关键词: 算法权威, 道德推脱, 自我责任归因, 可解释性, 道德冲突强度

Abstract: Algorithmic systems increasingly participate in high-stakes decisions in healthcare, judicial practice, and organizational management. When such decisions produce adverse consequences, human decision-makers may attribute their actions to algorithmic recommendations and distance themselves from moral responsibility. Yet empirical findings point in opposite directions: algorithmic authority can facilitate responsibility displacement and moral disengagement, but can also heighten moral vigilance and inhibit disengagement. Existing accounts explain reliance on and resistance to algorithms separately, leaving these contrasting consequences unresolved. This article integrates research on authority, automation bias, mind perception, responsibility attribution, and moral disengagement to propose a Dual Cognitive Appraisal Model of Algorithmic Authority and Moral Disengagement.
The model treats algorithmic authority as a subjective relation rather than an inherent technical property. It arises when individuals perceive an algorithm as accurate and professionally competent, grant its judgments interpretive priority, and become more inclined to rely on its recommendations. The model's central proposition is that algorithmic authority does not directly determine moral disengagement. Instead, it reshapes self-responsibility attribution, a context-dependent judgment about whether the individual remains the final decision-maker and is accountable for accepting an algorithmic recommendation and its consequences. Reduced self-responsibility attribution allows individuals to position themselves as executors, transfer responsibility to the algorithm or system, and activate state moral disengagement. Strengthened attribution reaffirms their status as responsible agents and constrains moral disengagement.
This mechanism connects two complementary pathways previously discussed separately. In the cognitive offloading and agentic-state pathway, the perceived objectivity and reliability of algorithmic advice encourage individuals to outsource judgment and reduce independent verification. This creates conditions for responsibility displacement and moral disengagement. In the value-conflict and responsibility-return pathway, individuals recognize that an algorithm lacks the experiential capacities required of a complete moral agent. When an algorithmic recommendation conflicts with salient moral values, they are more likely to scrutinize or reject the recommendation and to reclaim responsibility for the final decision. Algorithm aversion alone, however, does not demonstrate responsibility return because rejection may arise from doubts about accuracy, threats to professional autonomy, or general resistance to technology. Responsibility return requires stronger self-responsibility attribution.
The direction of self-responsibility attribution is jointly shaped by two appraisals. Explainability defines technical cognitive appraisal by determining whether individuals can understand and scrutinize the basis of an algorithmic recommendation. Moral conflict intensity defines moral cognitive appraisal by determining whether the recommendation creates tension among algorithmic advice, personal moral standards, and responsibility for the outcome. When explainability and moral conflict are both low, individuals have limited grounds and little motivation to examine the recommendation. They are therefore more likely to defer judgment to the algorithm, weaken self-responsibility attribution, and disengage morally through responsibility displacement. When both are high, individuals have both the capacity and the motivation to scrutinize the recommendation, recognize the moral significance of its consequences, and reaffirm their own responsibility, thereby inhibiting moral disengagement. The two mixed conditions reveal mismatches between scrutiny capacity and motivation. High explainability with low moral conflict provides the capacity to inspect the algorithm but may not motivate scrutiny; low explainability with high moral conflict motivates scrutiny but restricts its means. Responsibility attribution under these mixed conditions remains theoretically indeterminate and requires direct examination.
The proposed model advances current theorizing in three respects. First, it brings the facilitating and inhibiting effects of algorithmic authority into a single framework. Second, it identifies self-responsibility attribution as their shared proximal mechanism, shifting attention from whether people accept or reject algorithms to how they define their responsibility after an adverse outcome. Third, it formulates an interaction hypothesis in which technical and moral appraisals jointly determine the direction of responsibility attribution. This integration extends accounts centered on obedience to authority, automation bias, or mind perception by specifying when reliance may be disrupted and when moral responsibility may return to the human decision-maker. It also translates a system-level responsibility gap into an individual-level process of post-decision moral self-regulation.
The model is derived from indirect empirical evidence and theoretical reasoning, and its complete causal chain has not yet been directly tested. Future studies should manipulate algorithmic authority, explainability, and moral conflict intensity in morally consequential human-AI tasks while measuring self-responsibility attribution and state moral disengagement. They should also examine whether decision-control arrangements, perceived algorithmic mind, organizational responsibility structures, and cultural contexts alter the proposed pathways. By distinguishing the conditions under which algorithms become a shield for responsibility from those under which they prompt renewed human accountability, the model provides testable hypotheses for psychological research and a basis for designing human-AI systems that preserve meaningful scrutiny, clear responsibility boundaries, and human moral agency.

Key words: algorithmic authority, moral disengagement, self-responsibility attribution, explainability, moral conflict intensity