Abstract

This research investigates the relationship between Facebook and hate speech. In doing so, it deconstructs the governance structures of Facebook and analyses the principles and values that underpin Facebook’s schemes of intervention. This thesis argues that, from the perspective of Facebook, hate speech is approached less in terms of its substance and more in terms of a practical problem that needs to be resolved in an operational manner. They therefore conceive of relevant policy in terms of its fit within Facebook’s overall structure of governance, by which we mean the techniques or mechanisms used to internally order the various actors and actions. Theoretically, this thesis frames different approaches to hate speech regulation and adopts an understanding of governance as the means by which to regulate and order behaviours and actions, by using the work of Foucault and Miller and Rose to study governing systems. The research question concerns the ways (including the mechanisms, instruments, features and action sequences and above all the discourses) by which Facebook orders and regulates the creation and circulation of content when it comes to hate speech. The empirical materials upon which the thesis relied in order to identify the parameters of the governmentality of digital hate include: Mark Zuckerberg’s publications (May 2016 to November 2020), Facebook Principles and Values (2009 -2021), Facebook Community Standards (2016 2021), Content Standards (2018 2021), user’s settings (2016-2021), and the Oversight Board (2019-2021). To supplement these materials, the thesis makes use of three in depth interviews with Facebook’s Director of Public Policy, Campaigns and Programs (EMEA) Siobahn Cummiskey in 2016 and 2019 and with Facebook’s Public Policy officer Aibhinn Kelleher in 2017. Finally, the thesis makes use of secondary data that includes internal Facebook training materials for content moderators leaked to Pro-Republica and The Guardian. The key findings show that, while Facebook articulates its hate speech policies following a traditional liberal approach to hate speech, they operationationalise hateful content as a question of user safety. This results in an overall approach that is far removed from questions of social justice and emancipation. Instead, the focus is on procedural enforcement that produces more and more data. This emphasis on data, in turn, feeds into techno solutions relying on artificial intelligence and machine learning tools that are currently used moderately but which are planned as the preferred solution to what is constructed as a technical problem of content regulation.