refers to.198 Meta’s efforts to focus on alternative metrics in relation to hate speech has also been criticized as being “emblematic of a long-standing strategy by Facebook to obfuscate and omit information in transparency reports and other public statements”.199 The same document from the Facebook Papers further recognizes the limitations of Meta’s content moderation algorithms at capturing violating content such as hate speech, noting: Karen Hao, “The Facebook whistleblower says its algorithms are dangerous. Here’s why.”, MIT Technology Review, 5 October 2021, technologyreview.com/2021/10/05/1036519/facebook-whistleblower-frances-haugen-algorithms 192 Karen Hao, “The Facebook whistleblower says its algorithms are dangerous. Here’s why.”, MIT Technology Review, 5 October 2021, technologyreview.com/2021/10/05/1036519/facebook-whistleblower-frances-haugen-algorithms 193 Rev, “Tech CEOs Senate Testimony Transcript October 28”, 28 October 2020, rev.com/blog/transcripts/tech-ceos-senate-testimonytranscript-october-28, emphasis added. 194 The document is undated, but the comments on the document are all from July 2019. The Facebook Papers, “Demoting on Integrity Signals is not enough”, undated. 195 196 The Facebook Papers, “Demoting on Integrity Signals is not enough”, undated, emphasis added. Guy Rosen, “Hate Speech Prevalence Has Dropped by Almost 50% on Facebook”, Meta, 17 October 2021, fb.com/news/2021/10/hatespeech-prevalence-dropped-facebook ; Ranking Digital Rights, “Cross-checking Facebook: Five Lies Revealed by Frances Haugen”, 14 October 2021, rankingdigitalrights.org/2021/10/14/cross-checking-facebook-frances-haugen 197 198 Noah Giansiracusa, “Facebook Uses Deceptive Math” (previously cited), 15 October 2021. 199 Ranking Digital Rights, “Cross-checking Facebook” (previously cited). THE SOCIAL ATROCITY META AND THE RIGHT TO REMEDY FOR THE ROHINGYA Amnesty International 40