Preamble

Opportunity or Threat? A Meta-Analysis of the Impact of Human-AI Collaborative Systems on Employee Work Effectiveness

Abstract

As artificial intelligence (AI) technology continues to evolve, Human-AI Collaboration (HAIC) systems have become increasingly prevalent in modern organizational settings. However, academic consensus remains divided on whether these systems represent an opportunity for enhancement or a threat to the workforce. This study employs a meta-analytical approach to systematically evaluate the impact of human-AI collaborative systems on employee work effectiveness. By synthesizing existing empirical research, we explore the dual-pathway mechanisms through which AI collaboration influences performance and psychological well-being. Our findings aim to clarify the boundary conditions of these effects and provide theoretical and practical insights for organizations navigating the integration of AI into their workflows.

Introduction

The rapid advancement of machine learning and deep learning has transitioned AI from a simple automated tool to a collaborative partner capable of complex decision-making and task execution. Human-AI Collaboration (HAIC) refers to the synergistic interaction between human workers and AI systems to achieve shared organizational goals. Despite the potential for increased efficiency, the introduction of AI into the workplace has sparked significant debate regarding its impact on employee outcomes.

On one hand, proponents argue that AI serves as a powerful "opportunity," augmenting human capabilities, reducing cognitive load, and freeing employees from repetitive tasks to focus on high-value creative work. On the other hand, critics highlight the "threat" posed by AI, citing concerns over job displacement, algorithmic management, and the potential for increased work stress and reduced autonomy. This tension necessitates a comprehensive meta-analysis to reconcile conflicting findings and provide a clearer picture of the current state of human-AI collaboration.

Theoretical Framework and Hypotheses

This research utilizes the Job Demands-Resources (JD-R) model and Social Cognitive Theory to examine the impact of HAIC systems. We propose that AI collaboration functions as both a resource and a demand, leading to divergent effects on work effectiveness.

The Opportunity Pathway: AI as a Resource

When AI systems are designed to support human decision-making, they act as a job resource. By providing data-driven insights and automating routine processes, these systems can enhance an employee's self-efficacy and task performance. We hypothesize that high-quality human-AI interaction positively correlates with work engagement and objective performance metrics.

The Threat Pathway: AI as a Demand

Conversely, when AI systems are perceived as opaque, unpredictable, or evaluative, they function as a job

2 苏

School of Human Resources, Guangdong University of Finance and Economics, Guangzhou; School of Business Administration, Guangdong University of Finance and Economics, Guangzhou; School of Management, Guangdong Industrial University, Guangzhou. The rapid development of artificial intelligence (AI) has profoundly transformed social structures and production models. Consequently, the impact of AI applications within organizations on employee work performance has garnered significant attention from the academic community.

To explore the impact and underlying mechanisms of human-AI collaboration systems on employee work performance, this study conducted a meta-analysis of independent samples. The results indicate that the application of human-machine collaboration, anthropomorphism, and employee KSAOs (Knowledge, Skills, and Abilities) have a positive impact on employee work performance. Conversely, AI crisis awareness exerts a negative influence. Furthermore, trust in AI and job insecurity play mediating roles in the relationship between collaboration systems and employee work performance. The analysis further elucidates that employee category, industry attributes, and cultural background serve as significant moderators in these relationships.

The research findings demonstrate that human-AI collaboration has a "double-edged sword" effect: it can enhance employee work performance by fostering trust in AI, but it can also diminish performance by increasing job insecurity. Notably, the positive effects were found to be stronger than the negative effects. Within the framework of Conservation of Resources (COR) theory, this study clarifies the mechanisms and boundary conditions through which collaboration systems affect employee performance. These insights provide guidance for organizations to objectively evaluate the impact of human-AI collaboration systems and effectively leverage the value of AI.

关键词

The Impact of AI Collaboration and Job Insecurity on Employee Performance and Innovation

Abstract

As artificial intelligence (AI) continues to reshape the modern workplace, understanding the psychological and behavioral responses of employees to AI collaboration has become a critical area of organizational research. This study explores the complex relationship between human-AI collaboration, job insecurity, and AI trust, and their subsequent effects on work performance and employee innovation. By integrating theories of organizational behavior and machine learning adoption, we examine how the perceived threat of automation interacts with the necessity of technological cooperation.

1. Introduction

The rapid integration of machine learning and deep learning technologies into organizational workflows has fundamentally altered the nature of professional collaboration. While AI offers significant potential for enhancing efficiency and driving innovation, its implementation often triggers a dual response among employees: a recognition of its utility and a fear of displacement. This "AI paradox" necessitates a deeper investigation into how job insecurity and trust in AI systems mediate the relationship between human-AI collaboration and overall work outcomes.

2. Theoretical Framework and Hypotheses

2.1 Human-AI Collaboration and Work Performance

Collaboration with AI systems involves a synergistic relationship where human cognitive abilities are augmented by the computational power of machine learning algorithms. We hypothesize that effective collaboration leads to improved work performance by automating routine tasks and providing data-driven insights. However, the efficacy of this collaboration is contingent upon the employee's level of AI trust.

2.2 The Mediating Role of Job Insecurity

Job insecurity serves as a significant psychological stressor in the age of automation. When employees perceive AI as a substitute rather than a supplement, their sense of job stability decreases. This perceived threat can lead to a decline in both task performance and employee innovation. We model the impact of job insecurity using the following relationship:

$$ \text{Job Insecurity} = f(\text{AI Capability}, \text{Role Redundancy}, \text{Organizational Support}) $$

2.3 AI Trust as a Moderator

Trust in AI is defined as the employee's willingness to rely on the system's recommendations and actions. High levels of AI trust can mitigate the negative effects of job insecurity, fostering an environment conducive to innovation. Without trust, even the most advanced deep learning tools may fail to produce meaningful organizational value.

[FIGURE:1]

3. Methodology

This study employed a mixed-methods approach, utilizing survey data from 500 employees across various technology-

1 引言

Artificial Intelligence (AI), as a pivotal force driving technological progress, industrial upgrading, and productivity enhancement, is increasingly reshaping social structures and production models. It is widely regarded as a critical general-purpose technology for the future of society \cite{Brynjolfsson, Mcafee}. However, the rapid development of this technology has also brought about profound impacts and widespread concerns. The introduction of AI may replace traditional job roles, alter decision-making processes, and even influence employee work experiences and outcomes \cite{Felten, songyixiao, gdufe}.

This impact exhibits a significant duality. From a positive perspective, collaboration with humans through a division of labor allows for the realization of comparative advantages. It can liberate employees from low-value tasks, enabling them to focus on enhancing professional skills and engaging in more creative, high-value activities. This collaborative model not only improves work efficiency but also provides employees with greater room for development and opportunities for innovation.

However, from a negative perspective, if employees develop a negative perception of these systems—for example, viewing tracking and monitoring in the workplace as an invasion of privacy—it can undermine their sense of autonomy at work. This, in turn, can have an adverse effect on subsequent job performance \cite{2021; Savela}. Against this background, it is of great significance to analyze the impact of human-machine collaborative systems on employee work efficacy. Work efficacy is typically defined as...

Work performance is regarded as a key metric for measuring employee output. Specifically, it reflects whether an employee's work behavior and results meet the requirements and expectations of the organization.

organizational performance requirements (2023), while innovation is what helps enterprises stand out. Therefore, the core focus of this study is to comprehensively evaluate the dual impact of human-collaboration systems on employee job performance and innovation, thereby revealing their overall effect on employee work effectiveness.

Although academic discussions regarding the impact of human-AI collaborative systems have become increasingly prevalent, a consensus has yet to be reached. On one hand, some studies argue that human-AI collaboration yields significant positive effects. For instance, AI can undertake repetitive, high-risk, or high-precision tasks, allowing employees to focus on more complex work and thereby enhancing overall performance \cite{Zhang}. Furthermore, the application of AI provides employees with greater autonomy, stimulating creativity and facilitating the completion of innovative tasks \cite{Hauptman}.

On the other hand, some research emphasizes the negative effects of human-AI collaborative systems. These studies point out that employees may struggle to adapt to rapidly changing technologies or may even fear being replaced by AI, leading to a decline in performance \cite{2019}. Additionally, in contexts heavily dependent on data and algorithms, employees may rely excessively on AI and lack independent thought, which can undermine their autonomy and creativity \cite{Verma, Singh}. As research has deepened, some scholars have proposed the perspective that human-AI collaboration exerts a dual effect on innovation.

This may act as a job demand that negatively impacts innovative behavior by increasing employees' job insecurity; conversely, it may serve as a job resource that positively stimulates innovative behavior by enhancing employees' perceptions of autonomy. Therefore, an in-depth analysis of the mechanisms underlying human-AI collaboration from a systematic perspective will help deepen research on AI within organizations. Furthermore, such an analysis provides practical guidance for optimizing human-AI collaboration, thereby enhancing the overall effectiveness of the organization.

By reviewing the existing literature, it is evident that the underlying mechanisms through which human-AI collaborative systems influence employee work effectiveness still require further refinement. As previously mentioned, human-AI collaboration is characterized by complex interaction effects; however, due to the constraints of single empirical studies—both in terms of scope and methodology—existing research typically analyzes only a limited set of indicators. Furthermore, the current classification of dimensions within human-AI collaboration remains relatively coarse. This lack of granularity prevents a comprehensive clarification of how various multidimensional characteristics influence employee performance and the differing intensities of these effects.

A critical question remains: how do these systems evolve into effective collaborative partners? While some studies have begun to explore the specific processes by which human-AI collaboration impacts work effectiveness, most research to date has focused narrowly on isolated factors. Consequently, there is a need for a more integrated framework that accounts for the multifaceted nature of these systems and their nuanced impact on organizational outcomes.

Existing research has primarily focused on either positive or negative effects in isolation. Furthermore, these studies often rely on limited and fragmented samples, failing to provide a clear comparison of the relative magnitudes of different effects. Consequently, the underlying mechanisms of human-AI interaction effects require further in-depth investigation. Specifically, the boundary conditions under which AI systems transition into collaborative systems that influence performance remain to be comprehensively explored.

For instance, high-tech industries may experience distinct developmental trajectories due to their emphasis on innovation and flexibility. Additionally, factors such as cultural differences, employee gender, and age variations may also influence the outcomes generated by human-AI collaborative systems. To address these gaps in existing empirical research, this study constructs a theoretical model based on Conservation of Resources (COR) Theory.

Conservation of Resources (COR) theory emphasizes that individuals possess a strong inclination to maintain, acquire, and protect their resources; consequently, the perception of resource gain or loss directly influences their attitudes and behaviors \cite{Hobfoll}. In collaborative contexts, AI can serve as a resource gain tool by automating tasks to release employees' time and cognitive resources; conversely, it can act as a resource loss threat, triggering concerns regarding job displacement.

Drawing upon COR theory, this study innovatively selects job insecurity and trust in artificial intelligence as core mediating variables. First, from the perspective of the resource acquisition pathway, trust in AI reflects an employee's subjective evaluation of the technology's reliability and value.

Employees with high levels of trust in Artificial Intelligence (AI) are more inclined to view AI as a resource for expanding their professional capabilities; consequently, they actively explore its functions and integrate them into their workflows \cite{Glikson Woolley, 2020; Hauptman}. Conversely, from the perspective of the resource depletion path, job insecurity reflects an employee's perception of risks regarding their continued professional existence. When employees perceive AI as a threat to job replacement, it triggers psychological defense mechanisms, leading them to mitigate potential losses by reducing investment in innovation and avoiding technical learning \cite{Sharif}.

In terms of theoretical integration, these two mediating variables form a complementary relationship: AI trust focuses on explaining the positive effects of human-machine collaboration, while job insecurity focuses on the negative effects. Together, they constitute a comprehensive path of action. Regarding data sufficiency, the sample sizes for both variables are substantial, satisfying the requirements for structural equation modeling (SEM) of the full model. Therefore, this study selects job insecurity and AI trust as the mediating variables for the effects of human-AI interaction, exploring the mechanisms by which AI collaboration systems influence job performance and employee innovation through these dual resource paths. Specifically:

Main Effects of Collaborative Systems

The system analyzes the multidimensional characteristics of human-machine interaction, specifically examining how human-machine collaborative applications, system features, and employee characteristics impact job performance and employee innovation. This study aims to provide a deep understanding of the underlying influence mechanisms within human-machine collaborative systems. By employing Meta-Analytic Structural Equation Modeling (MASEM), we quantify the differing mediating roles of job insecurity and trust in artificial intelligence. This approach allows for a rigorous disclosure of the specific mechanisms through which these collaborative systems operate and affect organizational outcomes.

The boundary conditions under which human-robot collaboration (HRC) systems influence employees are examined through a three-dimensional moderating framework that integrates micro-, meso-, and macro-level factors. Specifically, this study employs meta-regression analysis, treating employee gender and age as continuous variables. Simultaneously, subgroup analyses are conducted by categorizing employee types, industry attributes, and cultural backgrounds as categorical variables. This comprehensive approach aims to explore how these factors moderate the relationship between variables associated with human-robot collaboration systems and overall employee work effectiveness.

2.1 变量定义

When exploring the relationship between technology and work, traditional perspectives typically view technology as either a tool or a medium. The tool perspective focuses on how employees utilize technological tools to enhance performance \cite{Nelson, Irwin}, while the medium perspective focuses on how technology facilitates team collaboration \cite{Bechky}. However, these perspectives are largely limited to the study of human individuals and tend to overlook the dynamic characteristics of the collaborative environment. To address this deficiency, \cite{Ajoudani} defines human-machine collaboration as a coupled dynamic system formed by the mutual contact of humans, machines, and the environment to complete specific tasks. \cite{Wang, Yao} point out that the core elements of human-machine collaboration include employee characteristics, robot characteristics, and environmental characteristics. \cite{Yin, Niu} further analyze the collaboration between AI and employees across four dimensions: AI technical factors, individual employee factors, organizational contextual factors, and task configurations. Synthesizing existing research, this study defines the human-machine collaborative system as an interactive system where employees collaborate with AI tools, algorithms, and related technologies within a specific organizational context to complete tasks. To analyze the impact of human-machine collaboration on employee work effectiveness, this study conducts an analysis across three dimensions: the organizational context of human-machine collaboration applications, employee characteristics, and AI crisis awareness. Human-machine collaboration application refers to the systematic practice in which employees continuously interact with AI systems—possessing autonomous learning, reasoning, and decision-making capabilities—within an organizational environment to collaboratively achieve core task objectives. This concept encompasses both the capability level of the technology itself and the breadth and depth of its invocation and integration within actual workflows. In empirical research, there are two common ways to operationalize this variable: one approach treats it as a categorical variable through experimental manipulation, such as the two-factor design used by \cite{2023} involving learning goal orientation to explore interactive effects on innovative behavior.

Another line of experimental design examines the impact of human-machine collaboration on task performance across different scenarios by manipulating the intelligence level of the assistant. Other studies employ continuous variables for measurement; for instance, some characterize it through behavioral indicators such as frequency of use and functional dependence. Others further integrate multiple items—including frequency of use, necessity, and interaction density—to more comprehensively capture the intensity of human-machine synergy. This diversified measurement approach provides a flexible operational foundation for empirical research and facilitates variable coding and theoretical comparison in subsequent meta-analyses.

The characteristics of AI primarily include two dimensions: first, autonomy, which refers to the ability of the AI to execute tasks and make decisions independently without explicit human intervention; and second, anthropomorphism, which refers to the degree to which the AI simulates human characteristics in terms of appearance, voice, and interaction methods \cite{Alabed}. These two types of characteristics significantly influence the naturalness of human-machine interaction and collaborative efficiency. Employee characteristics also include two key variables: first, Knowledge, Skills, and Abilities (KSA), specifically referring to the employee's experience, familiarity, and sensitivity regarding the use of artificial intelligence.

The second employee characteristic is AI crisis awareness, which refers to employees' perception of occupational risks—such as job displacement and skill devaluation—that may be caused by AI technology \cite{Brougham, Haar}. A high level of crisis awareness may have a negative impact on collaborative attitudes and work behaviors. Based on the inductive logic of meta-analysis, this study incorporates two highly scrutinized research outcomes—job performance and employee innovation—to evaluate the impact of human-machine collaboration on overall employee work effectiveness. To improve predictive validity, and with reference to reference books and related research, variables with similar meanings are merged into broader categories for analysis \cite{Alabed}. Specifically, adopting a two-dimensional view of performance, task performance and contextual performance are merged into the comprehensive variable of "Job Performance." Furthermore, given that both creativity and innovation drive organizational development, this study operationally defines employee innovation in a broad sense. Related variables such as employee creativity, employee innovative behavior, employee innovation outcomes, and employee innovation performance are unified into the aggregate variable "Employee Innovation." This refers to the ability of employees to create valuable new products, ideas, or processes within an organization, spanning from the generation of innovative ideas to the implementation of processes and the realization of final results \cite{Oldham, Cummings}. The analysis also considers factors across three levels: micro (employee gender, age, and type), meso (industry attributes), and macro (cultural background) regarding the application of human-machine collaboration.

1. 工作不安全感

Employees' fear of artificial intelligence replacing their job roles, as well as their concerns regarding the inability to adapt to new job requirements brought about by technological change.

2. 人工智能信任

The Impact of Reliability, Transparency, Unbiasedness, Safety, Adaptability, and Collaborative Intent on the Performance of Collaborative Systems

In the field of human-machine collaboration, the effectiveness of a system is not merely determined by its technical specifications, but by a complex interplay of several critical dimensions. These include reliability, transparency, unbiasedness, safety, and adaptability. Together with the underlying collaborative intent, these factors fundamentally shape the overall work performance and efficiency of collaborative systems.

Reliability and Transparency

Reliability serves as the cornerstone of any collaborative framework. A system must consistently perform its intended functions under stated conditions for a specified period to build user trust. When a system is reliable, human collaborators can delegate tasks with confidence, thereby reducing cognitive load and allowing for more strategic focus. Closely linked to reliability is transparency, which refers to the degree to which the system's internal logic, decision-making processes, and current states are observable and understandable to the user. High transparency enables users to predict system behavior and intervene effectively when necessary, preventing the "black box" effect that often hinders seamless cooperation.

Unbiasedness and Safety

As machine learning and deep learning become increasingly integrated into collaborative systems, the issue of unbiasedness has gained paramount importance. A system must operate without prejudice, ensuring that its outputs and decisions are not skewed by flawed training data or discriminatory algorithms. Unbiased systems foster equitable outcomes and maintain the integrity of the collaborative process. Simultaneously, safety remains a non-negotiable requirement. This encompasses both the physical safety of human operators in proximity to robotic systems and the functional safety of data and processes. A secure and safe environment is essential for maintaining the continuous operation of the system and protecting the well-being of all participants.

Adaptability and Collaborative Intent

The dynamic nature of modern work environments requires collaborative systems to possess high levels of adaptability. An adaptable system can adjust its strategies and behaviors in response to changing environmental conditions or evolving task requirements. This flexibility ensures that the system remains effective even in the face of uncertainty. Furthermore, the concept of collaborative intent—the shared understanding and alignment of goals between the human and the machine—is vital. When the system's actions are aligned with the user's objectives, the synergy between the two entities is maximized, leading to significantly enhanced work efficiency and task success.

Impact on Work Performance

The integration of these factors directly influences the overall performance of collaborative systems. When reliability, transparency, unbiasedness, safety, and adaptability are optimized, the system

2.2 理论框架

Conservation of Resources (COR) theory provides a systematic perspective for analyzing the complex effects of human-AI collaboration. The core of this theory lies in the acquisition, protection, and maintenance of resources by individuals. Resources are defined as objective entities (such as tools), conditions (such as job autonomy), or psychological traits that individuals perceive as valuable \cite{Hobfoll1989}. COR theory emphasizes that individuals lacking resources are more susceptible to "loss spirals"—where existing resources are continuously depleted and difficult to replenish—whereas resource-rich individuals tend to form "gain spirals," acquiring further resources through the accumulation of existing ones.

In human-AI collaborative systems, the introduction of AI affects an employee's resource status through two parallel pathways. On the resource gain pathway, when AI is perceived as a resource-supplementing tool, it creates a positive cycle by directly providing resources or enhancing resource utilization efficiency. For example, as a collaborative partner, AI's efficient information processing and predictive capabilities can provide effective support, thereby strengthening an employee's confidence and skills. In this scenario, employees believe that AI can enhance their job competence and autonomy. This positive perception protects their resources and increases intrinsic motivation, further reinforcing collaborative efficacy and innovative drive \cite{Huang2019, Gursoy2019}.

Conversely, on the resource depletion pathway, when AI is perceived as a resource threat, it triggers resource loss and defensive reactions. Employees may feel their occupational security is threatened by concerns that AI might replace their positions \cite{Li2019}. This job insecurity can lead to further resource loss; when individuals face resource threats, they often adopt conservative and defensive strategies, which may inhibit job performance and innovation \cite{Huang2019, Gursoy2019}. Therefore, based on the dual-pathway logic of COR theory, this study focuses on the triggering mechanisms of resource gain and loss within human-AI collaboration—including AI application contexts, AI characteristics, and employee characteristics—to explore how these factors influence job performance and innovative behavior through resource dynamics.

2.3 人与

Collaborative Systems and Employee Work Efficacy

Theoretically, the human-collaboration system functions as a multi-level dynamic interaction framework. Its core elements include human-machine collaborative applications, system characteristics, and employee characteristics. These elements collectively influence employee work performance and innovative behavior through the dual mechanisms of resource gain and resource depletion. When organizations systematically deploy and apply collaborative systems, employees can delegate routine, highly repetitive tasks to the system, thereby conserving time and cognitive resources. These resources can then be reallocated to tasks that require human judgment, complex reasoning, and interpersonal interaction. This effective release and redistribution of resources alleviate the employee's workload and contribute to improving the efficiency and quality of task completion, thus enhancing work performance. Furthermore, when employees actively adapt to technological changes in artificial intelligence and view them as opportunities for personal growth and innovation, their motivation for innovative exploration is effectively stimulated, leading them to more actively experiment with new methods and conceive new solutions. Therefore, we propose the following: Human-machine collaborative applications have a positive impact on work performance. Human-machine collaborative applications have a positive impact on employee innovation.

The inherent characteristics of a system facilitate the accumulation of employee resources, which in turn affects employee work efficacy. First, the ability of a system to independently complete complex tasks and make real-time decisions significantly reduces the cognitive load on employees by enhancing system usability and adaptability, helping them maintain limited cognitive resources \cite{Glikson & Woolley, 2020}. On the other hand, by replacing routine operations, it releases the employee's time and energy, providing the necessary foundational conditions for improving work performance and employee innovation \cite{Acqua et al., 2023}. Anthropomorphism provides socio-emotional support through human-like interactions (such as smiling, gazing, and voice interaction), thereby strengthening the emotional connection between the system and the employee and enhancing the employee's trust and willingness to collaborate with the technology. This positive emotional resource helps bridge the psychological gap in human-machine collaboration and promotes work performance \cite{Papadopoulos et al., 2016}. Anthropomorphic design interactions improve employees' acceptance of technology, motivating them to more actively explore new tools and address innovative challenges, thereby driving the emergence of innovative behavior \cite{Alabed et al., 2022}. Therefore, the following hypotheses are proposed:

Autonomy has a positive impact on work performance. Autonomy has a positive impact on employee innovation. Anthropomorphism has a positive impact on work performance. Anthropomorphism has a positive impact on employee innovation.

Individual skills and cognitive characteristics of employees can influence work efficacy through two pathways: resource gain and resource loss. In the resource gain pathway, as a core human capital resource, skills enable employees to efficiently utilize technology. While improving work efficiency, this also promotes a virtuous cycle of resources in human-machine collaboration. Employees with high artificial intelligence skills are better able to integrate professional knowledge with technological applications, allowing them to actively engage in innovative behavior and produce unique outcomes when resources are abundant. In the resource loss pathway, employees' concerns about the substitutability of technology can easily trigger job insecurity and the depletion of psychological resources. To avoid further resource loss, individuals may adopt conservative strategies, reducing work engagement or even exhibiting negative behaviors, which negatively impacts work performance \cite{Brougham & Haar, 2018}. At the same time, this sense of crisis consumes the psychological resources required for employee innovation and the use of technology, especially in high-risk and highly challenging innovative activities \cite{Ding et al., 2021; Verma & Singh, 2021}. This resource threat not only inhibits employees' willingness to innovate but may also reduce their enthusiasm for exploring new ideas.

Based on this, the following hypotheses are proposed: Skills have a positive impact on work performance. Skills have a positive impact on employee innovation. Employee artificial intelligence crisis awareness has a negative impact on work performance. Employee artificial intelligence crisis awareness has a negative impact on employee innovation.

2.4 人与

The Dual-Effect Mechanism of Human-AI Collaborative Systems: The Mediating Roles of Job Insecurity and AI Trust

First, from a practical perspective, the rapid iteration of technology and organizational transformation has intensified employees' job insecurity. Simultaneously, the development of human-AI collaboration is highly dependent on the level of trust employees place in AI \cite{McGrath}. These two variables provide an important empirical basis for explaining the dual-effects of human-AI collaboration on job performance and employee innovation from both negative and positive directions.

Second, from a theoretical perspective, both job insecurity and AI trust are closely related to an individual's resource state. Specifically, job insecurity reflects the resource loss threats perceived by employees; such threats may stem from risks such as skill devaluation or job displacement brought about by AI technology. Conversely, AI trust reflects the resource gains employees obtain through technology adoption, which helps enhance the effectiveness of human-AI interaction \cite{Glikson, Woolley}.

Finally, regarding mechanistic explanatory power, job insecurity and AI trust reveal the impact of human-AI collaborative systems on job performance and employee innovation from the perspectives of resource depletion and resource gain, respectively. Job insecurity inhibits employee performance and innovation by triggering anxiety and reducing work engagement \cite{Osborne}. In contrast, AI trust enhances collaborative efficiency by increasing technology acceptance and promoting knowledge sharing \cite{Glikson, Woolley}. Together, these two factors constitute the key psychological mediating mechanisms within human-AI collaborative systems. Therefore, it is of great significance to sequentially elaborate on the specific impacts of human-AI collaboration applications across three dimensions: job insecurity and AI trust.

2.4.1 工作不安全感的中介作用

Job insecurity refers to the concerns employees have regarding the application of technology and the subsequent value of their skills \cite{Shoss, 2017}. According to Conservation of Resources (COR) theory, when individuals perceive a potential loss of resources, a defense mechanism is triggered, leading them to adopt conservative strategies to avoid further loss. This, in turn, has a broad impact on work behaviors and outcomes. Research indicates that while the introduction of AI and systemic human-machine collaboration improves efficiency, it also redefines the boundaries of traditional job responsibilities, exacerbating employees' fears of being replaced. Specifically, job insecurity can cause anxiety, distraction, and psychological stress, reducing work engagement and task focus, which negatively affects job performance \cite{Sverke et al., 2002}. Furthermore, it inhibits employees' risk-taking spirit and creative thinking, making them prone to conservative work strategies and avoiding novel methods or ideas, thereby hindering innovative behavior \cite{Jiang and Lavaysse, 2018}. This insecurity triggered by human-machine collaboration leads to reduced work effort and fewer innovative attempts, ultimately decreasing overall job performance and innovation capacity.

Therefore, the following hypotheses are proposed:
$H_1$: Job insecurity mediates the relationship between human-machine collaboration applications and job performance.
$H_2$: Job insecurity mediates the relationship between human-machine collaboration applications and employee innovation.

As capabilities improve and the ability to perform routine tasks increases, employees are increasingly inclined to make comparisons, thereby evaluating the likelihood and consequences of being replaced.

When technology poses a threat, negative emotions significantly increase, further aggravating job insecurity \cite{Shoss, 2017}. In situations where resources are threatened, employees tend to adopt conservative strategies, reducing resource investment in high-risk tasks such as innovation activities. This protective behavior leads to a decline in work motivation and may suppress employees' innovative capabilities and performance \cite{Hobfoll, 1989}. Anthropomorphism may exacerbate employees' threat perception of artificial intelligence through its realistic appearance and emotional expression capabilities.

If technology is perceived as being too close to humans, employees may view it as a direct competitor rather than a mere support tool. This can trigger concerns about job substitutability, further intensifying job insecurity and exerting a negative impact on employees' psychological states and work behaviors \cite{Papadopoulos et al., 2020}. Therefore, the following hypotheses are proposed:
$H_3$: Job insecurity mediates the relationship between AI autonomy and job performance.
$H_4$: Job insecurity mediates the relationship between AI autonomy and employee innovation.
$H_5$: Job insecurity mediates the relationship between AI anthropomorphism and job performance.
$H_6$: Job insecurity mediates the relationship between AI anthropomorphism and employee innovation.

By enhancing technical adaptability and transformation capabilities, insecurity can be effectively mitigated. Employees who are proficient in AI typically have more confidence in their career development and are more inclined to view AI as an empowering tool rather than a substitutive threat.

A rich skill reserve not only helps maintain the stability of current positions but also provides a foundation for exploring new career opportunities, thereby significantly reducing job insecurity \cite{Huang and Rust, 2018}. On the other hand, employees' AI crisis awareness directly reinforces their sense of insecurity. Employees who are highly sensitive to AI substitution are more likely to fall into expectations of resource loss, triggering conservative and defensive behaviors \cite{Brougham and Haar, 2018}. Because these employees constantly perceive professional threats, they often actively avoid participating in risky innovative activities, ultimately suppressing their job performance and innovative output \cite{Shoss, 2017}. Therefore, the following hypotheses are proposed:
$H_7$: Job insecurity mediates the relationship between employee AI self-efficacy and job performance.
$H_8$: Job insecurity mediates the relationship between employee AI self-efficacy and employee innovation.
$H_9$: Job insecurity mediates the relationship between employee AI crisis awareness and job performance.

$H_{10}$: Job insecurity mediates the relationship between employee AI crisis awareness and employee innovation.

2.4.2 人工智能信任的中介作用

AI trust refers to an employee's positive beliefs regarding the technical competence, integrity, and benevolence of AI systems. As a critical psychological pathway for resource gain, it significantly promotes technology acceptance and collaborative behavior \cite{Glikson_Woolley}. Within a system, AI trust enhances employees' sense of psychological safety and control, prompting them to more actively invest cognitive and emotional resources into work tasks, thereby improving job performance \cite{Demiris}. Simultaneously, AI trust provides a psychological safety net for employees to experiment with new methods and innovative thinking, reducing the perceived risk of innovation failure and thus fostering innovative behavior \cite{Hengstler}. Human-machine collaborative applications contribute to the accumulation of positive interaction experiences by providing continuous and reliable technical support, leading employees to further recognize the collaborative value of AI \cite{McGrath}. This sense of trust not only bolsters employees' confidence in task execution but also motivates them to engage more actively in innovative exploration, ultimately exerting a positive influence on both job performance and innovation outcomes.

This trust not only enhances employees' confidence in performing tasks but also motivates them to engage more actively in innovative exploration, thereby exerting a positive impact on job performance and innovation performance.

Therefore, the following hypotheses are proposed:
$H_{1a}$: AI trust mediates the relationship between human-machine collaborative applications and job performance.
$H_{1b}$: AI trust mediates the relationship between human-machine collaborative applications and employee innovation.

Autonomy strengthens the shared awareness and collaborative intentions between employees and AI by enhancing the accuracy and independence of task execution, leading employees to place greater trust in the capabilities and roles of AI \cite{Bhaskara}. This competence-based trust not only provides employees with essential psychological resources to cope with challenges during technological change but also further enhances work motivation and efficacy by reducing workload and increasing the sense of support. Anthropomorphism facilitates emotional connections between employees and AI through highly simulated appearances, emotional expressions, and natural interactions, significantly mitigating friction during the collaboration process \cite{Munnukka}. This type of emotional trust helps employees adapt more quickly to collaborative environments and reduces the cognitive stress associated with uncertainty, thereby enhancing collaborative tacit understanding, promoting innovative exploration, and ultimately improving job performance and innovation outcomes.
$H_{2a}$: AI trust mediates the relationship between autonomy and job performance.
$H_{2b}$: AI trust mediates the relationship between autonomy and employee innovation.
$H_{3a}$: AI trust mediates the relationship between anthropomorphism and job performance.
$H_{3b}$: AI trust mediates the relationship between anthropomorphism and employee innovation.

By enhancing their understanding and sense of control over technology, employees further strengthen their trust in AI. Employees with high levels of AI literacy are better able to understand the operational logic and limitations of AI, thereby establishing a rational trust based on cognition \cite{Huang}. This trust bolsters their confidence in utilizing AI to handle complex tasks and conduct innovative work, which in turn contributes to the improvement of job performance and innovation levels. In contrast, employees' awareness of an AI crisis can generate negative emotions such as fear and anxiety. When employees perceive a resource threat and feel they lack the capability to cope with new technologies, they often exhibit lower levels of AI trust.

As a key psychological resource, AI trust not only enhances employees' sense of autonomy and control—promoting human-machine collaboration and improving job performance—but also helps employees better understand and adapt to human-machine collaboration within the organizational environment. By reducing perceptions of uncertainty, it incentivizes innovative behavior. Consequently, the following hypotheses are proposed:
$H_{4a}$: AI trust mediates the relationship between employee AI literacy and job performance.
$H_{4b}$: AI trust mediates the relationship between employee AI literacy and employee innovation.
$H_{5a}$: AI trust mediates the relationship between employee AI crisis awareness and job performance.
$H_{5b}$: AI trust mediates the relationship between employee AI crisis awareness and employee innovation.

2.4.3 人工智能信任与工作不安全感的中介作用比较

According to Conservation of Resources (COR) theory, individuals exhibit an asymmetry in their perception of resource gains versus resource losses (Hobfoll, 1989). Within collaborative systems, trust in AI represents more than just a positive belief in competence and benevolence; it signifies a fundamental expectation of continuous resource expansion and technological empowerment. Consequently, its impact may transcend short-term fluctuations, manifesting as a more profound and stable influence.

Empirical research demonstrates that employees' trust in AI significantly promotes technology adoption, collaboration satisfaction, and innovative intentions (Tams, 2018). These effects stem from the psychological safety and cognitive openness fostered by trust, which help build long-term, positive human-AI collaborative relationships. In contrast, while job insecurity may trigger immediate defensive behaviors and resource protection responses, its effects often diminish as individuals adapt or receive organizational support, exhibiting strong situational dependency (Shoss, 2017). That is to say, in the process through which human-AI collaborative systems influence work effectiveness, AI trust—as a gain-oriented psychological resource—may exert a stronger and more stable mediating effect than loss-oriented job insecurity. Therefore, we propose the following hypotheses:

H1: In the impact of collaborative systems on job performance, the mediating effect of AI trust is stronger than the mediating effect of job insecurity.
H2: In the impact of collaborative systems on employee innovation, the mediating effect of AI trust is stronger than the mediating effect of job insecurity.

2.5 潜在因素的调节作用

When examining the moderating effects of employee gender, age, type, industry attributes, and cultural background, data limitations prevent a differentiated analysis of certain moderating variables across all dimensions. Consequently, this study follows the approach of Duan Chenglong et al. (2025) by treating the human-machine collaboration system as an integrated construct. This construct encompasses three core dimensions—human-machine collaboration characteristics, task characteristics, and employee characteristics—to systematically grasp the holistic nature of human-machine collaboration and its underlying mechanisms at a theoretical level.

From a theoretical perspective, the moderating effect of gender on resource perception and transformation is closely linked to situational factors shaped by social culture. In the context of technical collaboration, traditional views on the gendered division of labor may lead organizations to develop implicit gender biases in resource allocation, granting certain groups more opportunities for deep interaction with AI \cite{VenkateshMorris2000}. In practical scenarios, these groups are often predominantly male employees. Due to societal gendered perceptions of technical workers, they frequently receive more technical training resources and opportunities to participate in high-autonomy tasks \cite{Russo}. Continuous technical practice allows them to accumulate extensive adaptive experience, making it easier for them to develop trust in artificial intelligence. This trust encourages them to integrate technology deeply into their workflows, efficiently transforming technical resources into psychological efficacy by optimizing task execution. As their proficiency grows, they ultimately achieve improved job performance and breakthroughs in innovative capabilities.

In human-machine collaboration scenarios, this group can rapidly adapt to technological changes, viewing AI as a vital tool for gaining resource advantages. Through resource accumulation, they effectively mitigate job insecurity and further consolidate work efficacy. Furthermore, research indicates that in human-machine collaboration roles within mechanical R&D, the proportion of male employees is significantly higher than that of females; their acceptance of technology is also higher, and they are more inclined to engage in challenging work, which further stimulates their motivation and creativity. In contrast, due to long-standing gender stereotypes in technical fields, female employees face numerous constraints in resource acquisition. They often must exert extra effort at work, not only to complete their primary tasks but also to contend with implicit societal questioning of their technical competence. This dual pressure forces them to consume more psychological resources to cope with external evaluations and challenges to self-identity during the collaboration process. Particularly in contexts where AI triggers professional crisis awareness, concerns about technological substitution and a lack of social identity can easily lead to the dissipation of psychological resources, thereby weakening work efficacy. Consequently, the following hypotheses are proposed:
$H_1$: Gender moderates the relationship between the human-machine collaboration system and job performance. The higher the proportion of male employees, the stronger the impact of the collaboration system on job performance.
$H_2$: Gender moderates the relationship between the human-machine collaboration system and employee innovation. The higher the proportion of male employees, the stronger the impact of the collaboration system on employee innovation.

2.5.2 年龄的调节效应

From a theoretical perspective, the impact of age on resource perception and transformation is essentially the result of the interaction between individual life-cycle characteristics and organizational contexts. When faced with resource threats or gains brought about by technology, individuals adopt differentiated resource acquisition and allocation strategies based on the resource reserves characteristic of their specific age stage, which in turn affects the effectiveness of human-machine collaborative systems \cite{Hobfoll, 2001}. In collaborative systems, different age groups exhibit significant differences in their resource transformation paths. Older employees, relying on long-accumulated domain knowledge and practical experience, tend to view technology as a complementary tool to compensate for declines in physical strength and reaction speed. The rule-following consciousness and steady work style developed throughout their careers make them more likely to integrate technology into existing workflows as a deterministic support for auxiliary decision-making. This cognitive pattern may strengthen their psychological resource reserves—for example, using collaboration to verify professional judgments—thereby enhancing their trust in artificial intelligence and their sense of professional competence. Research indicates that the stable trust older employees place in technology can significantly reduce job insecurity, thereby improving the continuity and precision of task execution \cite{Huang, 2022}. According to the resource conservation logic of theory, older employees' trust in technology may continuously improve the robustness of job performance by reducing the risk of resource loss.

In contrast, younger employees, drawing on the technological sensitivity of digital natives and open, inclusive cognitive traits, view technology as a breakthrough tool for stimulating innovative potential \cite{Dutta, 2022}. Driven by the autonomy and anthropomorphic characteristics of technology, younger employees are more likely to break free from the constraints of traditional work paradigms. By rapidly absorbing new technical knowledge, they transform technological resources into psychological resources such as creative thinking and self-efficacy. Furthermore, younger employees exhibit a stronger willingness to use emerging technological tools; they are more inclined to explore new technologies and accept new ways of working, potentially significantly increasing innovative output by actively expanding their resource boundaries. Based on this, the following hypotheses are proposed:

H1: Age moderates the relationship between human-machine collaboration and job performance. Specifically, the older the employee, the stronger the impact of the human-machine collaborative system on job performance.

H2: Age moderates the relationship between human-machine collaboration and employee innovation. Specifically, the younger the employee, the stronger the impact of the human-machine collaborative system on employee innovation.

2.5.3 员工类别的调节效应

In this study, organizational employees in the sample are categorized into non-knowledge workers and knowledge workers based on their specific knowledge domains. Non-knowledge workers primarily include employees with lower levels of education, such as frontline staff in service industries like hospitality and retail, as well as those engaged in repetitive manual labor. In contrast, knowledge workers refer to employees involved in knowledge processing and information management, such as professionals in high-tech industries, medical pharmaceuticals, the internet sector, and government or public institutions. Due to differences in the nature of their work and their resource endowments, these two categories of employees exhibit significantly different behaviors and outcomes within human-AI collaboration systems. Knowledge workers, leveraging their high educational attainment and deep knowledge reserves, view AI as an empowering tool for knowledge deepening and innovative breakthroughs.

When functions such as data analysis and complex task management are embedded into their workflows, knowledge workers optimize knowledge allocation by integrating technical resources, thereby stimulating creative thinking \cite{2023}. This collaborative mode reinforces the employees' perception that AI can extend their professional capabilities, leading them to more proactively explore the application of new technologies in innovative scenarios. Based on the resource gain logic of Conservation of Resources (COR) theory, knowledge workers continuously accumulate knowledge capital through AI collaboration, deepening their trust in artificial intelligence and forming a positive cycle of innovation capability enhancement. In contrast, non-knowledge workers, constrained by limited knowledge resources and physical energy, tend to view AI technology as a tool for replacing repetitive tasks and ensuring efficiency \cite{Chowdhury}. Automated processes and intelligent auxiliary tools effectively compensate for their resource deficiencies in standardized operations; by reducing work intensity and operational errors, these tools significantly improve task execution efficiency. This collaboration mode directly alleviates the employees' workload, allowing them to reinvest released psychological and physical resources into their work to improve job performance. Following the resource preservation logic of COR theory, AI collaboration for non-knowledge workers effectively mitigates the risk of resource loss, achieving a steady improvement in job performance. Therefore, the following hypotheses are proposed:

H1: Employee category moderates the relationship between human-AI collaboration and job performance. Specifically, the impact of human-AI collaboration systems on job performance is stronger for non-knowledge workers than for knowledge workers.

H2: Employee category moderates the relationship between human-AI collaboration and employee innovation. Specifically, the impact of human-AI collaboration systems on employee innovation is stronger for knowledge workers than for non-knowledge workers.

2.5.4 行业属性的调节效应

From a theoretical perspective, industry attributes exert a significant influence on the acquisition, allocation, and utilization of resources within human-AI collaboration systems. This study categorizes sample enterprises into three types based on industry attributes—high-tech, manufacturing, and service industries—to emphasize the differences in production, service, and value creation across different sectors. The high-tech industry is centered on knowledge-intensive activities, where production processes rely heavily on the innovation and iteration of complex knowledge \cite{Osborne,}. In this sector, employees can utilize AI to handle repetitive tasks and optimize complex data management, thereby freeing up psychological and temporal resources to focus on creative thinking and high-value activities. This reallocation of resources significantly reduces the rate of resource dissipation and enhances employees' innovative capabilities and work efficiency through resource concentration. Furthermore, employees in the high-tech industry typically possess higher levels of knowledge reserves and technical adaptability, enabling them to demonstrate a stronger sense of trust in AI during dynamic resource transformation, which further strengthens the efficacy of human-AI collaboration \cite{Osborne,}. In contrast, the manufacturing industry is primarily characterized by standardized production, where labor costs and efficiency are the core concerns.

In manufacturing, AI is mostly applied to the automation of production processes. While this can reduce operational errors and improve efficiency, the inherent rigidity of production workflows makes it difficult for employees to achieve large-scale knowledge innovation through AI. Meanwhile, the value creation process in the service industry is highly dependent on employee quality and customer feedback. This may lead service employees to focus more on how AI technology affects interpersonal interactions, as it could potentially undermine the human touch of the service and lead to decreased customer satisfaction. Such industry characteristics are likely to trigger concerns among employees regarding the loss of control over service quality, causing them to experience greater stress and anxiety during human-AI collaboration, which in turn reduces their work performance. Consequently, the following hypotheses are proposed:

H1: Industry attributes moderate the relationship between human-AI collaboration systems and work performance. Specifically, the impact of human-AI collaboration systems on work performance is stronger in the high-tech industry compared to the manufacturing and service industries.

H2: Industry attributes moderate the relationship between human-AI collaboration systems and employee innovation. Specifically, the impact of human-AI collaboration systems on employee innovation is stronger in the high-tech industry compared to the manufacturing and service industries.

2.5.5 文化背景的调节效应

Theoretical and cultural backgrounds, serving as macro-institutional resources, profoundly influence employees' resource perception, acquisition strategies, and conversion efficiency regarding collaboration. By shaping individual cognitive frameworks and behavioral norms, different cultural value systems lead employees to exhibit differentiated resource dynamic patterns during human-machine collaboration. In Western cultures, the tendencies toward individualism and low power distance emphasize individual value and autonomy; consequently, employees are more likely to perceive AI as a tool for resource gain. This cultural environment encourages employees to proactively apply technology autonomously—for instance, by utilizing data analysis tools to optimize personal workflows—thereby freeing up time and cognitive resources to invest in innovative activities \cite{DarwishHuber}. Simultaneously, employees view AI as a professional development opportunity, emphasizing its empowering role in personal competence and its ability to enhance individual efficiency and innovation. In contrast, the collectivism and high power distance prevalent in Eastern cultures place greater emphasis on team collaboration and hierarchical relationships. Within this cultural context, employees typically rely more on guidance from managers or the organization regarding AI application, tending to view AI as a tool for enhancing team effectiveness rather than merely a means of personal resource gain.

In such settings, employees may worry that over-reliance on AI will lead to the devaluation of their own skills or the weakening of interpersonal collaborative relationships, which directly intensifies job insecurity. Due to concerns over resource loss, employees in Eastern cultures often exhibit a more conservative attitude, demonstrating less autonomy and innovative drive. Therefore, by influencing employees' perceptions of resource gain or loss, cultural background moderates the relationship between human-AI collaboration systems and employee innovation.

Consequently, the following hypotheses are proposed:
H1: Cultural background moderates the relationship between human-AI collaboration systems and job performance. Specifically, the impact of human-AI collaboration systems on job performance is stronger in Western cultural contexts compared to Eastern cultural contexts.
H2: Cultural background moderates the relationship between human-AI collaboration systems and employee innovation. Specifically, the impact of human-AI collaboration systems on employee innovation is stronger in Western cultural contexts compared to Eastern cultural contexts.

3 研究方法

This study follows the meta-analysis procedures established by Lipsey and Wilson, which primarily include the following steps:

1. Literature Search and Screening

A comprehensive search was conducted across multiple academic databases to identify relevant empirical studies. Strict inclusion and exclusion criteria were applied to ensure the quality and relevance of the literature, focusing on studies that provided sufficient statistical data for effect size calculation.

2. Data Extraction and Coding

For each included study, key information was systematically extracted and coded. This process included recording study characteristics (such as author, publication year, and sample size), methodological details, and the specific statistical outcomes required to measure the relationships between variables.

3. Effect Size Calculation

To standardize the findings across different studies, effect sizes were calculated for each independent sample. Depending on the nature of the data provided in the original papers, appropriate metrics (such as Pearson's $r$ or Cohen's $d$) were utilized and subsequently transformed into a common metric to allow for aggregate analysis.

4. Statistical Analysis and Synthesis

The meta-analysis employed both fixed-effects and random-effects models to synthesize the data. This stage involved calculating the weighted mean effect size, assessing the confidence intervals, and performing heterogeneity tests (such as the $Q$ statistic and $I^2$) to determine the consistency of findings across the sampled literature.

5. Bias Assessment and Sensitivity Analysis

To ensure the robustness of the results, potential publication bias was evaluated using methods such as funnel plots or fail-safe numbers. Additionally, sensitivity analyses were performed to determine if the overall conclusions were significantly influenced by any single study or specific methodological choices.

3.1 文献检索与筛选

This study conducted a systematic literature search on human-AI collaboration. Building upon the search strategies employed in reviews by scholars such as Vaccaro and Jiang Jianwu, core keywords and thematic statements were selected. For Chinese databases, including CNKI, VIP, and Wanfang, thematic searches were conducted using keywords such as "human-machine collaboration," "human-machine interaction," and "human-machine relationship," combined with search strategies involving "artificial intelligence" and "employees."

For English databases, including Google Scholar, EBSCO, and ScienceDirect, precise matching searches were implemented using core combinations of "human-AI/robot/machine collaboration," "human-AI/robot/machine interaction," "human-AI/robot/machine relationship," and "human-AI/robot/machine cooperation," paired with "AI/artificial intelligence" and "employee/worker."

To ensure the breadth and depth of the research, this study also performed targeted searches for representative scholars in the field of human-AI collaboration to identify their published literature relevant to the current research topic.

Furthermore, following the approach of Su Tao et al. (2024), a systematic search was conducted for conference papers, dissertations, and working papers from major academic conferences in the fields of organizational behavior and management, both domestically and internationally.

The starting point for the literature search was set at [Year], as research following this period has focused more intensively on the impact of human-AI collaboration on employee performance and innovation, providing a more relevant theoretical foundation for this study. The search was conducted through [Month], resulting in a total of [Number] relevant documents, including [Number] Chinese and [Number] English publications.

Following the initial search, literature was screened according to the following criteria: (1) the study must simultaneously address [Variables]; (2) the literature must be empirical, excluding reviews, case studies, and purely theoretical papers; and (3) the literature must report sample sizes and include correlation coefficients or provide sufficient data to calculate them, such as $t$-values or Chi-square statistics.

Additionally, the literature must be based on independent studies using different samples to avoid duplicate publication. After screening and exclusion, this study compiled a total of [Number] Chinese and English documents ([Number] Chinese, [Number] English), encompassing [Number] independent empirical studies and [Number] samples, with a total of [Number] effect sizes.

To test mediation effects, correlation coefficients between other variables were also required. Following the methodology of Li Chaoping et al. ([Year]), correlation coefficients for job insecurity with job performance, job insecurity with employee innovation, job insecurity with trust, and job performance with employee innovation were obtained from the research of Sverke and Cheng, respectively. For variables where correlation coefficients were not found, this study supplemented the data by collecting [Number] additional papers and conducting a meta-analysis to obtain the necessary coefficients. The literature search and screening process is illustrated in [FIGURE:1].

Chinese Keywords: Human-machine collaboration, human-machine interaction, human-machine relationship, human-machine synergy, artificial intelligence, and employees.
Databases: CNKI, VIP, Wanfang, China Doctoral/Master's Dissertations Full-text Database.
English Keywords: human-AI/robot/machine collaboration/interaction/relationship/cooperation, AI/artificial intelligence, employee/worker.
Databases: Google Scholar, EBSCO, ScienceDirect.
Other Sources: Representative scholar searches; relevant conference papers, dissertations, and working papers in the fields of organizational behavior and management.

Total number of documents initially retrieved: $n = 4035$ (Chinese: $n = 1365$, English: $n = 2670$).

Documents excluded: $n = 2811$.

Number of documents after initial screening of titles, keywords, and abstracts: $n = 1224$ (Chinese: $n = 306$, English: $n = 918$).

Documents excluded: $n = 1039$.

Number of documents for full-text eligibility review: $n = 185$ (Chinese: $n = 76$, English: $n = 109$).

Documents excluded: $n = 106$.

Documents included in direct effect analysis: $n = 79$ (Chinese: $n = 21$, English: $n = 58$).

Empirical articles on human-AI collaboration applications and [Variable]: [Number]; empirical articles on employee innovation: [Number]. Empirical articles on AI autonomy and [Variable]: [Number]; empirical articles on employee innovation: [Number].

Empirical articles on AI anthropomorphism and [Variable]: [Number]; empirical articles on employee innovation: [Number]. Empirical articles on [Variable]: [Number]; empirical articles on employee innovation: [Number]. Empirical articles on AI crisis awareness and [Variable]: [Number]; empirical articles on employee innovation: [Number].

Among these, duplicate documents involving multiple dimensions of human-AI collaboration systems in relation to job performance and employee innovation were identified. The final number of documents used for the Meta-Analytic Structural Equation Modeling (MASEM) analysis was [Number].

3.2 数据编码与处理

This study developed a comprehensive coding manual, and two researchers independently conducted the coding process. From the screened literature, we extracted research information (such as authors, titles, and journals) and effect size statistics (including correlation coefficients, reliability coefficients, and sample sizes). Although most empirical studies contained only a single independent sample, several papers included multiple independent samples; in these cases, each sample was coded separately.

Furthermore, the study detailed the coding of five potential moderating variables: employee gender, age, employee category, industry attributes, and cultural background. During the initial coding review, the inter-rater agreement reached 86.74%. The primary reasons for inconsistencies were clerical coding errors and differing interpretations of specific coding criteria. Following a thorough review and discussion, all discrepancies were resolved, resulting in a finalized and complete coding table.

4.1 发表偏倚检验

To ensure the robustness of the results, this study employed the Fail-Safe N ($N_{fs}$) test, Egger’s regression coefficient test, and Rank Correlation tests to further evaluate potential publication bias. As shown in [TABLE:1], the results of the Fail-Safe N and correlation tests did not reach statistical significance ($p > 0.05$), indicating that there is no serious publication bias in this study.

Egger’s test results suggested a potential degree of publication bias between AI crisis awareness and employee performance. Consequently, a p-curve analysis was further employed to examine the impact of publication bias on the meta-analytic results regarding AI crisis awareness and employee innovation. The data exhibited a right-skewed distribution, with significant samples supporting the existence of a true effect. Since the significant results were primarily concentrated at low p-values, this suggests that the study does not suffer from severe publication bias. The Egger’s test also estimated the number of unpublished studies required to nullify the observed effect.

Employee characteristics; AI crisis awareness; Fail-safe N ($N_{fs}$) represents the fail-safe coefficient; $k$ represents the number of effect sizes.

4.2 同质性检验与主效应分析

Homogeneity tests typically utilize statistical indicators to evaluate the level of homogeneity within a sample. When the $Q$ value is significant, the sample exhibits significant heterogeneity, necessitating the use of a random-effects model; otherwise, a fixed-effects model is employed. As shown in [TABLE:N], the results for all variables indicate $p < 0.001$, demonstrating clear heterogeneity across the variables. Furthermore, the $I^2$ values for all variables exceed the threshold, confirming significant heterogeneity and justifying the adoption of the random-effects model. Simultaneously, the main effect analysis reveals that the point estimates are all statistically significant. These results indicate that human-machine collaborative applications, anthropomorphism, and employee self-efficacy have positive impacts on job performance, while AI crisis awareness has a negative impact on job performance. Similarly, human-machine collaborative applications, anthropomorphism, and employee self-efficacy positively influence employee innovation, whereas AI crisis awareness negatively influences it. Thus, hypotheses H1a through H3d are supported.

Homogeneity Test and Main Effect Analysis

In this analysis, $k$ represents the number of effect sizes; $N$ denotes the number of independent samples; $RE$ refers to the random-effects model; $Q$ is the homogeneity test statistic; $df(Q)$ represents the degrees of freedom; $p$ is the significance level; $I^2$ indicates the proportion of observed variation attributable to real differences in effect sizes; $Tau^2$ represents the between-study variance used for weight calculation; and $SE$ denotes the standard error.

4.3 中介效应检验

The mediation effect test was conducted using a meta-analytic structural equation modeling (MASEM) approach. This analysis examined the potential mediating effects of job insecurity and trust in artificial intelligence (AI) on the relationship between various indicator variables within the human-AI collaboration system and the outcomes of job performance and employee innovation. The process consisted of two primary stages. In the first stage, a pooled correlation matrix was obtained through multivariate meta-analysis. In the second stage, Mplus software was used to input this pooled correlation matrix into a structural equation model to test the mediation model. Based on the results of the mediation analysis, Monte Carlo confidence interval tests for indirect effects were performed using R software.

$X_1$ represents Human-AI Collaboration (HAC) applications; $X_4$ represents employee AI Knowledge, Skills, and Abilities (KSA); $X_5$ represents AI Awareness (AIA); $M_1$ represents Job Insecurity (INS); $M_2$ represents Trust in AI (AIT); $Y_1$ represents Job Performance (JP); $Y_2$ represents Employee Innovation (IP). The scales for job insecurity were adapted from Cheng, Sverke, Hellgren, and Naswall \cite{cite_key}; unstandardized correlation coefficients, the number of independent samples, and total sample sizes were calculated for this study.

*** denotes $p < 0.001$, ** denotes $p < 0.01$, * denotes $p < 0.05$; the same applies hereafter.

The results indicate that job insecurity is significantly and negatively correlated with both job performance and employee innovation. Conversely, trust in AI is significantly and positively correlated with both job performance and employee innovation.

Human-AI Collaboration (HAC) Applications

AI awareness significantly and positively influences employee job insecurity, while employee KSA significantly and negatively influences job insecurity. Human-AI collaboration applications significantly and positively influence trust in AI, whereas employee AI

awareness significantly and negatively influences trust in AI ($\beta = -0.46, p < 0.001$).

As shown in the results for human-AI collaboration applications, job insecurity, trust in AI, and AI awareness, the mediation analysis regarding the mechanisms influencing employee work effectiveness within human-AI collaboration systems reveals several key findings. The indirect effects of human-AI collaboration applications, anthropomorphism, employee KSA, and AI awareness on job performance through job insecurity were significant, as were the indirect effects on employee innovation. Job insecurity was found to play a partial mediating role, thus supporting the hypothesis. Furthermore, the indirect effects of human-AI collaboration applications, anthropomorphism, employee KSA, and AI awareness on job performance through trust in AI were significant, as were the indirect effects on employee innovation. Trust in AI also played a partial mediating role. This suggests that the indicator variables of the human-AI collaboration system can produce dual-sided effects—both negative and positive—by influencing employee job insecurity and trust in AI through two distinct pathways.

The Bootstrap 95% confidence intervals for the indirect effects are as follows: IND1 (HAC-INS-JP): [0.01, 0.03]; IND2 (HAC-AIT-JP): [0.13, 0.15]; IND3 (HAC-INS-IP): [0.01, 0.02]; IND4 (HAC-AIT-IP): [0.15, 0.17]; IND5 (AUT-INS-JP): [0.02, 0.04]; IND6 (AUT-AIT-JP): [0.20, 0.22]; IND7 (AUT-INS-IP): [0.07, 0.06]; IND8 (AUT-AIT-IP): [0.18, 0.20]; IND9 (ANT-INS-JP): [0.00, 0.01]; IND10 (ANT-AIT-JP): [0.15, 0.17]; IND11 (ANT-INS-IP): [0.01, 0.02]; IND12 (ANT-AIT-IP): [0.17, 0.19]; IND13 (KSA-INS-JP): [0.01, 0.02]; IND14 (KSA-AIT-JP): [0.17, 0.19]; IND15 (KSA-INS-IP): [0.01, 0.02]; IND16 (KSA-AIT-IP): [0.19, 0.21]; IND17 (AIA-INS-JP): [0.05, 0.07]; IND18 (AIA-AIT-JP): [-0.20, -0.18]; IND19 (AIA-INS-IP): [0.06, 0.05]; IND20 (AIA-AIT-IP): [-0.22, -0.20]. Total indirect effects: TOTALIND1 (HAC-JP): [0.13, 0.14]; TOTALIND2 (HAC-IP): [0.11, 0.13]; TOTALIND3 (AUT-JP): [0.19, 0.21]; TOTALIND4 (AUT-IP): [0.17, 0.19]; TOTALIND5 (ANT-JP): [0.14, 0.16]; TOTALIND6 (ANT-IP): [0.10, 0.12]; TOTALIND7 (KSA-JP): [0.16, 0.18]; TOTALIND8 (KSA-IP): [0.26, 0.28]; TOTALIND9 (AIA-JP): [-0.15, -0.13]; TOTALIND10 (AIA-IP): [-0.28, -0.26]. Finally, following the method of Wen et al. \cite{cite_key}, a comparative test of mediation effects was conducted. When two mediation paths have opposite signs, the difference between the absolute values of the effects should be used for comparison. Compared to job insecurity, human-AI collaboration applications, anthropomorphism, and employee KSA exert a stronger positive influence on job performance through trust in AI; the differences in mediation effects between trust in AI and job insecurity were [0.13, 0.15], [0.19, 0.21], [0.14, 0.16], and [0.16, 0.18], respectively. Similarly, these variables exert a stronger positive influence on employee innovation through trust in AI, with mediation effect differences of:

[0.12, 0.13], [0.17, 0.19], [0.10, 0.12], and [0.13, 0.14]. Furthermore, compared to job insecurity, AI awareness exerts a stronger negative influence on both job performance and employee innovation through trust in AI, with differences of [-0.19, -0.17] and [-0.17, -0.15], respectively. Thus, the hypothesis is supported.

4.4 调节效应检验

This study examines the moderating effects of two demographic characteristics—gender and age—on the relationship between human-AI collaboration systems and employee work effectiveness. Meta-regression analyses were conducted using the proportion of male employees and the mean age as predictor variables.

The results of the moderation analysis are presented in [TABLE:N]. The findings indicate that employee gender does not significantly moderate the relationship between the various dimensions of human-AI collaboration systems and either job performance or employee innovation (the confidence intervals for $\beta$ include 0). Similarly, employee age does not exert a significant moderating effect on these relationships (the confidence intervals for $\beta$ also include 0). Consequently, these hypotheses were not supported. To further explore the relationships between human-AI collaboration systems, job performance, and employee innovation, this study follows the methodological approaches of Duan Chenglong et al. (2025) and others (2024) to test the moderating effects of contextual variables. As shown in [TABLE:N], three contextual variables were selected from micro, meso, and macro perspectives—employee category, industry attributes, and cultural background—to examine their moderating roles. The results are as follows:

At the micro level, employee category significantly moderates the relationship between human-AI collaboration systems and job performance ($r = 0.21$ and $r = 0.44$ for knowledge workers and non-knowledge workers, respectively). These correlation coefficients indicate that human-AI collaboration systems have a stronger positive impact on the job performance of non-knowledge workers compared to knowledge workers. Furthermore, employee category also significantly moderates the relationship between human-AI collaboration systems and employee innovation ($r = 0.47$ and $r = 0.24$ for knowledge workers and non-knowledge workers, respectively). This suggests that human-AI collaboration systems have a more pronounced positive effect on the innovative behavior of knowledge workers than on non-knowledge workers. At the meso level, industry attributes significantly moderate the relationships between human-AI collaboration systems and both job performance and employee innovation. For job performance, the correlation coefficients across high-tech, manufacturing, and service industries are $0.57$, $0.38$, and $0.27$, respectively. For employee innovation, the corresponding coefficients are $0.50$, $0.34$, and $0.27$.

These results demonstrate that the impact of human-AI collaboration systems on job performance and employee innovation is significantly stronger in the high-tech industry than in the manufacturing and service industries, supporting the relevant hypotheses. At the macro level, cultural background significantly moderates the relationship between human-AI collaboration systems and job performance ($r = 0.51$ for Western culture and $r = 0.26$ for Eastern culture). This indicates that the impact on job performance is stronger in Western cultural contexts than in Eastern ones. However, cultural background does not significantly moderate the relationship between human-AI collaboration systems and employee innovation; thus, this specific hypothesis was not supported.

Human-AI Collaboration Application: 95% CI [-0.8, 0.53], [-1.80, 1.35], [-1.01, 0.29], [-2.54, 0.51], [-1.52, 1.52], [-0.56, 1.67], [-1.0, 0.29]. AI Crisis Awareness: 95% CI [-1.61, 0.16], [-1.8, 0.46]. Human-AI Collaboration Application: $\beta$ coefficients [-0.01, 0.02], [-0.00, 0.07], [-0.02, 0.01], [-0.06, 0.02], [-0.03, 0.01], [-0.01, 0.05]. AI Crisis Awareness: $\beta$ coefficients [-0.01, 0.03], [-0.06, 0.03].

5 结论与讨论

This meta-analysis, based on 68 independent samples from domestic and international literature, examines the impact of Human-AI Collaboration (HAIC) on employee work effectiveness. The results indicate that HAIC applications, anthropomorphism, and employee AI self-efficacy have a significant positive impact on work effectiveness. This may be attributed to the fact that HAIC applications allow employees to fully leverage the potential of AI tools, thereby enhancing job performance and innovation capabilities \cite{Soomro, Pitafi}. The autonomy of AI not only reduces human error but also improves work quality and efficiency; meanwhile, by sharing repetitive tasks, it frees up employee time to focus on high-value activities and skill enhancement. Furthermore, AI anthropomorphism enhances the employee work experience through natural interaction methods, fostering problem-solving abilities and creative thinking \cite{Zhang}. Additionally, employees with high AI self-efficacy can more effectively apply AI technologies to improve performance and innovation due to their technical skills and understanding of AI \cite{AL-Khatib}. However, this study also found that AI crisis awareness has a significant negative impact on work effectiveness. Employees may develop resistance and distrust due to concerns that AI threatens their job security, which in turn reduces work motivation and inhibits efficiency and innovation \cite{2024}. Regarding the specific underlying mechanisms, this study found that in HAIC systems, employee work effectiveness is significantly influenced by the dual-path mediation of job insecurity and AI trust. In the complex ecosystem of human-machine collaboration, job insecurity often stems from the initial stages of technology introduction, where employees experience stress reactions due to the unknown nature of professional boundary reshaping, which impacts work effectiveness in the short term \cite{2024}.

However, as employees gradually adapt to technological developments and recognize the potential value of AI, their trust in AI increases, thereby compensating for and eventually exceeding the negative effects of insecurity. AI trust not only helps employees use AI tools more proactively but also enhances their sense of control over work tasks and their trust in the organization, thus promoting work effectiveness \cite{2024, Shahzad}. A comparative analysis revealed that the mediating effect of AI trust is stronger than that of job insecurity. Related research suggests that while job insecurity reduces performance, this negative effect is often phase-specific—a primary reaction to the shock of new technology. Over time, as cognitive understanding of AI deepens, the positive effects of established trust gradually emerge and become dominant, explaining why the mediating effect of trust is significantly stronger than that of insecurity. Furthermore, the study found that employee gender and age do not significantly moderate work effectiveness within HAIC systems, reflecting the combined effects of gender equality and technological ubiquity. In recent years, with the continuous advancement of gender equality, contemporary women have become comparable to men in terms of technical application and adaptability, weakening traditional gender stereotypes. Women's acceptance and learning capacity for new technologies in the workplace have approached or reached the levels of their male counterparts \cite{Huyer}. Additionally, HAIC systems provide more standardized and structured workflows, which may further reduce the impact of gender differences during the technological adaptation process. The non-significant moderating effect of age may be related to the narrowing of the intergenerational gap. With the popularization of information technology education and the promotion of lifelong learning in aging societies, the gap in technological adaptability across different age groups has significantly decreased \cite{Ranta, Ylinen}. Moreover, compared to knowledge workers, HAIC systems have a more significant effect on improving the job performance of non-knowledge workers. This may be because the work content of non-knowledge workers typically consists of repetitive and standardized tasks; the introduction of AI technology can significantly reduce their workload and improve efficiency. Simultaneously, AI technical support serves as an external resource providing real-time feedback, helping non-knowledge workers adapt to job requirements faster and compensating for deficiencies in skills and experience. Conversely, HAIC systems have a more significant effect on enhancing the innovation capabilities of knowledge workers compared to non-knowledge workers. This is likely because the work of knowledge workers involves high complexity and creativity; AI can provide data analysis, pattern recognition, and predictive support to stimulate their creative thinking. Furthermore, knowledge workers possess higher skill levels and technology acceptance, enabling them to better transform AI technology into innovative outcomes.

Regarding macro factors such as industry and culture, the moderating effect of HAIC is more significant in high-tech industries compared to the manufacturing and service sectors. This may be because high-tech enterprises often possess advanced technical infrastructure, higher R&D investment, and flexible organizational structures that effectively support the innovation needs and challenges of employees. The work environment in high-tech industries is typically characterized by uncertainty and complexity, which helps stimulate employee creativity and proactivity. At the same time, companies in these industries generally encourage innovation through their organizational culture and structure, allowing employees to fully utilize human-machine collaboration technologies to enhance work effectiveness. Cultural background only moderated the relationship between HAIC systems and job performance, failing to moderate the relationship between HAIC systems and employee innovation.

The reason may be that, regardless of Eastern or Western culture, organizational applications and resource allocations for human-machine collaboration are typically standardized, which reduces the moderating effect of cultural background on innovation \cite{Tenakwah}. Additionally, the evaluation of job performance is usually objective and quantitative, making the influence of cultural background more easily manifested through these standards. In contrast, the evaluation of innovation involves more subjective and complex factors, making the influence of cultural background more indirect.

Finally, globalization and the exchange between Eastern and Western cultures have significantly increased the acceptance of technology in Eastern cultures, leading to a gradual weakening of the impact of cultural differences on the relationship between human-machine collaboration and employee innovation.

5.1 理论贡献

Theoretical Contributions

The theoretical contributions of this study are as follows: First, it systematically explores the impact and differences of human-AI collaborative systems on employee work effectiveness. While existing research has largely focused on the positive impacts of human-machine collaboration—emphasizing the role of AI tools in enhancing performance \cite{Vaccaro}—there has been less focus on the "double-edged sword" effect on human workers. This study explicitly identifies the positive roles of human-AI collaboration applications, anthropomorphism, and employee engagement in driving work performance and innovation. These findings theoretically echo the model of "technology empowerment leading to capability release" proposed by He et al. Simultaneously, this research reveals the negative impact of AI crisis awareness on these outcomes. This discovery supplements psychological theories regarding employees in human-machine collaboration scenarios and reveals the dynamic balance of dual-path mediation.

This study finds that human-AI collaborative systems not only enhance employees' trust in artificial intelligence—thereby improving work performance and stimulating innovation—but may also lead to increased job insecurity, which in turn reduces performance and innovation. However, overall, the positive impact of human-AI collaboration on performance and innovation through AI trust is more significant. Although AI may trigger feelings of insecurity, the establishment of long-term trust can significantly offset these negative effects. This result extends the application of relevant theories within human-machine collaboration contexts and reveals a compensation mechanism for declining security, providing a new perspective for explaining the dynamic process of technology acceptance.

From micro, meso, and macro levels, this study clarifies the moderating roles of the individual, organizational, and environmental dimensions. At the micro-individual level, the research challenges traditional assumptions that gender and age dictate technological adaptation. It finds that the advancement of gender equality \cite{Huyer} and the popularization of lifelong learning \cite{Poquet_Laat} have weakened generational differences and narrowed the gap in technological adaptation across different gender and age groups. Furthermore, the study identifies distinct efficiency enhancement paths for knowledge-based and non-knowledge-based employees: non-knowledge workers significantly improve efficiency through the replacement of standardized tasks, while knowledge workers stimulate innovative thinking through AI assistance in creative tasks. At the meso-organizational level, the study confirms that high-tech industries—due to their sophisticated technical infrastructure and active innovation cultures—are better positioned to unleash the effectiveness of human-AI collaboration, thereby supplementing the moderating mechanisms of industry characteristics on technology application. In the macro-environmental context, cultural background was found to moderate performance but not innovation. Specifically, the moderation of performance by cultural background is explicit and context-dependent, whereas the impact on innovation is implicit and value-oriented. This provides a critical analytical dimension for understanding the cross-cultural adaptation of human-AI collaboration.

5.2 实践启示

Practical implications include the necessity for managers to acknowledge the dual effects brought about by human-AI collaborative systems. On one hand, organizations should foster a supportive culture and uphold human-centric philosophies. This can be achieved through training programs—such as establishing real-time feedback mechanisms and sharing success stories—to build trust and enhance employees' sense of control when utilizing artificial intelligence. On the other hand, it is essential to establish psychological safety buffers. Intervention strategies, including career development planning and reskilling initiatives, should be employed to alleviate employees' sense of crisis. By positioning AI as a partner for capability expansion rather than a replacement, organizations can ensure that employees maintain dominance over work autonomy and decision-making, thereby reducing the emergence of negative emotions.

Furthermore, managers should implement differentiated design and support schemes. Rather than treating all employees as a single category, managers should design distinct system application plans based on categorical differences among staff. For instance, for non-knowledge workers, the focus should be on the automation of repetitive tasks, providing standardized tools to improve efficiency while bridging skill gaps through real-time technical support.

In contrast, for knowledge workers, the emphasis should be on the innovative auxiliary functions of AI. Leveraging their high level of technology acceptance, organizations can design open collaborative interfaces to facilitate human-machine co-creation. By rationally allocating employee resources, organizations can form highly complementary teams that promote the flow and sharing of knowledge, ultimately achieving a dual enhancement of work performance and innovation capabilities.

Finally, the advantages of human-machine collaboration should be leveraged in a manner tailored to specific industry characteristics and cultural backgrounds. Governments should introduce policies that support the application of AI technology across various sectors, such as tax incentives and technical subsidies. Simultaneously, enterprises should increase investment in research and development to respond rapidly to technological iterations. By relying on robust technical infrastructure and a vibrant culture of innovation, organizations can drive the research and application of cutting-edge technologies to fully release the potential of human-machine collaborative efficiency.

5.3 研究不足与展望

This study has several unavoidable limitations that should be addressed in future research. First, the literature collection was limited to Chinese and English documents, excluding studies in other languages, which may lead to selection bias. Although this study included relevant results from public preprints and applied the same quality screening process as published literature, unpublished non-preprint documents were not included due to technical barriers in cross-database retrieval, access restrictions, and the risk of publication bias. Future research could explore collaborations with academic institutions to obtain internal reports or utilize translation tools to expand multilingual analysis, while remaining cautious regarding quality verification and sample representativeness of non-public documents. Furthermore, while this study systematically categorized the interaction system into human-AI collaborative applications, autonomy, and anthropomorphism, and examined employee characteristics and AI crisis awareness, it did not comprehensively cover all potentially influential variables. For instance, factors from the Technology Acceptance Model (TAM), such as perceived usefulness and perceived ease of use, may significantly impact the effectiveness of collaborative systems. Future research could further explore other machine and employee characteristics (e.g., AI transparency, employee personality), organizational contextual factors (e.g., AI readiness), and task types (e.g., interaction modes and task objectives).

Although meta-analysis can synthesize results from multiple studies, it is difficult to fully reveal the complex relationships between variables. To better understand these effects and their interactions, future research should consider combining meta-analysis with other statistical methods.

In addition, this study primarily focused on the impact of human-AI collaborative systems on job performance and employee innovation, without considering other outcome variables. For example, it did not examine how human-AI collaboration affects interpersonal collaboration and interaction within organizations \cite{2019}. Given that technological progress is often a major driver of the evolution of organizational interpersonal relationships \cite{Chen 2022}, and that the aggregate emergence of human capital resources for performance and innovation relies heavily on effective collaboration and skill complementarity based on interpersonal networks \cite{Huang}, future research could pay more attention to the substitution and competition between human-AI collaboration systems. Exploring how these dynamics affect organizational performance and innovation would be a significant and compelling topic.

This study explored the relationships between variables based on cross-sectional data and did not fully account for dynamic changes within the context of rapid technological iteration. As employees accumulate experience and organizational transparency mechanisms improve, the mediating paths of job insecurity and AI trust may exhibit dynamic evolutionary characteristics. Future research could employ longitudinal tracking or multi-wave research designs to construct dynamic models that reveal the patterns of change in human-machine collaboration mechanisms across different stages.

This study only considered the moderating effects of employee gender, age, employee type, industry attributes, and cultural background, while omitting factors such as AI type, person-job fit, and task workload. For example, the type of AI system (e.g., chatbots, large language models) may lead to differentiated impacts of human-AI systems on job performance and employee innovation. Future research should further examine the influence of these potential moderating variables through meta-analysis to provide a more comprehensive understanding.

Furthermore, in the studies included in the meta-analysis, the gender and age of respondents were often not randomly selected, which affects the rigor of the moderating effects. Therefore, the moderating effects of the variables in this study should be interpreted with caution.

This research focused on the moderating effects within the main effects and did not deeply explore their influence on mediating effects (moderated mediation). The primary reasons for this are: first, existing literature on moderated mediation in human-AI collaborative systems is relatively scarce, lacking a mature theoretical framework to serve as an analytical basis; second, limited by the data structure and statistical methods of meta-analysis, it is difficult to directly and effectively test moderated mediation models. Future research could combine empirical studies and utilize methods such as structural equation modeling (SEM) to deeply analyze the complex interaction mechanisms among these variables.

Acknowledgments: We would like to thank the two anonymous reviewers and the editors for their many valuable suggestions for the continuous improvement of this study. We also thank Associate Professor Chen Hongzhi from the School of Management, Fudan University, for his advice and support.

参考文献

(2022). The Impact of Human-Machine Relationships on Task Performance in Intelligent Workplaces: An Explanation Based on Self-Concept. Science & Technology Management Research, (19), (2019).

From "Grabbingism" to Collaborative Symbiosis: The Integration of Human Resource Management and Social Network Research. Human Resource Management Review, (2025). The Relationship Between Leadership Style and Employee Deviant Innovation: A Meta-Analysis. Science & Technology Progress and Policy. How the Use of AI Marketing Technology Affects the Digital Creativity of Sales Employees. Journal of Zhejiang University of Finance and Economics, (2021).

Research on the Impact of Digital Human-Computer Interaction Knowledge and Skills on Task Performance. Master's Thesis, Xi'an University of Science and Technology, (2004). Employee-Machine Relationships: A Study Based on Manufacturing Enterprises, (2022).

Research on the Impact of Human-Machine Collaboration Identification on Employee Job Performance. Master's Thesis, (2004). A Meta-Analysis of the Impact of Workplace AI Applications on Employees. Advances in Psychological Science, (10), (2022).

Research on the Non-linear Impact of Artificial Intelligence Effects on Employee Job Performance in Manufacturing Enterprises. Master's Thesis, Xi'an University of Technology, (2023). The Impact and Mechanism of Family-Supportive Supervisor Behavior on Employees: Evidence from a Meta-Analysis, (2024).

The Mechanism of Human-Computer Interaction's Influence on Innovative Work Behavior: Evidence from Smart Power Plants. Foreign Economics & Management, (10), (2023).

The Impact of Human-Machine Collaboration Degree on Employee Job Crafting in Application Scenarios: The Mediating Role of Psychological Empowerment. China Personnel Science, (2025). Research on the Impact of Technology Adoption on the Innovation Performance of Scientific Researchers. Studies in Science of Science, (2024).

Research on the Mechanism of Artificial Intelligence Technology Application on Employee Innovative Behavior, (17), (2024). Threat or Challenge? The Double-Edged Sword Effect of AI Use on Employee Innovation Performance. Journal of Zhejiang University of Finance and Economics, (2023). A Study on the Double-Edged Sword Effect of AI Use on Employee Work Behavior. Chinese Human Resource Development.

(2022). Research on the Impact of Human-Machine Relationship Perception on Job Crafting. Master's Thesis, University of Electronic Science and Technology of China, (2022). How the Application of Artificial Intelligence Technology Improves Employee Task Performance. Journal of Beijing Union University (Humanities and Social Sciences Edition), (2024).

Research on the Impact of Human-Machine Collaboration on Project Team Performance: A Moderated Mediation Model. Journal of Management, 86+94, (2024). New Drivers of Innovation: Research on the Impact of Human-Machine Relationships on Employee Innovative Behavior in the Intelligent Era. Western Forum on Economy and Management, 53+63, (2021).

Workplace Ostracism: A Meta-Analysis in the Chinese Context. Nankai Business Review, (2017). Evidence from Meta-Analysis. Nankai Business Review, (2024). A Meta-Analysis of the Double-Sided Effects of Narcissistic Leadership on Subordinate Effectiveness. Advances in Psychological Science, (2023).

Does Obedience Lead to Good Performance? The Mediating Role of Psychological Entitlement and the Moderating Role of Communication Openness. Business Economics and Administration, (2023). Research on the Impact of Introducing Artificial Intelligence on Employee Work Behavior. Master's Thesis, Beijing Foreign Studies University, (2022).

A Review of Research on the Impact of Workplace Human-Machine Collaboration on Employees. Foreign Economics & Management, (2022). Methodological Research on Mediating Effects in China. Advances in Psychological Science, 30(8):1692, (2021).

Human Resource Management in the Digital Era: A Perspective Based on Human-Technology Interaction, 216+13, (2022). Research on the Influencing Factors and Mechanisms of Human-Machine Team Performance. Master's Thesis, (2024).

A Dual-Path Study on the Impact of AI Device Use on Employee Innovative Behavior. Contemporary Manager, (2024). The Impact of Human-Machine Relationships on Employee Service Innovation Behavior in the Digital-Intelligence Context: Based on Social Cognitive Theory. Journal of Hubei University of Technology, (2024).

Human-Machine Interaction from a Systematic Perspective. Advances in Psychological Science, (2023). Gain or Loss? The Impact of AI Technology Application on Employee Innovative Behavior. Science & Technology Progress and Policy, (18), (2021).

The Impact of Robot Use on Job Performance and Its Mechanism: A Study of Employees in Low-to-Medium Skill Positions. China Soft Science, (2024). AI-Powered E-Learning for Lifelong Learners.

Impact Performance Knowledge Application. Sustainability (20), Sung, (2022). effect gender stereotypes artificial intelligence recommendations.

Journal Business Research Ajoudani, Zanchettin, Ivaldi, Albu-Sch ffer, Kosuge, Khatib, (2018).

Progress prospects human robot collaboration. Autonomous robots Alabed, Javornik, Gregory-Smith, (2022). anthropomorphism effect users' self-congruence self-ai integration: theoretical framework research agenda.

Technological Forecasting Social Change AL-Khatib, (2024).

Prioritizing factors generative artificial intelligence-based innovation adoption hospitality industry.

Management Decision

Atalla, El-Ashry, Mohamed Sobhi Mohamed, (2024). moderating ethical awareness relationship between nurses artificial intelligence perceptions, attitudes, innovative behavior: cross-sectional study.

Nursing Bechky, (2003). Sharing meaning across occupational communities: transformation understanding production

floor. Organization S cience , 14 (3), 312 – 330.

Bhaskara, Skinner, Loft, (2020) Agent transparency: review current theory evidence.

Transactions Human-Machine Systems Brougham, Haar, (2018).

Smart technology, artificial intelligence, robotics, algorithms STARA Employees perceptions future workplace.

Journal Management Organization Brynjolfsson, Mcafee, (2017).

Artificial intelligence, real. Harvard Business Review Chen, (2024).

Member performance human hybrid teams: perspective adaptability theory.

Information Technology People Chen, Mehra, Tasselli, Borgatti, (2022).

Network dynamics organizations: review research agenda.

Journal Management 48(6): Chen, Zhao, Wang, (2024 ffect emands rtificial ntelligence mbeddedness mployees erformance: oderated ouble- odel.

Behavioral Sciences (10), Cheng, Kong, (2023). Challenge hindrance? organizational artificial intelligence adoption influences employee crafting.

Journal Business Research Cheng, Chan, (2008). suffers insecurity? analytic review.

Applied Psychology Chowdhury, Budhwar, Joel-Edgar, Abadie, (2022).

AI-employee collaboration business performance: Integrating knowledge-based view, socio-technical systems organisational socialisation framework.

Journal Business Research Darwish, Huber, (2013). Individualism collectivism different cultures: cross-cultural study.

Intercultural Education Dell'Acqua, McFowland Mollick, Lifshitz-Assaf, Kellogg, Rajendran, Lakhani, (2023).

Navigating jagged technological frontier: Field experimental evidence effects knowledge worker productivity quality.

Harvard Business School Technology Operations Working Paper (24-013).

Visser, Monfort, Goodyear, Hara, Krueger, (2017). little anthropomorphism Effects trust, compliance, performance automated agents.

Human Factors Diab, Demiris, (2025). TICK: Knowledge processing infrastructure cognitive trust human robot interaction.

International Journal Social Robotics (2021). Employees challenge-hindrance appraisals toward stara awareness competitive productivity: micro-level Case.

International Journal Contemporary Hospitality Management, Dong, Tian, Wang, (2025). knowledge workers double-edged sword effects adoption innovative behavior.

Journal Knowledge Management Dutta, Mishra, Tyagi, (2023).

Augmented employee voice employee engagement using artificial intelligence- enabled chatbots: field study.

International Journal Human Resource Management (12), Felten, Seamans, (2021).

Occupational, industry, geographic exposure artificial intelligence: novel

potential uses. Strategic Management Journal (12), Frey, Osborne, (2017). future employment: susceptible computerisation.

Technological Forecasting Social Change Glikson, Woolley, (2020).

Human trust artificial intelligence: Review empirical research.

Academy Management Annals Good, Hughes, Kirca, McGrath, (2022). self-determination theory-based meta-analysis differential effects intrinsic extrinsic motivation salesperson performance.

Journal Academy Marketing Science Gull, Ashfaq, Aslam, (2023).

Workplace: Uncovering Impact Employee Well-being Cognitive Insecurity.

International Journal Business Economic Affairs (IJBEA), Hauptman, Schelble, McNeese, Madathil, (2023).

Adapt overcome: Perceptions adaptive autonomous agents human-AI teaming.

Computers Human Behavior Teng, Song, (2023). Linking employees challenge-hindrance appraisals toward service performance: influences crafting, insecurity knowledge.

International Journal Contemporary Hospitality Management Zheng, Chen, (2024). mixed blessing leaders artificial intelligence (AI)-oriented change behavior: implications employee performance unethical behavior.

Applied Research Quality Hengstler, Enkel, Duelli, (2016).

Applied artificial intelligence trust autonomous vehicles medical assistance devices.

Technological Forecasting Social Change Hobfoll, (1989).

Conservation resources: attempt conceptualizing stress.

American Psychologist Hobfoll, (2011). Conservation resources theory: implication stress, health, resilience Oxford handbook tress, ealth, oping, Huang, Chen, Chan, (2024).

Pairing anthropomorphized artificial agents: Leveraging employee creativity service encounters.

Journal Academy Marketing Science Huang, Gursoy, (2024). technology integration affect employees proactive service behaviors? transactional theory stress perspective.

Journal Retailing Consumer Services Huang, (2023).

Understanding continuance intention artificial intelligence anchor:

Based expectation confirmation theory. Systems Huang, Chen, Cheng, (2025).

Human capital resource emergence multiple project participation settings.

Academy Management Proceedings, Huyer, (2022). Breaking through silicon wall:

Gendered opportunities risks technologies. Gender, Technology Development Hwang, (2021, May).

IdeaBot: investigating social facilitation human-machine creativity.

Proceedings conference human factors computing systems Jeong, Jeong, Driving creativity AI-enhanced workplace: roles self-efficacy transformational leadership.

Current Psychology Fang, Liao, (2024). artificial intelligence augments employee creativity.

Academy

Management Journal (2024). Architecting future: exploring synergy AI-driven sustainable conscientiousness, employee engagement.

Discover Sustainability Jiang, Lavaysse, (2018). Cognitive affective insecurity: meta-analysis primary study.

Journal Management, (2014). Empirical

analysis

changes human creativity people humanoid robots their avatars.

Learning Collaboration Technologies. Designing Developing Novel Learning Experiences:

First International Conference, 2014, International 2014, Heraklion, Crete, Greece, 22-27, 2014, Proceedings, 281).

Springer International Publishing. Kahn, Kanda, Ishiguro, Gill, Shen, Ruckert, Gary, (2016, March).

Human creativity facilitated through interacting social robot.

ACM/IEEE International Conference Human-Robot Interaction (HRI) 180).

IEEE. Kang, Shin, (2023). Smart technology service employees crafting:

Relationship between STARA awareness, performance pressure, receiving giving help, crafting.

Journal Retailing Consumer Services Khan, Soomro, Pitafi, orkplace:

Driving employee performance through enhanced knowledge sharing engagement.

International Journal Human-Computer Interaction (17), Wirtz, (2022).

Service robots: applying social exchange theory better understand human-robot interactions.

Tourism Management (2023). heads better one?: effect student-AI collaboration students' learning performance.

TechTrends Kong, Chon, Yuan, (2024). artificial intelligence enhance hospitality employee innovation? roles exploration, trust, proactive personality.

Journal Hospitality Marketing Management Kumar, (2025).

Artificial intelligence (AI)-augmented knowledge management capability clinical performance: implications marketing strategies health-care sector.

Journal Knowledge Management Zhao, Zhang, Zhang, (2024). algorithmic management undermine employee creativity?: perspective focused theory.

Journal Organizational Computing (JOEUC), Bonn, (2019).

Hotel employee's artificial intelligence robotics awareness impact turnover intention: moderating roles perceived organizational support competitive psychological climate.

Tourism Management Zhang, (2024). autonomy human-robot collaboration affect hotel employees health outcomes? insecurity person-job International Journal Hospitality Management Zong, Cheng, (2024). mpact edical xplainable rtificial ntelligence urses nnovation ehaviour: tructural quation odelling pproach.

Journal Nursing Management Liang, Gong, (2022). Investigating double-edged sword effect awareness employee's service innovative behavior.

Tourism Management (2024). Meta-analytic review hospitality tourism employees creativity innovative behavior.

Tourism Management Lipsey, Wilson, (2001). Practical Publications Akbar, Chen, (2022, June). effect human-robot extroversion matching individual performance.

International Conference Human-Computer Interaction 570).

Springer International Publishing. Zhang, Wang, (2024). organizations contributes employee competitive advantage: moderating perceived organization support.

Technological Forecasting Social Change Tang, Koopman, McClean, Zhang, Cremer, (2022). conscientious employees intelligent machines: integrative approach inspired complementarity theory theory.

Academy Management Journal Marikyan, Papagiannidis, Rana, Ranjan, Morgan, (2022).

Alexa, about productivity impact digital assistants productivity.

Journal Business Research Marimon, Mas-Machuca, Akhmedova, (2024).

Trusting generative Catalyst employee performance engagement workplace.

International Journal Human-Computer Interaction (11), McGrath, Duenser, Lacey, Paris, Collaborative human-AI trust (CHAI-T): process framework active management trust human-AI collaboration.

Computers Human Behavior: Artificial Humans Mulcahy, Riedel, Keating, Beatson, Letheren, (2024).

Avoiding excessive service agent anthropomorphism: examining delivering news.

Journal Service Theory Practice Munnukka. Talvitie-Lamberg, Maity. (2022).

Anthropomorphism social presence human-virtual service assistant interactions: dialog length attitudes.

Computers Human Behavior Nelson, Irwin, (2014). Defining do-all again Occupational identity, technological change, librarian/Internet-search relationship.

Academy Management Journal (2017). Transformational leadership performance outcomes:

Analyses multiple mediation pathways. Leadership Quarterly Zhang, (2023).

Experimental evidence productivity effects generative artificial intelligence.

Science (6654), Oldham, Cummings, (1996). Employee creativity:

Personal contextual factors work. Academy Management Journal Paliga, (2022).

Human cobot interaction fluency cobot operators performance. mediating engagement: survey.

Robotics Autonomous Systems Paliga, (2023). relationships human-cobot interaction fluency performance satisfaction among Cobot operators moderating workload.

International Journal Environmental Research Public Health Papadopoulos, ster, Corrigan, Kappas, Castellano, (2016). relative positions proxemics affect engagement human-robot collaborative scenario?.

Interaction Studies Wang, Peng, (2024). Saving face:

Leveraging artificial intelligence based negative feedback enhance employee performance.

Human Resource Management Poquet, Laat, (2021). Developing capabilities:

Lifelong learning British Journal Educational Technology Presbitero, Teng-Calleja, (2022). attitudes career behaviors relating employees' perceived incorporation

artificial intelligence workplace: career self-management perspective.

Personnel Review, Wang, (2022). impact AI-enabled service attributes service hospitableness: employee physical psychological workload.

International Journal Contemporary Hospitality Management Ranta, Ylinen, (2023).

Board gender diversity workplace diversity: machine learning approach.

Corporate Governance: International Journal Business Society 23(5), Rebensky, Carmody, Ficke, Carroll, Bennett, (2022).

Teammates instead tools: impacts level autonomy mission performance human agent teaming dynamics multi-agent distributed teams.

Frontiers Robotics Russo, Romano, Clemente, Iacovone, Gladwin, Panno, (2025).

Gender differences artificial intelligence: artificial intelligence anxiety.

Frontiers Psychology Savela, Kaakinen, Ellonen, Oksanen, (2021).

Sharing robots: negative effect robot workers in-group identification team.

Computers Human Behavior Shahzad, Zahid, (2025). Exploring impact generative AI-based technologies learning performance through self-efficacy, fairness ethics, creativity, trust higher education.

Education Information Technologies Sharif, Zhang, Asif, Alshdaifat, Hanaysha, (2025).

Artificial intelligence employee outcomes: Investigating insecurity technostress hospitality industry.

Psychologica Shoss, (2017). insecurity: integrative review agenda future research.

Journal Management, Soomro, Pitafi, (2024). Workplace:

Driving Employee Performance Through Enhanced Knowledge Sharing Engagement.

International Journal Human-Computer Interaction (17), Sowa, Przegalinska, (2019, October).

Digital coworker: human-AI collaboration environment, example virtual assistants management professions.

Collaborative innovation networks conference Digital Transformation Collaboration (pp.179 201).

Springer International Publishing. Xiong, Zeng, (2024). artificial intelligence usage facilitates performance.

Social Behavior Personality: International Journal (10), Sverke, Hellgren, swall, (2002). security: meta-analysis review insecurity consequences.

Journal Occupational Health Psychology Tahir, Javed, Akhtar, Wang, (2024).

Employees friend: artificial intelligence employee outcomes.

Service Industries Journal (11-12),1100 Tams, Thatcher, Craig (2018). trust matters post-adoptive usage: mediating roles internal external self-efficacy.

Journal Strategic Information Systems Tang, Koopman, Cremer, Zhang, Reynders, Chen, (2023). person island:

Unpacking after-work consequences interacting artificial intelligence.

Journal Applied Psychology (11), Tang, Koopman, Cremer, Zhang, Reynders, (2023). regulatory consequences dependence intelligent machines work:

Evidence field experimental studies. Human Resource Management Baykal, Aydin, Yakupo kuner, (2023). mployees' Artificial Intelligence ttitudes

ffect ndividual usiness erformance?. Journal Organisational Studies Innovation Tenakwah, Tenakwah, Amponsah, Eyaa, Boateng, Okhawere, (2022).

Adoption sustainable technologies during crisis: examining employees perception readiness across cultures.

Sustainability Teng, Zhou, Zheng, (2024). Artificial intelligence awareness withdrawal: evaluating chained mediation through negative work-related rumination emotional exhaustion.

International Journal Contemporary Hospitality Management Tong, Fang, (2021).

Janus artificial intelligence feedback: Deployment versus disclosure effects employee performance.

Strategic Management Journal Vaccaro, Almaatouq, Malone, (2024). combinations humans useful: systematic review meta-analysis.

Nature Human Behaviour 8(12), Venkatesh, Morris, (2000). don't directions?

Gender, social influence, their technology acceptance usage behavior. uarterly Verma, Singh, (2022 employees intention artificial intelligence-based hybrid environments.

Transactions Engineering Management Verma, Singh, (2022 Impact artificial intelligence-enabled characteristics perceived substitution crisis innovative behavior employees high-tech firms.

Computers Human Behavior Verma, Singh, Tudoran, Bhattacharyya, (2024).

Elevating employees psychological responses performance through responsible artificial intelligence.

Information Technology People Tang, Jiang, (2023). development digital technology contribute innovation performance China's high-tech industry.

Technovation Wang, (2023). Double-edged sword effect flexible arrangements employee innovation performance: demands resources individual effects perspective.

Sustainability (13), Wang, Zhang, Chen, Zhang, (2022).

Influence artificial intelligence robotics awareness employee creativity hotel industry.

Frontiers Psychology Weng, Yuan, Huang, (2024). Leveraging responsible artificial intelligence enhance salespeople well-being performance.

Service Industries Journal, (9-10), Liang, Wang, (2024). buffering workplace mindfulness: insecurity human-artificial intelligence collaboration impacts employees life-related outcomes.

Journal Business Psychology Tang, Jackson, Gray, (2023). robots increases insecurity maladaptive workplace behaviors:

Multimethod evidence. Journal Applied Psychology Yang, (2022).

Transition Robot Identity Partner Competitor Implications Human Robot Interaction.

International Journal Social Robotics Jiang, (2024). really help? double-edged sword effect assistant employees innovation behavior.

Computers Human Behavior Duffy, Tepper, (2018). Consequences downward envy: model self-esteem threat, abusive supervision, supervisory leader self-improvement.

Academy Management Journal Zang, Shao, (2024). Challenge hindrance: paths usage effects Chinese employee innovative behaviour.

Pacific Business Review

Zhang, (2025). generative affects employee incremental radical creativity: empirical study China.

European Journal Innovation Management Zhang, Liang, (2025 human-like characteristics assistants affect employee creativity: social network perspective.

International Journal Human-Computer Interaction (10), Zhou, Chen, Cheng, (2024).

Bringing employee learning stress research: moderated mediation model.

Technological Forecasting Social Change Opportunity threat? meta-analysis impact human-AI collaboration systems employee effectiveness Yixiao Mingzhuo School Human Resources Guangdong University Finance Technology Guangzhou China) School Business Administration Guangdong University Finance Technology Guangzhou China) School Management Guangdong University Technology Guangzhou China

Abstract

rapid development rtificial ntelligence profoundly changed social structures production models, application organizations attracted significant attention scholars regarding impact employee efficiency. investigate impact underlying mechanisms human-AI collaboration systems employee ffectiveness study conducted meta-analysis independent samples 54,726) derived studies. findings reveal following: uman- collaboration applications, autonomy, anthropomorphism, employee positively influence employee efficiency, representing "opportunities." Conversely, awareness exerts negative effect, perceived "threat" trust insecurity mediating roles relationship between human-AI collaboration ystems employee efficiency, further elucidating pathways "opportunity" "threat" Additionally, mployee categories, industry characteristics, cultural contexts moderate these effects. research concludes human-AI collaboration systems double-edged sword effect enhance employee efficiency through trust reduce insecurity, positive effect outweighing negative. study, within framework Conservation esources heory, clarifies mechanisms boundary conditions impact human-AI collaboration systems employee efficiency, providing guidance organizations effectively leverage value while correctly understanding impact.

Keywords

human-AI ollaboration insecurity, trust, performance, employee innovation