Evaluating Public Plans for AI Systems: Comparing Premortem and Promortems

Effective planning requires anticipating risks before they occur, yet planners often underestimate potential failures due to optimism bias. This research compared three reasoning techniques—critique, premortem, and promortem—for evaluating a proposed automated high-speed rail plan across two experiments (individual, N = 108; group, N = 30). Participants rated their confidence and understanding of the plan before and after generating reasons, then generated potential solutions. Premortem reasoning reduced participants’ confidence in the plan and then supported confidence recovery once solutions were generated, whereas promortem reasoning increased confidence throughout the session, consistent with optimism bias; effects were more pronounced in group settings. Critique and premortem elicited a higher proportion of reasons and solutions related to the plan’s AI than promortem. Findings support the premortem as a tool for calibrating confidence and generating actionable, domain-relevant risks during early-stage technology planning.