Rising with the machines: A sociotechnical framework for bringing artificial intelligence into the organization

Source: Erin E. Makarius, Debmalya Mukherjee, Joseph D. Fox, Alexa K. Fox; Journal of Business Research 120:262-273 · 2020

Artificial intelligence (AI) is increasingly being adopted by organizations, yet implementation is often carried out without careful consideration of the employees who will be working along with it. If employees do not understand or work with AI, it is unlikely to bring value to an organization. The purpose of this paper is to investigate the ways in which employees and AI can collaborate to build different levels of sociotechnical capital. Accordingly, we develop a model of AI integration based on Socio-Technical Systems (STS) theory that combines AI novelty and scope dimensions. We take an organizational socialization approach to build an understanding of the process of integrating AI into the organization. Our framework underscores the importance of AI socialization as a core process in successfully integrating AI systems and employees. We conclude with a future research agenda that highlights the cognitive, relational, and structural implications of integrating AI and employees.

Introduction. “All too often, AI projects start by trying to implement a particular technical approach, and, not surprisingly, front-line managers and employees don’t find it useful, so there’s no real adoption and no ROI” -Roman Stanek, CEO of GoodData (Satell, 2018).

Approximately 80% of large companies have adopted some form of artificial intelligence (AI) into their core business, an increase of 70% in five years (Ghosh, Daugherty, Wilson, & Burden, 2019). AI is relevant for almost every organizational function and is ranked first on the Society for Industrial and Organizational Psychology’s top 10 workplace trends list (SIOP, 2020). Scholars agree that AI is likely to become an integral part of the future of work (Huang & Rust, 2018; Wilson, Daugherty, & Davenport, 2019). AI technologies are associated with a plethora of benefits that range from greater efficiency, faster and more accurate results, and reduced error rate at the process level, to more effective and improved strategic outcomes at the organization level (Davenport & Kirby, 2015; Davenport, Guha, Grewal, & Bressgott, 2020; Paschen, Pitt, & Kietzmann, 2020). AI refers to “a system’s ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation” (Kaplan & Haenlein, 2019, p. 17). Ideally, such systems should sense, comprehend, act, and learn (Jarrahi, 2018; Kolbjørnsrud, Amico, & Thomas, 2017), mimicking a person applying intelligence (Haenlein, Kaplan, Tan, & Zhang, 2019). Indeed, today’s organizations use AI for highly complex tasks that traditionally require human intelligence, such as selecting applicants for job positions or scheduling logistics (Lawler & Elliot, 1996; von Krogh, 2018), and even for distributing payments to exchange partners in a complex supply chain system through blockchain enabled smart contracts (Murray, Rhymer, & Sirmon, 2020). Similarly, a hospital in Boston successfully utilized machine learning and other AI technologies to create a hyper-local alert system that helped them successfully forecast COVID-19-related clinical demands during the ongoing pandemic crisis (Stevens, Horng, O’Donoghue, Moravick, & Weiss, 2020). Despite the potential for AI to drive organizationally-valued outcomes, however, there is evidence that this often is not the case (Canhoto & Clear, 2020). Many companies invest time, effort, and resources into AI, yet do not experience the anticipated benefits – ultimately deeming AI initiatives a failure (Fountaine, McCarthy, & Saleh, 2019). For instance, a survey of executives by Boston Consulting Group and MIT finds that seven out of ten AI projects generated little impact and that AI implementation plans dropped from 20% in 2019 to 4% in 2020 (The Economist, 2020). Similarly, in a study of senior managers working on 152 AI projects, Deloitte (2017) reports that 47% of respondents find it difficult to integrate AI with existing people, processes, and systems. Thus, it is clear that in order for AI to be successful, employees must embrace, interact with, and integrate their behavior with AI systems (Glikson & Woolley, 2020; Lichtenthaler, 2018). Indeed, scholars indicate that “organizations are entering a landscape characterized by unprecedented collaboration among managers and intelligent machines. There are no maps available yet for navigating through this challenging and unknown terrain” (Kolbjørnsrud et al., 2017, p. 6). Therefore, the interaction between AI and people should be the focal point of AI research in management. This leads to the following broad research question that motivates our current analysis: How do we bring AI into the organization and successfully integrate such systems and employees to create a sustainable competitive advantage? Answering this question is important as organizational scholars indicate that the combination of AI with human ingenuity offers a fertile ground for novel research (Fleming, 2019; Haenlein & Kaplan, 2019; Murray et al., 2020; von Krogh, 2018). The significance of AI and employee integration is clear, especially in practitioner literature and anecdotal evidence (Shrestha, Ben-Menahem, & von Krogh, 2019); yet the investigation of this integration is less developed in the management and organizational academic literature, with few AI related articles published in top management journals that are even “remotely connected” to people and work (Phan, Wright, & Lee, 2017). Moreover, much of the existing research on AI focuses on its technical application (e.g., Barrett, Oborn, Orlikowski, & Yates, 2012). However, AI systems can be active agents in improving strategic decision-making and problem solving (Davenport et al., 2020; Murray et al., 2020), rather than passive receivers of human inputs. Unlike prior technology, AI will have the capability to collaborate, learn from, and adapt to employee interactions.

Related work. AI has been suggested as the ‘Fourth Industrial Revolution’ (Syam & Sharma, 2018), which is characterized by a shift in decision-making from humans to machines. Indeed, the fundamental distinguishing feature of AI is its ability to make decisions and to learn and interact accordingly (Haenlein & Kaplan, 2019; Makridakis, 2017). While differentiating AI from prior technological advancements, Syam and Sharma (2018) opine that the close interaction between the physical, digital, and biological worlds sets AI apart from its technological predecessors (p. 136). While the prior technological advancements focused on altering or replacing routine manual tasks, AI involves cognitive, relational, and structural complexities (Kaplan & Haenlein, 2020; Syam & Sharma, 2018). Thus, the changes that occur with AI integration are uniquely different than those of prior industrial revolutions. Therefore, as we review the literature on AI, we identify how prior research fits into these cognitive, structural, and relational themes, as shown in Table 1. Studies that contribute to the cognitive theme focused more on some of the practical applicability of systems that were built to think or act like humans. Studies with structural themes have contributed to broader discussions of environmental factors involved in the AI adoption process. Last, nascent research in the relational theme briefly summarizes relationships between people and AI, especially along the themes of trust and collaboration.

Mimicking elements of human behavior and higher order thinking provides a goalpost for many that work in development of AI capabilities, like sensemaking work in natural language processing (Weizenbaum, 1966) (NLP) and computer vision (Papert, 1996). The application of novel techniques like machine learning, neural networks, and deep learning is an attractive area for many businesses looking to improve their organizational decision-making ability (Haenlein & Kaplan, 2019; Shrestha et al., 2019; Meinhart, 1966). The combination of man and machine have not eliminated flaws in decision-making such as biases, which appear in models, algorithms (Choudhury, Starr, & Agarwal, 2020), training sets (Cowgill & Tucker, in press), and input incompleteness (Choudhury et al., 2020). Fair, transparent, and explainable AI serve as a step to help mitigate some of these issues (Satell & Sutton, 2019), yet the creation of context-spanning systems that satisfy these criteria are challenging to construct. Nevertheless, the literature shows that newer AI systems are continuously becoming more intelligent and moving from simple task automation activities to having better contextual awareness (Davenport et al., 2020).

Method. Our discussion of the pertinent AI literature demonstrates that AI has shifted from the use of simple tools to highly sophisticated systems (Davenport & Kirby, 2015; Glikson & Woolley, 2020) that can have wide ranging effects on the future workplace and employee behavior. Consequently, we believe that it is important for management scholars to better understand the dynamics of different AI systems and employee relationships. STS theory offers one perspective that can help build understanding of how the integration of AI and employees can lead to a competitive advantage for organizations. STS develops a framework to examine how people within a given system interrelate with its technology to affect joint outcomes while being enveloped in a contextual environment (Emery, 1959). Thus, STS theory seeks to “describe and explain the behavior of organizations and their members while providing critical insights into the relationships among people, technology, and outcomes” (Kull, Ellis, & Narasimhan, 2013, p. 66). As such, STS theory helps build the case for sociotechnical capital, or what we define as the combination of AI technology and people in organizations that leads to a source of competitive advantage for an organization. We extend work by Resnick (2001), who indicates that sociotechnical capital refers to “productive combinations of social relations and information and communication technology” (p. 3). In his conceptualization, Resnick (2001) describes sociotechnical capital as a subset of social capital focused on the technological medium of interactions. He argues that individuals can foster sociotechnical relations through email, social media, and other technological applications, particularly in the community and in building civic virtual engagement. We propose, however, that the sociotechnical capital construct can be expanded to the work domain and modified to reflect the ability of the organization to jointly optimize the use of AI and people. Sociotechnical capital is likely to result from the successful integration of AI technology and employees where both the entities act as a tightly coupled system exhibiting increased responsiveness (Morgan-Thomas, Dessart, & Veloutsou, 2020; Orton & Weick, 1990). Like other forms of capital, sociotechnical capital can be accumulated over time with intentional investment. As such, it can be considered an intangible resource that is valuable, rare, difficult to imitate, and organization-specific that may be utilized to build a sustainable competitive advantage (Barney, 2001). Prior research suggests that capital is valuable only to the extent to which “an organization is able to integrate and utilize” the resource (Raffiee & Byun, 2020, p. 37). As intelligent systems continue to evolve and become more pervasively adopted by organizations for a broader range of tasks, they will change employee roles and the nature of the AI-employee relationship also is likely to become more complex. Thus, the novelty of such systems and the scope of their applications in the adopting organization are two salient dimensions in this regard. As shown in Table 2, we develop a “Model of AI Integration” in the workplace that combines the level of AI novelty (low/moderate/high) and the scope of AI (content changing/ incremental AI versus context changing/radical AI) dimensions. “Content changing” refers to the changes occurring in a narrower task domain, mainly based on the component knowledge of the associated system (Davenport et al., 2020; Dewar & Dutton, 1986; Henderson & Clark, 1990). “Context changing” refers to the changes that may occur in broader task domains (and sometimes across the value chain creating new ecosystems) (Davenport et al., 2020) based on the architectural knowledge of the system (Mukherjee, Lahiri, Ash, & Gaur, 2019). The typology highlights the types of employee-AI relationships and human roles in working with different types of AI. The resultant quadrants entail six types of AI and associated human-AI relationships and human roles. Each of these combinations produces various levels of sociotechnical capital in organizations.

Discussion. Our analysis and the socialization model of AI is aimed at stimulating discussion about how to successfully integrate AI into the organization. Additionally, the ideas we present pave the path for future scholars to meaningfully contribute to this discussion. It is widely accepted that cognition and behavior are central constructs driving organizational goals and outcomes (Phan & Wright, 2018). These two constructs manifest themselves through strategic decision-making, while influencing relational outcomes, and shaping organizational design and structure at different levels. Thus, while AI-centric research in the management literature can take many shapes and forms, consistent with the theme of our paper, we highlight the cognitive, relational, and structural implications of integrating AI and employees (see Table 3 for an overview). First, our framework for AI socialization creates opportunities for research on cognition at the individual (decision-making) and organizational (learning) level, as well as for knowledge management. Second, our framework provides relational questions about integrating AI onto a team, drawing attention to issues related to teamwork, identification, and coordination. Third, we build upon our framework to provide recommendations for structural human resource and organizational issues such as training and job design. Behavioral theory of the organization (Cyert & March 1963) and its related theoretical frameworks of organizational learning (March, 1991) and knowledge management literatures (Grant, 1996) should form the foundation of cognitive scholarship in AI. A key area of cognitive scholarship on AI should is decision-making (Kaplan & Haenlein, 2020; Shrestha et al., 2019). Although some AI systems can make decisions on their own, there are questions as to how decision makers trust the output that they are receiving from an AI program and understanding how the system arrived at that outcome (Gunning, 2017). What happens when a system suggests something that a lifelong employee sees as foolish? Will the system or the employee be given favor?

What weights should be given to such systems as they gain increased control over various processes? How can one implement proper controls in a decision-making process when an AI system encounters an abnormality that requires human interaction? Future scholars should explore how managers manage the shrinking gap between AI and employee capabilities and how AI and employees proactively and jointly alter strategic decision-making. Organizational cognition is another crucial research domain in the current context. It is now widely accepted that organizations do learn (Huber, 1991; March, 1991). However, adoption and subsequent integration of AI in organizations may mean deep and disruptive changes in their learning architectures. While many companies are already using big data inputs obtained from platforms such as social media, the integration of AI systems such as ANNs and deep learning models may imply fundamental shifts in the way organizations learn and absorb new knowledge. Newer AI technologies will compel companies to shift from the focused learning mindset (where learning occurs at specific stages of the value chain or in a particular task domain) to more complex and transformative learning (in which learning would occur across the value chain and may include partner organizations). The positive spillover effect of such learning will be jointly shared by the network/cluster organizations (Mudambi, Mudambi, Mukherjee, & Scalera, 2017).

Conclusion. AI is set to fundamentally transform the future of work. We are currently witnessing a critical shift in the usage of AI systems during the ongoing COVID-19 pandemic crisis, a time that has, in many cases, required companies to increase AI-employee integration to continue safe and effective operations. For instance, in the tourism industry, the usage of Robots for reception and concierge services and AI-enabled chatbots for standard customer communication and questions, has substantially increased (Sigala, 2020), allowing greater safety for employees who can then focus on more complex interactions with customers. AI can be used to support essential jobs, such as upgraded ventilators that nurses and doctors can operate remotely. Robots in grocery stores perform cleaning tasks so that employees can focus on stocking shelves and ensuring that customers have the products they need in a time of crisis (Howard & Borenstein, 2020). The COVID-19 pandemic highlights the need for further AI-employee integration to enhance safety and obtain organizationally and socially valued outcomes worldwide. Our analysis makes two distinct contributions to extant research. First, our proposed typology using the AI novelty and scope dimensions points toward different types of AI systems and AI-employee relationships that can be leveraged in studying the evolving complexity and resulting integration of AI and employee roles in the workplace. This is particularly important because although AI is increasingly being adopted in organizations, managers and employees have negative perceptions related to job substitution, training, and uncertainty (Frey & Osborne, 2017), lack understanding of why and how the AI should be used (Raisch & Krakowski, 2020), and have issues in trusting AI systems (Gunning, 2017).