From Assistance to Eclipse: Telos and the Levels of AI–Person Interaction

A companion article established five axioms grounding the non-delegability of moral judgment. This article extends the inquiry to the full spectrum of AI–person interaction. We propose a taxonomy of five levels — from AI as passive tool (Level 0) through decision-support (Level 1), a multi-dimensional degradation space (Level 2), person-as-subagent (Level 3), to full autonomy (Level 4) — and argue that Level 2 is not a single transitional stage but a space whose coordinates are the statuses of the five axioms of personhood. Our central finding is that the telos — the constitutive purpose for which a technology is deployed — determines whether the same intervention reinforces or undermines the axiomatic system. A cochlear implant conceived to restore perceptual engagement strengthens the axioms; the same device conceived to increase a worker's receptive bandwidth for algorithmic instructions weakens them. We trace degradation trajectories through four domains — algorithmic labour management, healthcare automation, content moderation, and gig work — showing that different orderings of axiom negation converge on the same endpoint: the eclipse of the person as judging subject. The axiomatic system admits no stable equilibria under partial negation: internal dependencies among axioms produce a cascade effect whereby the loss of any single axiom generates pressure toward the loss of the remaining four. From this analysis we derive a Finality Principle: the legitimacy of any AI deployment depends not on the properties of the system but on whether the institutional arrangement is constitutively oriented toward the person as end. This article is the second in a triptych on AI and personhood. The foundational manifesto is available at https://doi.org/10.5281/zenodo.18650485.