Modern societies increasingly rely on algorithmic systems, social platforms, and large language models (LLMs) as de facto shared references amid the erosion of traditional anchors such as religion, nation-states, and local communities. This paper reframes both historical religious systems and contemporary generative models as knowledge operation mechanisms rather than truth containers. It introduces a two-layer knowledge architecture separating implicit normative weights from explicit external references, interprets sectarian schisms as checkpoint conflicts over evaluation authority, and identifies pseudo-scripture canonization as a recurring failure mode in modern reference systems. By approaching polarization, dependency, and radicalization as structural design failures rather than moral or technological flaws, this work proposes preventive principles for preserving plurality, calibration, and human agency in AI-mediated societies.
Kimura Hinano (Tue,) studied this question.