abstract
This report analyzes the issue of fairness in automated decision-making systems, highlighting its challenges, importance and some potential solutions. The study conducted delves into the various definitions of fairness, some types of bias, and sources of discrimination that can occur in artificial intelligence systems. It also examines state-of-the-art approaches to identifying and mitigating bias, including pre-processing, in-processing and post-processing techniques. In addition, the report proposes a modular library, FairLib, designed to aid in the analysis and mitigation of bias in artificial intelligence systems. The library aims to provide comprehensive tools for bias analysis and intervention. Ultimately, the approaches and methods presented in this work are envisioned to contribute to a future where AI systems operate in an increasingly fairer and transparent manner. This will gradually foster trust and reduce inequalities in various domains, including finance, healthcare, justice, and autonomous systems.
outcomes
