Avaliação de impacto algorítmico: uma abordagem contextual a partir do tripé de precaução, accountability e transparência

Abstract

Artificial intelligence (AI) systems are increasingly present in activities and decisionmaking processes that are part of people's daily lives. With the promise of generating gains in productivity and efficiency, such systems can also impact the exercise of human rights, especially in the face of biases and inaccuracies. Against this backdrop, debates are emerging about tools for identifying and mitigating risks in the development or implementation of AI systems, such as Algorithmic Impact Assessment (AIA). In this context, this dissertation aims to investigate the legal outlines of AIA and analyze its role as a tool for protecting human rights. This is an exploratory study, carried out through bibliographic and documentary analysis. Based on the bibliographic survey, it is understood that the AIA should be structured on the basis of a tripod based on accountability, precaution and transparency. In addition to these parameters, this paper seeks to demonstrate that, in order for it to be an effective risk mitigation instrument, it is necessary to adopt a contextual approach in the development of AIA, which recognizes that the impacts generated by AI systems can vary considerably depending on the social, economic and ethical particularities of the development and implementation context, including in relation to the people and groups affected.