EXPOSING ISLAMOPHOBIA IN MACHINE LEARNING: A CRITICAL ANALYSIS OF THE EXISTING THEORIES AND BIASES

Abstract

With the proliferation of Artificial Intelligence (AI) across different sectors, ethical challenges, such as bias in machine learning, have raised concerns for the AI models’ performance. Based on the datasets on which AI systems are trained, AI systems exacerbate and mimic prejudices during the training, where the datasets are inherently biased. Hence, resulting in unfair treatment of individuals based on their gender, race, and religious affiliations. In the digital age, the surge of Islamophobia is a global challenge. Islamophobic bias is a negative feeling towards Muslims, stemming from misinformation about Islam and its followers. The researchers have contributed various mitigating strategies and theories to overcome this challenge. This paper critically investigates the following theories evolved in machine learning and their inherent limitations: (1) Algorithmic Bias, (2) Fairness through Unawareness, (3) Disparate Impact, (4) Equalized Odds and Equal Opportunity, (5) Counterfactual Fairness, and (6) Intersectional Fairness.  The research contributes further to mitigating strategies to address the Islamophobic bias in the age of generative AI.