|
Title
|
|
|
|
Explaining prediction models to address ethical issues in business and society
| |
|
Author
|
|
|
|
| |
|
Abstract
|
|
|
|
| The field of artificial intelligence (AI) has experienced explosive growth in recent years, with applications ranging from medical diagnosis to financial forecasting. However, as these technologies become increasingly integrated into decision-making processes, it is crucial that we also consider the ethical implications of their use. In particular, the transparency, fairness and privacy of AI systems are major concerns, as these systems can have far-reaching impacts on individuals and society. In this PhD thesis, I focus on the ethics of explainable AI (XAI). XAI refers to the development of techniques that are able to provide human-understandable explanations for AI models. My research explores the importance of explainability in the context of ethical decision-making and investigates both opportunities and challenges that arise from the use of Explainable AI. In this PhD thesis, I categorize my research contributions into three pillars: Transparency, Fairness and Privacy. Within the transparency pillar, I study the trade-off between transparency and performance of machine learning models, and investigate the manipulation issues that Explainable AI techniques can induce. Next, within the fairness pillar, I demonstrate how XAI techniques can be used to measure discrimination in machine learning models, and I discuss the opaqueness surrounding the impact of bias mitigation methods. Within the final pillar of privacy, I analyse the privacy issues of XAI techniques, and conduct an applied study to show the trade-off between privacy and personalization on a dataset of Facebook likes. |
| |
|
Language
|
|
|
|
English
| |
|
Publication
|
|
|
|
Antwerp
:
University of Antwerp, Faculty of Business and Economics
,
2024
| |
|
DOI
|
|
|
|
10.63028/10067/2079070151162165141
| |
|
Volume/pages
|
|
|
|
xxx, 210 p.
| |
|
Note
|
|
|
|
:
Martens, David [Supervisor]
:
Sörensen, Kenneth [Supervisor]
| |
|
Full text (open access)
|
|
|
|
| |
|