Infrastructure as Inherited Knowledge: Artificial Intelligence and the Governance of Long‐Life Infrastructure

ABSTRACT Epistemic orphaning, responsibility drift, loss of institutional memory, and non‐stationarity are not new problems in infrastructure governance—they have accompanied long‐life infrastructure throughout its history. What artificial intelligence changes is not the existence of these problems but their scale, opacity, and the structural difficulty of recognizing them once they have taken hold. This paper argues that AI in long‐life infrastructure should be understood not primarily as a source of new risk but as a magnifier of existing governance vulnerabilities, demanding a more deliberate and durable framework for stewardship than the field has previously required. Using a double bow‐tie governance framework applied to an integrated urban subway system, the paper develops three distinct governance regimes: a construction regime concerned with knowledge creation, a stewardship transition regime concerned with knowledge continuity, and a long‐term operations regime concerned with knowledge interpretation and renewal. Construction and operations are two fundamentally different regimes, not two phases of one system, connected by a stewardship transition that is the highest governance risk point in the lifecycle. Three interlocking concepts structure the analysis: normalization of plausibility , the characteristic failure mode of construction, in which AI outputs appear credible even when their underlying assumptions are wrong; epistemic orphaning , the failure mode of the stewardship transition, in which records and AI outputs survive but the context required to interpret them is permanently lost; and normalization of deviance , the failure mode of long‐term operations, in which gradual drift from design conditions goes unrecognized because each step looks acceptable relative to the last. The framework introduces two governance innovations absent from current infrastructure practice: the AI Steward , an organizational role responsible for the interpretability and accountability of AI‐generated knowledge across the asset's operational life; and model reconciliation and arbitration , a formal function for resolving conflicts between construction‐phase AI predictions and operational observations. Together these constitute a stewardship architecture designed to keep accountability aligned with AI influence across the infrastructure lifecycle. Long‐life infrastructure is not simply built and operated; it is inherited. In the age of artificial intelligence, governing what is inherited may prove more important than governing how it is created.