Answers are grounded in this paper and cite their sources.
Generative artificial intelligence is increasingly used to support technical program managers in identifying delivery risks, but improved productivity does not necessarily imply more complete risk recognition. A critical failure mode arises when a manager initially identifies a material risk and later removes it after encountering an AI-generated assessment. This paper develops a human–AI decision-comparison method for detecting such omissions without equating disagreement with error. The framework records an unaided initial risk register, randomly assigns decision makers to a human-only second pass, ordinary AI assistance, or AI assistance with a structured challenge step, and assesses final decisions against independently adjudicated scenario risks. It distinguishes risk discovery, risk retention, and severity-weighted coverage using explicit transition measures and a multilevel comparison strategy. Domain-specific scenarios span cloud migration, data platforms, regulated delivery, and automotive system validation. The contribution is a reproducible measurement architecture for identifying whether AI assistance is associated with excess removal of risks that were previously recognized. The paper specifies testable causal contrasts and validity controls; it does not claim that an experiment has already been conducted.
No takes yet. Share an insight, caveat, or question.
Nandkishore R Sawant (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: