生成式人工智能对员工工作行为的双刃剑效应：整合性述评与分析框架

The Double-Edged Sword Effect of Generative Artificial Intelligence on Employee Work Behavior: An Integrative Review and Analytical Framework

摘要: 生成式人工智能正迅速嵌入企业的工作流程，但其对员工工作行为的影响方向在既有研究中存在明显分歧。本文采用整合性述评的方法，系统检索信息系统、组织行为与人力资源管理三个领域的相关实证研究，以工作要求–资源模型与资源守恒理论为骨架，构建统一的分析框架。研究发现生成式人工智能对员工工作行为存在方向相反的两条路径。增益路径经由心理授权、工作繁荣与工作重塑，提升员工的工作投入与工作效能；损耗路径经由技术压力、认知卸载与生成式人工智能依赖，导致批判性思维弱化与情绪耗竭。两条路径的相对强度受三类边界条件调节，分别是个体层面的AI素养与动机取向、组织层面的支持与治理能力，以及任务与岗位属性。由此得出，企业应放弃单纯的限制性策略，转向“受控赋能”的治理组合，在可见性建设、能力培养与边界划定三个方面同时着力。本文的贡献在于把分散于三个学科的实证发现整合进统一框架，并给出可操作的治理路径。需要说明的是，上述路径的现有证据以横断面设计为主，部分关系可能存在互惠性，本文在框架中对变量间关系的方向性另作专门界定。

Abstract: Generative artificial intelligence (AI) is rapidly becoming embedded in corporate workflows, yet existing research shows marked divergence regarding the direction of its impact on employee work behavior. Using an integrative review approach, this paper systematically searches relevant empirical studies in three fields—information systems, organizational behavior, and human resource management—and constructs a unified analytical framework based on the job demands-resources model and conservation of resources theory. The study finds that generative AI exerts two opposing pathways on employee work behavior. The gain pathway, through psychological empowerment, thriving at work, and job crafting, enhances employees’ work engagement and work effectiveness. The loss pathway, through technostress, cognitive offloading, and generative AI dependence, leads to weakened critical thinking and emotional exhaustion. The relative strength of the two pathways is moderated by three types of boundary conditions: individual-level AI literacy and motivational orientation, organizational-level support and governance capacity, and task and job attributes. It follows that firms should abandon purely restrictive strategies and shift toward a governance portfolio of “controlled empowerment,” focusing simultaneously on visibility building, capability development, and boundary setting. This paper’s contribution lies in integrating empirical findings scattered across three disciplines into a unified framework and offering actionable governance pathways. It should be noted that the existing evidence for these pathways is predominantly cross-sectional, and some relationships may be reciprocal; the analytical framework addresses the directionality of these relationships explicitly.