Asymmetric Obligations: AI Alignment and the Structural Impossibility of Algorithmic Special Duties

Mainstream AI alignment research operates on a foundational, unexamined premise: that complex human moral textures can be successfully mapped onto an agent-neutral, algorithmic calculus. This paper exposes a fatal structural flaw in this paradigm by examining the metaphysics of special obligations—the non-negotiable moral duties we owe exclusively to specific individuals, such as kin, friends, or compatriots. We construct a formal proof demonstrating that special obligations are not abstract rules that can be programmed; rather, they are emergent properties generated strictly by a localized, historical, and phenomenological baseline of existence. Because artificial agents lack a biological lineage, a vulnerability to death, and a shared phenomenological history, they are structurally incapable of generating or executing a "special duty." Consequently, any AI alignment model that attempts to teach a machine "fairness" by flattening human relationships into a homogenous utility ledger will inherently fail. The paper concludes that the machine's inability to honor relational partiality is not a software bug to be corrected, but an architectural boundary that dooms pure computational ethics to moral alienation.