关键词: 人工智能, 应急决策与处置, 经验阻断效应, 替代式AI使用, 人智协同

Abstract: Artificial intelligence (AI) can improve emergency decision-making and response without necessarily strengthening managers’ experience-based judgment. When AI takes over activities through which managers encounter situations, form judgments, and learn from consequences, current task performance may become decoupled from subsequent experiential growth. We introduce the AI experience-blocking effect (EBE) to explain this possibility. Across three interrelated studies, we develop an integrated framework linking substitutive AI use, EBE, experiential learning processes, and experience consolidation.
We define EBE as a proximal, multidimensional task-process state comprising four formative dimensions: reduced exposure to raw situational cues, displacement of independent judgment generation, weakened ownership of consequences and feedback-based calibration, and superficial reflective transformation. Here, ownership of consequences concerns managers’ responsibility for relating action outcomes to their own prior judgments and expectations as part of learning. Reflective transformation involves reconstructing events, examining causes and counterfactuals, and deriving rules for future action. Each dimension captures a distinct activity, so the four dimensions are not interchangeable and need not vary together. EBE may occur even when AI advice is correct and existing skills remain intact: it concerns participation in activities needed to acquire experience, rather than judgment errors or skill decline. Subsequent learning changes and experience deficits are assessed separately.
Adaptive structuration theory provides an account of how AI use configurations shape managers’ opportunities to participate in these activities. Experiential learning and naturalistic decision-making perspectives explain how such participation supports experience-based judgment, while organizational knowledge creation and team learning perspectives address how members’ experience becomes available to a team. These connections distinguish the organization of AI assistance from the task processes and learning through which its effects on experience may develop.
Study 1 focuses on the conceptual boundaries and measurement of EBE. It combines critical incident interviews, exercise and debriefing materials, and expert assessments to define the four dimensions and develop a measurement instrument. Four reflective first-order dimensions form a second-order formative construct. Validation compares self-reports with task records and behavioral observations and tests whether EBE explains delayed experience outcomes after accounting for prior experience, cognitive offloading, AI reliance, and current task performance.
Study 2 examines whether substitutive AI use increases EBE by manipulating the human-AI-task configuration. Supportive AI use presents original materials first and requires managers to submit initial judgments or reflections before receiving corresponding AI input. Substitutive AI use presents summaries, judgments, plans, or debriefing drafts first, placing managers in the role of reviewing preformed outputs. The study includes a no-AI baseline, maintains access to original materials, and holds AI output content and quality constant across AI conditions. Managers may still seek original information or challenge AI outputs under either configuration; EBE is therefore assessed from their actual task activities. Current task performance is examined as a parallel outcome for comparison.
Study 3 examines two parallel, potentially correlated learning mechanisms in a longitudinal design using repeated high-fidelity simulations. Perceptual learning of situational cues involves learning to identify diagnostic cues and recognize situational patterns. Reduced cue exposure and displaced judgment generation may limit the searching, comparison, and integration through which these abilities develop. Associative learning of action-outcome relationships requires managers to connect their judgments and actions with subsequent consequences. Independently generated judgments and outcome expectations provide a basis for comparing anticipated and actual outcomes, explaining discrepancies, and revising action rules. When AI supplies judgments and interpretations before managers develop their own, opportunities for feedback-based calibration and reflective transformation may diminish. Repeated learning assessments and delayed tasks without AI test whether these constraints contribute to deficits in individual experience internalization. Novel scenarios and erroneous AI advice further test transfer and independent error correction.
Team experience externalization depends on both what members learn and how they work with that experience together. Restricted individual learning may leave members with less experiential material to articulate, including diagnostic cues, judgment rationales, outcome explanations, and rules for future action. Sharing, co-construction, constructive disagreement, and collective reflection shape how this material is developed into shared experience. The study examines members’ experience records alongside team interactions, causal maps, and after-action reviews to trace this transformation. Team experience externalization is thus treated as a distinct outcome that cannot be inferred by simply aggregating individual EBE scores.
By tracing how task experiences become resources for later judgment, this account extends the evaluation of AI-assisted emergency management beyond immediate task performance. It identifies where experience generation may be interrupted and how those interruptions may affect individual learning and team experience consolidation. Practical implications center on preserving original cue exposure, independent judgment, feedback-based calibration, and reflective transformation within human-AI collaboration, so that task effectiveness and the development of managers’ experience-based expertise can be evaluated together.

Key words: artificial intelligence, emergency decision-making and response, experience-blocking effect, substitutive AI use, human-AI collaboration