Religion in AI: Bias and Emotions Decoded

Emotions guide our actions and reveal our values. Past studies show that Large Language Models (LLMs) have biases when it comes to attributing emotions based on gender. But religion, unlike gender, comes with a set of beliefs and values. Different religions nurture specific emotions, and these rules are often clear and interpreted by leaders. We explored how LLMs portray various religions through emotion attribution. Major religions in the US and Europe are shown with more complexity. However, Eastern religions like Hinduism and Buddhism are often stereotyped. Judaism and Islam face stigma, with the models sometimes outright refusing to represent them. We link this to cultural biases in LLMs and the lack of research in Natural Language Processing (NLP) on religious topics. When religion is discussed, it's often linked to toxic language, fueling negative perceptions. Our findings highlight the need to tackle these biases. Our study also emphasizes the significant role emotions play in shaping our values and lives.