Overcoming algorithmic bias

: the role of bias awareness, knowledge, and minority status on human decision-making

Abstract

Algorithms, whether augmenting or automating human decisions, are used by an increasing number of organizations to execute several tasks. Algorithms create value because they can analyse massive volumes of data, explore, and identify nonlinear and counterintuitive relationships, being more consistent with their decisions than humans. However, this decision-making process also has the potential to perpetuate and exacerbate historical and societal biases, since the algorithms are trained on data that reflect human biases. This dissertation reviews the limitations and associated challenges of human and algorithmic decision-making and explores how human decision-making can be leveraged to address and mitigate the harmful effects of algorithmic bias. Particularly, this dissertation explores the effects of being aware of biased decision-making, being knowledgeable in Artificial Intelligence (AI), and belonging to a minority, on human’s ability to perceive and include potentially biasing variables that lead to unfair and discriminatory algorithmic decisions. The results suggest that a simple intervention of bias awareness can be an effective debiasing strategy; being AI knowledgeable can help to perceive the potential for bias, yet it does not necessarily translate to exclude potential sources of algorithmic bias; belonging to a minority may not be an advantageous factor to improve decision-making. Therefore, the solution proposed by this dissertation to overcome algorithmic bias lies in promoting bias awareness as a viable strategy to improve human decision-making.

Date of Award	26 Apr 2021
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Filipa de Almeida (Supervisor)