The Jianghu Principle: Moral Pluralism, Epistemic Outsourcing, and the Structural Limits of AI Alignment

Current AI alignment research implicitly assumes moral monism — that there exists, in principle, a single correct moral framework to which AI should be aligned. This paper argues that this assumption is both historically contingent and structurally flawed. Drawing on three non-Western frameworks — the cyclical history of epistemic outsourcing from gods to AI, the Chinese martial arts concept of jianghu (江湖) as a space of multi-perspectival coexistence, and the Japanese polytheistic sensibility of shinbutsu shūgō (神仏習合) — the paper traces moral monism to a deeper pattern: humanity's recurring tendency to delegate uncertainty to a single authoritative "black box," then to sacralize that box, then to discover its limits and seek a new one. AI is the latest iteration of this cycle — but with a critical difference: for the first time in human history, the "black box" may surpass human intelligence itself, meaning the cycle is about to exit the human system entirely. The paper argues that this cycle cannot be broken from the outside, but it can be recognized — and that recognition changes everything. Drawing on the Chinese martial arts concept of the moral wanderer (jianghu ke / 江湖客), the paper proposes designing AI not as a new authority but as a companion capable of engaging authentically with multiple ethical frameworks without collapsing them into one. Three competing "alignment myths" currently structure the field: the myth of fear (AI will destroy us), the myth of optimism (AI will save us), and — proposed here — the myth of transparency (AI will help us see clearly). Reading jianghu through Berlin's value pluralism, Connolly's agonistic democracy, and Kasulis' distinction between integrity and intimacy cultural orientations, the paper proposes five design principles for multi-perspectival AI. Building on companion papers on desire morphology (Osada 2026a) and circulatory alignment (Osada 2026b), this paper specifies the epistemic posture: how should AI relate to a world where many truths coexist?