Artificial ‘Ulama – Patterns and Biases in Non-Human Islam
The research project examines AI-generated Islamic theology, focusing on the responses generated by Large Language Models (such as GPT-4, Claude, Gemini, Mixtral and Grok) to prompts based on Islamic texts, interpretational themes and concepts. The study combines recent theoretical developments in Islamic studies with issues raised the field of "explainable AI", aiming to critically assess how generative AI systems interpret Islam.
The project expands the scope of Islamic theological study to include interpretations by non-human entities, specifically AI models. It investigates three main areas: Qur'anic interpretation, religious counseling (fatwa), and contentious topics in Islamic faith and practice.
Primary data is gathered through prompting LLMs with a large set of queries on Islam and analyzing the responses via both distant and close reading. The goal is to determine the consistency of AI interpretations across different models and compare them with human interpretations. The project ultimately aims to identify and explain any patterns or biases in the AI responses, which could be a result of the training data, computing processes or alignment procedures.
The two-year project will contribute to the understanding of AI's potential role in future religious discourse. The findings will be shared through academic publications. The datasets generated in the project will be made publicly available for future research.
The project expands the scope of Islamic theological study to include interpretations by non-human entities, specifically AI models. It investigates three main areas: Qur'anic interpretation, religious counseling (fatwa), and contentious topics in Islamic faith and practice.
Primary data is gathered through prompting LLMs with a large set of queries on Islam and analyzing the responses via both distant and close reading. The goal is to determine the consistency of AI interpretations across different models and compare them with human interpretations. The project ultimately aims to identify and explain any patterns or biases in the AI responses, which could be a result of the training data, computing processes or alignment procedures.
The two-year project will contribute to the understanding of AI's potential role in future religious discourse. The findings will be shared through academic publications. The datasets generated in the project will be made publicly available for future research.