The Epistemic Alibi: Opacity, Delegation, and the Structural Erosion of Dignity in AI-Mediated Decisions. Article 10

When AI-generated judgment operates at a complexity that systematically exceeds the human subject's capacity to verify it, something beyond accountability is at stake. The Axiomatic Theory of Tragic Subjecthood (ATTS) has established that delegation does not sever the ontological connection between a normative subject and the space of losses opened by its decision (T1), and that concealing this connection undermines legitimacy (S6, S7). These results concern the visibility of responsibility. This article addresses a different problem: not whether the bearer is visible, but whether the decision can remain the bearer's own when its inferential grounds are structurally inaccessible. The central claim is this: structural opacity in AI systems produces a condition — the epistemic alibi — that is distinct from the algorithmic alibi, the moral crumple zone (Elish 2019), and the responsibility gap (Matthias 2004). The algorithmic alibi conceals the bearer. The epistemic alibi does not: it erodes the ontological condition of dignity under A10, which holds that dignity is the capacity to bear the irreversible consequences of one's own choices. Own in A10 is an agentive predicate, not a causal one. The agentive condition requires evaluative access to inferential grounds — grounded independently through hierarchical theories of autonomy (Frankfurt 1971; Christman 1991, 2009). The article operationalises structural opacity through a Verification Capacity Ladder (L1–L4), establishes L3 (interpretable reconstruction) as the minimum threshold for dignity under A10, and explains why post-hoc approximation methods cannot satisfy it. Three governance criteria follow: the verification floor, the interpretability threshold, and the dignity-preserving deployment test. These are applied to COMPAS and mapped onto GDPR Article 22 and EU AI Act requirements. The article also addresses the tragic trade-off between epistemic alibi and the risk of unaided human judgment error, locating it as a structural tragic tension — irreducible by optimisation — through the Axiom of Tragedy (A6). The falsification condition is stated explicitly: the argument would be defeated by a demonstration that structural inaccessibility of inferential grounds is compatible with full dignity under A10.