The Epistemological AI Turn: From JTB to KnowledgeS

In this paper, we examine whether large language models (LLMs) can be said to possess knowledge in the sense defined by the Justified True Belief (JTB) framework, and if not, whether any alternative form of knowledge can meaningfully be attributed to them. While LLMs perform impressively across various cognitive tasks—such as summarization, translation, and content generation—they lack belief, justification, and truth-evaluation, which are essential components of the JTB model. We argue that attributing human-like knowledge (in the JTB sense or its variants) to LLMs constitutes a category mistake. Accordingly, LLMs should not be regarded as epistemic agents with human-like capacities, but rather as machine tools that simulate certain functions of human cognition. We acknowledge, however, that when used critically and ethically, these tools can enhance human cognitive performance. To distinguish the capacities of LLMs from human cognitive agency, we introduce the term knowledgeS to denote the structured linguistic outputs produced by LLMs in response to complex cognitive tasks. We refer to the emergence of knowledgeS as marking an “epistemological AI turn.” Finally, we explore the theological implications of AI-generated knowledge. Because LLMs lack conscience and moral sense, they risk detaching knowledge from ethical grounding. Within normative traditions such as Christianity, knowledge is inseparable from moral responsibility rooted in the faith of a religious community. If AI-generated religious texts are mistaken for genuine spiritual insight, they may promote a form of “algorithmic gnosis”—content that mimics sacred language while remaining spiritually hollow. Such developments could erode the moral and spiritual depth of religious expression. As AI systems assume increasingly authoritative roles, society must guard against confusing knowledgeS with genuine, embodied, and ethically accountable knowing, which remains unique to human agency.