Beyond Linear AI Education: A Task-Oriented Competency Graph Architecture for Interdisciplinary Vocational Learning and Assessment

Artificial intelligence (AI) education must address diverse professional objectives: comprehending system limits, operating generative tools, verifying synthetic outputs, and, for technical roles, designing and governing systems. Monolithic, flat competency taxonomies provide broad subject coverage but fail to resolve task-specific curriculum selection or explain why certain capabilities should be prioritized or deferred. This paper proposes a task-oriented competency graph architecture G = (V, E, Phi, T) that formally models prerequisite constraints, symbiotic co-development, and cognitive offloading interference across foundational (V_core), task-specific (V_task), and specialized (V_spec) cognitive tiers. Grounded in cognitive load theory, evaluative judgment, and design science research, the architecture features an explainable selection function Phi that derives an induced subgraph G_tau under strict prerequisite transitive closure, preventing fragmented or ungrounded skill acquisition. The conceptual model is evaluated against empirical and institutional precedents (including UNESCO, AAMC 2026, and SAIL) alongside an illustrative medical education case study with three distinct clinical roles and simulated performance assessments. We demonstrate that while multi-tier taxonomies have existing antecedents, our authentic contribution lies in the formal graph selection mechanism, transparent decision records, and explicit demarcation between AI-assisted performance and unassisted human competence.