The Epistemological AI Turn: From JTB to Knowledge*

Abstract

This paper critically examines whether Large Language Models (LLMs) possess knowledge in the sense of the Justified True Belief (JTB) framework. While LLMs excel at tasks like summarization, translation, and content generation, they lack belief, justification, and truth-evaluation—key components of JTB. Attributing human-like knowledge to LLMs is a category mistake. To clarify this distinction, we introduce knowledge⁎: a term for the structured linguistic outputs of LLMs, which simulate cognition without understanding. LLMs are not epistemic agents but tools that can augment human thought when used critically and ethically. This “epistemological AI turn” calls for reevaluating what counts as knowledge in AI systems.We also consider theological implications of LLM generated knowledge. LLMs, lacking conscience or moral sense, risk detaching knowledge from ethical grounding. In normative traditions like Christianity, knowledge is inseparable from moral responsibility. If AI-generated religious texts are mistaken for genuine insight, they may foster a form of “algorithmic gnosis”—stylized but hollow content that mimics sacred language without meaning. Such use could erode the spiritual depth and moral seriousness of religious expression. As AI takes on authoritative roles, society must guard against confusing knowledge* with true, embodied, ethically accountable knowing.