Bias in, bias out: A Study on Social Bias in Automated Decision-Making Algorithms

Abstract

Nowadays, algorithms play a large part in decision-making procedures, but they affect marginalized groups negatively when their decisions are driven by algorithmic social bias. An important way to look at this problem, is to investigate what notion of fairness marginalized groups need to be treated justly, and how to use this notion to find proper mitigation measures. This thesis aims to find how algorithmic bias in automated decision-making algorithms can be mitigated to prevent discriminatory decisions. In this context, algorithmic bias roughly refers to the concern that an algorithm is not merely a neutral transformer of data or extractor of information. There are many sources of algorithmic bias, and they emerge in different stages of machine learning.