AI Hallucination vs. Tawthīq: A Critical Epistemological Review of the Reliability of Large Language Models (LLMs) in Fatwa Generation

Authors

  • Rully Agung Firmansyah Ahmad Dahlan Institute Probolinggo
  • Benny Prasetiya Institut Ahmad Dahlan Probolinggo

Keywords:

AI hallucination, Tawthīq, islamic epistemology, large language models (llms), sanad, AI governance

Abstract

The revolution of Generative Artificial Intelligence (AI) and Large Language Models (LLMs) has shifted the paradigm of Islamic knowledge acquisition from a traditional chain-of-transmission (sanad) and verification (tawthīq) framework to algorithmic probabilities. This reliance poses a severe epistemological threat known as "AI hallucination", output that appears fluent and authoritative yet is factually incorrect, manifesting as hallucinations in isnad, matn, takhrij, fatwas, and manhaj. This study aims to analyze the computational mechanics of AI hallucination, evaluate the epistemic authority of LLMs based on Hadith and Usul al-Fiqh criteria for news acceptance (khabar), and formulate an Islamic AI governance framework. Employing a Critical Epistemological Review methodology, this research contrasts the probabilistic mechanics of LLMs ("stochastic parrots") with the tawthīq principles of the Salafus Shalih. The findings reveal that LLMs fundamentally operate on token statistical probabilities without semantic comprehension (fahm), moral integrity (‘adālah), legal memory precision (ḍabṭ), or continuous chains of transmission (sanad muttaṣil). Epistemologically, AI is classified as majhūl al-ḥāl wa al-‘ain, meaning it cannot serve as a binding proof (ḥujjah) or a mufti, but strictly as a search tool (adāt al-baḥth). As a solution, this study formulates the "Digital Tawthīq" framework, integrating Hadith criticism methodology into modern AI architecture through ten technical components, including Closed Corpus Constraints, Retrieval-Augmented Generation (RAG) with digital sanad metadata, Knowledge Graph Sanad, and Human-in-the-Loop scholar verification. This framework repositions AI as a supportive instrument while safeguarding the purity of Islamic revelation in the digital era.

Downloads

Published

2026-11-10