Algorithmic Discrimination - The Challenge of Unveiling Inequality in Brazil

• Ano de defesa: 2019
• Autor(a) principal: Mattiuzzo, Marcela
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Algorithmic discrimination; Algorithmic governance; Artificial intelligence; Big data; Discriminação algorítmica; Equality; Governança algorítmica; Igualdade; Inteligência artificial
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/2/2134/tde-16072020-174508/

The objective of this work is to provide some clarity on what the role of the Law can be in shedding light upon algorithmic discrimination, as well as how legal instruments could help minimize its risks, with a specific focus on the Brazilian jurisdiction. To do so, it first engages in a debate about what algorithms indeed are, and how the emergence of the data-driven economy, Big Data, and machine learning have leveraged the use of automated systems. Next, it conceptualizes discrimination, and suggesting a typology of algorithmic discrimination that takes statistics into account to provide a rationalization of the debate. It moves on to discussing the path towards enforcing legal norms against discriminatory outcomes running from the use of algorithms. Because legislation specifically aimed at fighting automated systems is still scarce (or application of the current legislation to the problem is contentious), it engages in a debate about the horizontal effects of fundamental rights - given that a relevant part of discriminatory practices occur among private parties, and the most basic defense an individual has against discrimination is the constitutional right to equality. It then analyzes ordinary legislation in three jurisdictions, the United States of America, Germany, and Brazil, that could also be enforced against discriminatory practices running from algorithms, with a special focus on the Brazilian legislation. The legislative debate concludes with the presentation of two concrete cases of algorithmic discrimination, one concerning the unemployment policy in Poland, and the other regarding credit scoring in Brazil. The cases are presented so that the applicability of Brazilian legislation to deal with algorithmic discrimination can be discussed. The final chapter is focused on debating the path forward and what can and should be done by experts, legislators, and policymakers to foster algorithmic innovation without losing sight of its potential for discrimination. It first presents the literature on algorithmic governance and the many proposals for dealing with the problem - dedicating a specific section to the challenges brought about by machine learning - and then sets out an agenda for Brazil.