whether such temporary measures would be sufficient to prevent and mitigate harms which Meta’s contentshaping algorithms have helped to enable over a long-term period. The Facebook Papers reveal that Meta has deployed BTG measures in Myanmar on at least one occasion. In the case (referenced in Section 5.3, above) wherein Meta found that 70% of the views gained by a video by hate figure U Wirathu had been actively chained to by Meta’s algorithms, the company did launch “a BTG measure to stop video chaining” in Myanmar in response.324 However, Amnesty International could find no evidence of Meta deploying any similar measures – aimed at reducing algorithmic harms – in Myanmar ahead of the 2017 violence perpetrated against the Rohingya. Despite the repeated warnings and requests which the company received from civil society, Meta continued to deploy its content-shaping algorithms which were actively amplifying harmful content in Myanmar. Amnesty International wrote to Meta and asked what measures the company took prior to 2017 to prevent its algorithmic systems from being used to amplify advocacy of hatred against the Rohingya. Meta responded that the company could not provide information concerning the period leading up to 2017 because the company is “currently engaged in litigation proceedings in relation to related matters”.325 In addition to the possibility of using its algorithmic levers to reduce the direct amplification of harmful content, Meta could have taken far more comprehensive action in respect of content moderation. As Section 5.4, above, details, Meta drastically underinvested in its content moderation for Myanmar, resulting in a situation where most reports of violating content either went ignored or were subject to patently incorrect determinations that the reported content did not violate Facebook’s community standards. Even a well-resourced approach to content moderation, in isolation, would likely not have sufficed to prevent and mitigate these algorithmic harms. This is because content moderation fails to address the root cause of Meta’s algorithmic amplification of harmful content (See Section 6.1 for further analysis). Effective content moderation can nevertheless be a relevant mitigation tactic when it is combined with other measures that seek to directly prevent and mitigate algorithmic amplification. Ultimately, however, Meta’s content moderation operations in Myanmar ahead of the 2017 violence were deeply inadequate. As such, Meta’s use of content moderation cannot be deemed an effective mitigation measure ahead of the 2017 atrocities in Northern Rakhine State. Amnesty international asked Meta what action the company took in Myanmar ahead of and during the crisis in 2017, as well as what steps the company took to address the risks of its “core product mechanics” amplifying advocacy of hatred and other harmful content. Meta did not provide any information because the company is “currently engaged in litigation proceedings in relation to related matters”.326 8.3.1 AN “INESCAPABLE” CONCLUSION: META SUBSTANTIALLY CONTRIBUTED TO HUMAN RIGHTS HARMS SUFFERED BY THE ROHINGYA “When can we say that a company like Facebook is ‘contributing to’ human rights harm?... Unwittingly getting even severely consequential cases wrong once or twice is one thing. But persistent refusal to substantially change what the company does to reduce its role in others’ promotion of social strife and violence makes the attribution of ‘contribution’ inescapable.” John Ruggie, author of the UN Guiding Principles, 2018327 The independent human rights impact assessment for Myanmar commissioned by Meta did not overtly address whether the company had caused, contributed, or was directly linked to adverse human rights impacts against the Rohingya. This is despite the fact the HRIA was conducted in the immediate aftermath 324 The Facebook Papers, “Video Time Spent in Myanmar”, 12 November 2020, on file with Amnesty International, p.16 325 Meta, letter to Amnesty International, 31 May 2022, on file with Amnesty International. 326 Meta, letter to Amnesty International, 31 May 2022, on file with Amnesty International. John G. Ruggie, “Facebook in the Rest of the World,” John F. Kennedy School of Government, Harvard University, November 15, 2018, https://media.business-humanrights.org/media/documents/files/documents/John_Ruggie_Facebook_15_Nov_2018.pdf, cited in Mark Latonero, and Aaina Agarwal, ‘Human Rights Impact Assessments for AI: Learning from Facebook’s Failure in Myanmar’, Carr Center For Human Rights Policy Harvard Kennedy School, 2021. 327 THE SOCIAL ATROCITY META AND THE RIGHT TO REMEDY FOR THE ROHINGYA Amnesty International 62