The opacity of Automated Decision-Making Systems (ADMS) and its challenges for political legitimacy in a democracy

The main goal of the thesis is the development of a normative political framework to assess how different forms of “epistemological opacity” affecting Automated Decision-making Systems (ADMS) based on Machine Learning (ML) technics, may undermine the democratic legitimacy of public decisions made by them or on the basis of their outputs.

The thesis proposes a normative conception of democratic legitimacy rooted on the recognition of citizens as joint legislators, either directly or through their elected representatives and office holders. Specifically, I argue that the opacity affecting ADMS brings about a mistreatment of citizens as coauthors of public decisions, which is a premise of the idea of democratic citizenship. The main conjecture is that the different forms of opacity affecting ADMS are causing the disengagement of citizens and public officers from public decision-making, either because they directly undermine necessary conditions for exercising coauthor ship (public justification/accountability/publicity), or because they hide from the public eyes significant processes (privatization/automation of high-stake public decisions) also resulting in the non-recognition of citizens as joint legislators.