Co-working with AI is a Double-sword in Technostress? An Integrative Review of Human-AI Collaboration from a Holistic Process of Technostress

. An accumulating body of research has demonstrated that human-AI (Artificial Intelligence) collaboration is an indistinguishable future of work. The study of employee-perspective, affect-related responses, and job-related consequences from the adoption and use of AI is gaining attraction in the field of business and management. Some research has examined the technostress creators or negative responses related to human-AI collaboration, however, a holistic process of technostress that considers both positive and negative impacts has not been adopted. This study integrates the findings to examine the relationship between technostress and human-AI collaboration, based on the technostress creators (challenge technostressors and hindrance technostressors), positive and negative psychological responses, and outcomes in work-related well-being and intention to use AI. Future research opportunities on technostress generated by human-AI collaboration are discussed alongside implications to theory and management practices.


Introduction
Artificial Intelligence (AI), a symptom of technological advancements and breakthroughs, is different from previous information technology applications and products.When the machine mimics human thinking and actions through data analysis, it can be called AI as the word "intelligence" represents that the machine can learn and solve problems [1].
Although AI has been studied since the 1950s, the power of AI was yet to be discovered and adopted by organizations to boost business performance in the past twenty years, with the development of machine learning, deep learning, and neural networks [2].AI has been initially adopted to accelerate the production line in the manufacturing field and subsequently to perform as a coworker in supporting human decision-making in various business fields, changing the future of work [1].For example, predictive analytics that can help markets generate dynamic and competitive pricing strategies has been a critical marketing effort in the service and retail industry [3].Chat-box and virtual assistants that use advanced natural language processing and voice recognition have also been applied as communications tools to enhance work productivity [4].Therefore, the question of whether AI will replace humans in the workplace has been intensely discussed within academia.Yet an agreement on AI job replacement has not been reached.On the one hand, the pessimistic view presented that AI would cause a large proportion of people to lose their jobs due to job replacement theory [5].On the other hand, some researchers argued that AI, freeing people from repetitive and low-efficiency jobs, would foster employee innovation and generate more new job opportunities [6].
Considering the characteristics of tasks, much literature suggests the development of human-AI collaboration (e.g., [1] [4] [7]).According to AI job replacement theory, tasks in organizations can be divided into four types: mechanical tasks (such as administering daily routines and tracking attendance), thinking tasks (such as analyzing customer preferences and scheduling logistics), intuitive tasks (such as thinking creatively and adjusting effectively to novel situations) and empathetic tasks (such as empathizing with customers and advising therapies to patients) [5].When it comes to most knowledge work in the business context, it involves both thinking and empathetic tasks, such as stock pitching requires both data analysis and the part of selling to others, and recruiting involves resume screening and communication with the candidates [8].While organizations have already adopted AI to perform not only mechanical tasks but also thinking tasks, humans can provide an intuitive and emotional approach that AI can not [7].Therefore, the human-AI collaboration that allows humans and AI each to leverage their advantages in the business processes will be the future for knowledge workers.
Despite the increasing research attention on human-AI collaboration, several literature gaps remain.First, previous literature has identified technostress as one of the negative determinants in human-AI collaboration from the individual level, but they have not explored technostress creators in a synthesized way [9].Second, though previous research has suggested considering a holistic process of technostress, research has yet to examine both positive and negative sides of technostress related to co-working with AI in the workplace [10].Third, most literature has investigated different models of human-AI collaboration and its relation to job outcomes through team and organizational levels, but research has not examined how technostress influences an individual's job outcomes [11].This review aims to fill the gap by synthesizing the findings of previous studies and answering the questions: Which, why, and how do technostress creators influence human-AI collaboration?
The rest of this paper is organized as follows.The first section presents a review of technostress research and clarifies a holistic process of technostress as the framework of this paper.The next section presents a review of previous studies in three parts: technostressors, psychological response, and outcome.The following section outlines the theoretical and practical implications of human-AI collaboration.Conclusions and future research directions are discussed in the final section.

Conception of technostress
Technostress is first proposed by Craig Brod as "a modern disease of adaptation caused by an inability to cope with new computer technologies in a healthy manner" [12].With the increased use of ICT (Information and Communications Technology) in the professional work environment, a majority of studies have investigated the phenomenon of technostress (e.g., [13] [14] [15]).Tarafda et al. conducted one of the most recognized studies to explain the antecedents of technostress in the organization environment by adopting socio-technical theory and role theory [16].They identified five technostress creators: techno-overload, techno-invasion, techno-complexity, techno-insecurity, and techno-uncertainty [16].Later, Ayyagari et al. argued that the previous conceptualization blurred the boundary between technology characteristics and stress [13].Building upon the person-environment fit model, their work showed that the rise of stress was due to two types of misfits between the characteristics of the person and the environment: the supply-value misfit and demand-ability misfit [13].The study constructed technostress as five work stressors, which are work overload, role ambiguity, invasion of privacy, work-home conflict, and job insecurity, and found three technology features as the antecedents of technostress: usability, intrusiveness, and dynamism [13].
Aligning with the perspective of the "dark side" of technology, many researchers studied the negative impact of technostress, e.g., decreased job productivity and hindered job satisfaction [14] [16].However, Califf et al. argued from an organizational stress perspective and posited that a holistic technostress process was constituted by both the techno-eustress subprocess and the technodistress subprocess [10].According to the holistic technostress process, an individual cognitively evaluates whether a stressor is positive or negative and then views the stressor as either a challenge stressor (beneficial to workplace tasks) or a hindrance stressor (harmful to workrelated tasks), the process of which is called appraisal process [10].After that, the individual responds to a challenge stressor or a hindrance stressor with a positive or negative psychological reaction, respectively.Finally, the individual performs or changes his or her behavior in the workplace [10].
This study uses the model from Califf et al. as a reviewing framework to synthesize the findings on the relationship between co-working with AI in the workplace and technostress.

Technostress generated by human-AI collaboration
This section presents a review of the relationship between human-AI collaboration and technostress.It initially discusses research on technostress creators.Subsequently, studies that elaborate on both positive and negative psychological responses are presented.Final part reviews studies of work-related well-being and behavior outcomes.

Construct of stressors
When an individual passes through the appraisal process, the individual classifies environmental stimuli into either hindrance technostressors or challenge technostressors [10].This section reviews literature related to hindrance technostressors (unreliability, complexity, uncertainty, insecurity, overload) and challenge technostressors (usefulness, tech support), respectively.

Hindrance technostressors
Several studies identified the challenges while working with AI, which were mostly aligned with the hindrance technostessors (e.g., [9] [17]).For example, through a qualitative analysis of 24 frontline service employees who worked with robots, Meyer et al. confirmed the issues of work overload, job insecurity, unreliability, and complexity [18].Arslan et al. also summarized key challenges of human workers-AI team interaction: fear of losing jobs, complexity, overload, and unreliability of the potential of AI-powered team members [19].
The below review presents research that focuses on two of five hindrance technostressors, namely, insecurity and complexity.

Insecurity
Among those five hindrance technostressors, job insecurity has been the most studied work stress when investigating the negative influence of AI adoption (e.g., [20] [21]).
Job insecurity describes the degree of an individual's perceived uncertainty about job stability and continuity [20].When organizations adopt AI, humans experience a mix of hope and fear that contributes to the feeling of uncertainty [22].On the one hand, people are worried that AI will replace their current jobs in the future, leading to either layoffs or a decline in the employee treatment [20].On the other hand, AI handles tedious and repetitive tasks, calling for talent upskills [6].
Despite the above research mainly focused on employees' perspectives, one survey investigated whether AI would reduce the workforce in the next five years from the managers' view; the result showed that 47% of the managers agreed, representing an even mix of perceptions [22].

Complexity
The complex feature of AI is one of the barriers for employees to understand and perform their job duties [10].Thus, employees may experience feelings of role ambiguity [13].In addition, Bedué and Fritzsche suggested that the underlying rules of AI seemed unapparent to the users, thus the results produced by AIgenerated can not be interpreted, which was considered the obstacle to developing trust in AI [23].

Challenge technostressors
Two challenge technostressors have been investigated through previous studies, which are usefulness and tech support.

Usefulness
Usefulness is the most recognized challenge technostress in the previous studies [17] [24].For example, Malik et al. found the perception of overall enhancement in job performance when using AI in the workplace [9].Similarly, Sowa et al. found that a majority of people agreed that cooperation with AI will enhance their job productivity [4].In the research that focuses on a specific function, the perception level of AI's ability to change human resources management was high because AI could help save the time spent on mundane tasks [25].

Tech support
Tech support refers to the degree of organizational support for employees in using AI [10].Cao et al identified that tech support had an insignificant influence on behavioral intention [24].Providing employees with the knowledge to work with AI can also be viewed as tech support, which was one of the determinants to increase trust in AI [23].

Psychological response
An individual generates a positive or negative psychological response as the reaction to a challenge stressor or a hindrance stressor [10].Since a holistic process consists of two subprocesses, both negative and positive psychological responses are reviewed.

Fear
Numerous studies have investigated whether people perceive the fear of losing jobs and discovered contradictory findings.For example, a piece of qualitative research on 32 working professionals in multinational firms with an average work experience of 7.6 years confirmed job insecurity was the second-highest work stressor through the thematic analysis [9].However, a recent study analyzed the perceptions of 821 human resource employees and managers in large 500 companies, over half of whom demonstrated "not worried about being unemployed" [25].
To further dive into the different findings of fear of job replacement, researchers have found individual differences, job type differences and task type differences.For example, young generations experienced a little higher job insecurity than senior workers [4].The differences also existed in job types: office and operation workers, business analysts, and experts and specialists are more likely to experience job insecurity caused by AI [22].Additionally, the relationship between fear of job replacement and type of job tasks is also investigated.Recently, Vorobeva et al. performed a series of studies on employees in the service industry and discovered that people experienced a higher level of fear of job replacement when engaging in thinking tasks than feeling tasks [28].Furthermore, when AI was salient, thinking tasks had a similarly negative effect on employees' fear of replacement [28].
Therefore, based on the above literature on negative psychological responses to job insecurity, age differences, job type differences and task type differences have been found when deploying AI in the workplace.

Concern
Some studies suggested that when employees used AI, they could face two concerns: personal concerns and technology concerns.Cao et al. identified two personal concerns: personal development concern and well-being concern [24].Personal development concern is referred to the use of AI that may hinder personal skill development and career prospect; well-being concern is considered another issue due to some possible consequences like depression and cynicism if using AI [24].In addition to personal concerns, technology concerns were found related to privacy concern, and technology-reliance concern [29] [4].

Threat
Cao et al. hypothesized that AI threat was positively influenced by perceived susceptibility to bad decisionmaking due to the use of AI, and the perceived severity of negative consequences [24].The result supported the hypothesis and also suggested a negative relationship between AI threat and an individual's attitude towards AI.

Positive psychological response
Trust is the only psychological response that has been discussed when employees working with AI as trust can be viewed from both cognitive and emotional perspectives [30].Some studies have successfully identified antecedents of trust in AI through technology features and personal features.In a review paper that covers approximately 200 publications in the year between 1999 and 2019, the important antecedents of cognitive trust were AI's tangibility, transparency, reliability and immediacy and one specific determinant of emotional trust was AI's anthropomorphism [30].

Outcome
Psychological responses lead to the actions, which is the outcome.The outcome can be positive or negative regardless of its psychological responses [10].For example, positive psychological responses can result in a negative outcome.
This section further classified negative and positive outcomes into two streams: well-being and behavior.Firstly, building upon dimensions of well-being, this section reviews the impact of AI adoption on employees' well-being, including emotional well-being (e.g., anxiety, stress, burnout) and professional well-being (e.g., job engagement) [31].In addition, the reviews of behavior outcomes mainly focus on the acceptance and resistance of AI.

Negative impact on well-being.
Negative impacts caused by AI technostressors have been found on affective well-being (e.g., job burnout, emotional exhaustion), and professional well-being (e.g., job engagement, change readiness).Firstly, a study on call center employees indicated that perceived stress from working with AI decreased affective well-being in the service industry [27].Additionally, Kong et al. presented that the higher level of AI awareness contributed to job burnout based on the definition of AI awareness as employees' perception of job insecurity [21].In this study, burnout was measured in three dimensions: emotional exhaustion, depersonalization and reduced personal accomplishment.Similarly, Wu et al. discovered that job insecurity had a significant effect on emotional exhaustion [20].In addition to affective well-being, research on professional well-being focused on work engagement and change readiness.Braganza et al. investigated AI adoption as a moderating factor that decreased the positive relationship between psychological contract and job engagement [32].The same study also argued that the negative effects of AI on job engagement may be caused by job insecurity [32].Moreover, a study on HR managers confirmed that AI anxiety negatively influenced change readiness and a high-performance work system moderated the effect [26].

Negative impact on behavior
Several negative psychological responses have been confirmed the negative influence on AI adoption.For example, both well-being concern and personal development concern negatively affected an individual's intention to use AI [24].Similarly, data privacy concern was another negative factor in adopting AI [17].Choi discovered the moderating effect of privacy concern between role clarity and employees' willingness to AI adoption, which also illustrated the negative effect of concern on behavior intention [29].

Positive outcome 3.3.2.1 Positive impact on well-being
Literature has found that working with AI generates a positive impact on professional well-being, including change readiness, and meaningful work.For example, Smids et al. proposed that when AI takes over the boring and repetitious tasks in the workplace, employees can have an increased feeling of pursuing meaningful work [33].Additionally, challenge technostressors like beliefs about AI have also been found to positively influence change readiness for AI adoption [26].

Positive impact on behavior
Challenge technostressors and positive psychological responses have both been studied in the relationship with AI adoption.Perceived usefulness and tech support as challenge technostressors were found to have a positive effect on an individual's intention to use AI [24] [17].Additionally, trust, as one of the positive psychological responses related to the use of AI positively, could influence employees' willingness to accept AI technology [29] [17].

Implications
This paper provides several literature and practical implications.In terms of literature contributions, this paper provides a synthesized review base on Califf et al.'s model of both positive and negative sides of technostress when co-working with AI [10].As a result, this paper provides clear pathways to understand an individual's technostress process when co-working with AI in organizations.Second, this paper also delivers practical implications for facilitating human-AI collaboration strategy and the related training and support within organizations.A series of dark and bright side variables and outcomes can be combined to minimize the negative influence associated with employees' technostress when they collaborate with AI.For example, tech support could be a challenge technostressor that generates a positive psychological response and a high-performance work system could moderate the negative effect on change readiness.

Future research
There are three possible future research directions.
Firstly, this paper suggests that future studies should shed light on balanced research on both the bright and dark sides of technostress related to human-AI collaboration.For example, in terms of psychological response, much literature investigated negative emotions like fear, threat and anxiety.More research on the eustress subprocess will facilitate a clearer understanding of the effects of technostress related to using AI in the workplace on employees' positive organizational behavior, such as employee innovation, and organizational commitment [34].
Another interesting domain for future research is the exploration of technostress related to AI-aid decisionmaking in a broad of industries and job functions.The development of AI has stepped into supporting decisionmaking in various business domains, including marketing, healthcare, human resources, etc. [35].Future studies can focus on the effect of technostress in fostering human-AI collaboration in decision-making.For example, a conceptual model constructed by Solberg et al. explored the relationships among trust, perceived risk, and reliance on AI Decision Aids recently [36].
Lastly, a majority of literature focused on the perception of AI adoption (e.g., [4] [32]) and few studies examined the effect of human-AI collaboration through actual co-working with AI.Thus, field research in more industries and functions is needed to investigate the effects of human-AI collaboration in the feeling economy, especially in that have a high level of empathetic skills, such as business and financial operation, and first-line managers [1].

Conclusions
Despite increasing interest in human-AI collaboration, there is little research on how working with AI drives technostress and its impact on psychological well-being and organizational behavior.The goal of the current study was to present a holistic review of technostress in relation to human-AI collaboration.In closing, this paper argues that it is important to understand the both risks and potential of human-AI collaboration in today's organizations.