Abstract: The rapid advancement of artificial intelligence (AI) has profoundly reshaped social structures and production models. Its widespread integration within organizations has attracted increasing scholarly attention regarding its influence on employees' work effectiveness. However, the academic community has yet to reach a consensus on how and under what circumstances human-AI collaboration systems affect employees' work effectiveness. The conclusions of existing studies remain inconsistent, reflecting the complexity of human-AI interaction dynamics. Although some research has explored the mechanisms through which human-AI collaboration influences work effectiveness, most studies have focused primarily on either positive or negative consequences, without thoroughly examining the “double-edged sword” effect of human-AI collaboration systems or the boundary conditions under which these effects occur.
To systematically investigate the effects and underlying mechanisms of human-AI collaboration systems on employee work effectiveness, this study conducted a comprehensive meta-analysis synthesizing findings from 79 domestic and international studies, encompassing 106 independent samples (n = 54,726). The analysis reveals that human-AI collaboration, AI autonomy, AI anthropomorphism, and employees' KSAs (knowledge, skills, and abilities) exert significant positive effects on work effectiveness, representing “opportunities”. Conversely, AI awareness exerts a negative effect, perceived as a “threat”. Furthermore, the study identifies AI trust and job insecurity as key mediating variables that jointly explain the dual pathways through which human-AI collaboration influences employees' work effectiveness. Moreover, in this process, as a gain-oriented psychological resource, AI trust exerts a stronger and more stable mediating effect than job insecurity, which functions as a loss-oriented mechanism. This dual mediation framework illustrates the coexistence of “opportunity” and “threat” mechanisms within human-AI collaboration systems.
The results also demonstrate the presence of several moderating factors, including employee categories, industry characteristics, and cultural contexts, which shape the strength of these effects. Specifically, human-AI collaboration exhibits a stronger positive influence on employee innovation among knowledge workers (compared with non-knowledge workers) and within high-technology industries (compared with manufacturing and service sectors). When examining job performance, the positive effect of human-AI collaboration is more pronounced among non-knowledge workers, within high-technology industries, and in Western cultural contexts than in Eastern ones.
Overall, the findings substantiate the double-edged sword effect of human-AI collaboration systems. On the one hand, such systems enhance employees' work effectiveness through mechanisms of AI trust; on the other hand, they can reduce effectiveness through mechanisms of job insecurity. Importantly, the meta-analytic evidence indicates that the positive effects outweigh the negative ones, suggesting that with appropriate organizational design and management, the benefits of AI-enabled collaboration can be maximized while its risks can be mitigated.
Theoretically, this study is grounded in the Conservation of Resources (COR) theory, which provides a robust framework for understanding how individuals strive to acquire, maintain, and protect valuable resources in the face of technological change. By situating human-AI collaboration within this theoretical lens, the study clarifies the resource-based mechanisms underlying the observed opportunity-threat duality. Practically, the findings offer actionable insights for organizations aiming to implement AI technologies responsibly, emphasizing the importance of fostering employee trust in AI systems, strengthening KSAs through training, and reducing job insecurity through supportive management practices.
In summary, this research contributes to the literature by (1) providing an integrated empirical synthesis through meta-analysis to reconcile prior inconsistent findings, (2) elucidating the dual pathways of AI trust and job insecurity through which human-AI collaboration affects work effectiveness, and (3) identifying critical boundary conditions that determine when and for whom AI serves as a performance enhancer or inhibitor. The results not only advance theoretical understanding of human-AI collaboration in the workplace but also offer practical guidance for organizations seeking to harness AI's transformative potential while safeguarding employee well-being and productivity.

Key words: human-AI collaboration, job insecurity, AI trust, job performance, employee innovation