Conceptualizing a Socio-technical Model for Evaluating AI-driven Technology

Abstract

Socio-technical models have been developed to address challenges inherent to complex systems. The goal of this paper was to conceptualize a comprehensive socio-technical model, integrating technological and measurement-based dimensions of AI technology with socio-technical dimensions. This model addresses current limitations in operationalizing AI-driven technology interventions in complex adaptive environments. Building off of previous models, eight dimensions were identified to account for factors influencing the success of AI-driven technology interventions in complex, adaptive organizations, including hardware and software; content; user interface; people; workflow and communication; organizational policies, procedure, and culture; external rules, regulations, and forces; and measurement and monitoring. This model provides new avenues of thinking regarding socio-technical dimensions for adopting AI-driven technology by organizations and how these can be accounted for in AI-driven technology interventions and research.

Keywords

Socio-Technical Model Artificial Intelligence Evaluation AI-driven Technology

Introduction

With Industry 4.0, digital information technology has transformed how we work and fundamentally changed how we do business and interact with each other. Industry 4.0 has been enabled by simultaneous advances in many fields over the past decade, including artificial intelligence (AI), machine learning, robotics, Internet-of-Things (IoT), autonomous vehicles and self-driving cars, 3D printing, virtual and augmented reality, wearables, additive manufacturing, and others that are blurring traditional boundaries and creating new business models (Vollmer, 2018). These and other technologies have had a disruptive impact on both personal and working lives. With Industry 4.0 being concerned with the ongoing transformation to create a highly automated, interconnected, and data-driven working system, Industry 5.0 directs the focus on the human element by creating a working system that puts people at the center, with machines and technology playing a supporting role (Bednar & Welch, 2020).

It is suggested that Industry 5.0 will be characterized by human intelligence working collaboratively with cognitive computing to produce more value-added goods and services, with humans focusing on tasks that require creativity, problem-solving, and decision-making skills (McEwan, 2013). For instance, organizations embracing the vision for Industry 5.0 set out to have humans and AI interact continuously to manage processes effectively, with digital assistants supporting the monitoring and management of complex systems in dialogue with human managers, using AI to give expert advice to optimize workflows (McDonnell, 2018). These developments will undoubtedly create new challenges for enterprises and service providers, leading to a greater focus on interactions among stakeholders and between stakeholders and intelligent, integrated systems (Bednar & Welch, 2020).

An ongoing challenge to designing, developing, implementing, and evaluating AI-driven technology interventions is to operationalize their application within complex adaptive environments (e.g., environments that consist of distributed settings, are high-stressed, and are fast-paced; Sittig & Singh, 2010).

While models have previously been applied, for instance, Venkatesh’s unified theory of acceptance and use of technology (Venkatesh et al., 2003; Kijsanayotin, Pannarunothai, & Speedie, 2009; Holden & Karsh, 2010; Weger et al., 2022), Reason’s Swiss Cheese Model (Reason, 2000; Paulin et al., 2023) or Norman’s 7-step human–computer interaction model (Norman, 1988; Malhotra et al., 2007; Xu et al., 2021; Tuli et al., 2022). All of these models account for one or more critical facet of technology implementation; their utility to address the full range of factors that should be considered during the design, development, implementation, use, and evaluation of AI-driven technology interventions is somewhat incomplete. For instance, these models were not designed to address the complex relationships between hardware, software, information content, and the user interface nor address measurement and monitoring structure (e.g., AI analytics and methods to continuously collect data, create or review digital reports, or track defined outcomes).

Based on these limitations, this paper aims to conceptualize a comprehensive socio-technical model to integrate technological and measurement-based dimensions of AI-driven technology with other socio-technical dimensions. Particularly with the embarkation of Industry 5.0, it is apparent that a socio-technical perspective is needed for opportunities to be pursued in a way that balances the needs and desires of different stakeholder groups and ensures that the full potential of intelligent technologies is harnessed for the benefit and safety of all.

Socio-Technical Systems

The effectiveness of any purposeful activity is considered a socio-technical phenomenon to bring about a desired outcome that requires technical and social elements to be considered. For instance, humans use tools to be productive, and tools are designed for use. Thus, the socio-technical approach does not pursue two separate (social and technical) strands for evaluation, but one, integrated whole (Mohr & van Amelsvoort, 2016).

The term was first introduced in the 1950s by researchers at the Tavistock Institute of Human Relations in the UK (Trist, 1981). In a socio-technical system, the social and technical dimensions are interdependent and influence each other. The social dimensions include the people, their roles, and the relationships between them, while the technical dimensions include the tools, equipment, and technology used to support their work. For example, a hospital can be viewed as a socio-technical system. The social dimensions include the doctors, nurses, other healthcare professionals, and patients and their families. The technical dimensions include medical equipment, electronic health records, and other technologies used to support healthcare delivery.

Pan and Scarbrough (1998) stated, “the socio-technical perspective thus adopts a holistic approach which highlights the interweaving of social and technical factors in the way people work”. The fit between the social and technical systems builds an organization. Laudon and Laudon (2004) expanded on the previous perspective by stating that: “In a sociotechnical perspective, the performance of a system is optimized when both the technology and the organization mutually adjust to one another until a satisfactory fit is obtained.” Thus, a socio-technical system is a way of thinking about how technology and people interact in complex systems.

While understanding the interdependence of social and technical dimensions, organizations can design more effective, efficient, and resilient systems, yet current models often fail to break down the technology dimension into individual components that would allow for dissecting the use problems or identifying specific solutions. Previous research by Sittig and Singh (2010) has identified that models often fail to consider technology-specific elements such as the interplay of hardware, software, information content (e.g., Big Data and AI-generated decision support), and user interfaces or attempt to treat them separately, which hinders the overall understanding of intelligent technology-related challenges.

For instance, well-constructed, robust user interface content terminology can make all the difference to an operator struggling to quickly and accurately enter a time-sensitive order for a critically ill patient (Rosenbloom et al., 2006). Without a multidimensional understanding of the technological dimensions that drive the AI-application for the order, one may incorrectly conclude that the AI-driven software or the user were the cause for failure when in actuality, the poorly designed content terminology might have been the root cause.

Optimizing the performance of the social and technical dimensions requires a holistic approach that considers how changes to one dimension will impact the other. For example, introducing new AI-driven technology may require changes to how people work, and changes to work processes may require changes to the technology used to support them. However, most models do not account for the unique monitoring processes or governance structures required for designing, developing, implementing, or using AI-driven technologies.

For example, identifying who will make the decision on what, when, and how AI-driven technology interventions will be added; developing a process for monitoring the effect of a new AI-driven technology, such as an AI-generated decision support application on the systems’ response time; building tools to track the AI-generated decision support that is in place; developing an approach for testing AI-generated decision support; defining approaches for identifying rules that interact; developing robust processes for collecting feedback from users and communicating new system fixes, features, and functions; and building tools for monitoring the AI-generated decision support system itself (adopted from Sittig and Singh, 2010).

Socio-Technical Model for AI-driven Technology

To study the design, development, use, implementation, and evaluation of AI-driven technology, we adapted the 8-dimensional model by Sittig and Signh (2010) and conceptualized a socio-technical model that accounts for factors influencing the success of AI-driven technology interventions in complex adaptive organizations. As with other complex adaptive systems, these interacting dimensions need to be viewed in relation to each other and cannot be viewed as a series of interdependent steps. While some dimensions depend more on one another (e.g., hardware, software, content etc.), the social dimensions strongly influence the technical elements. Therefore, the AI-driven technology intervention must be perceived in the context of simultaneous effects across multiple dimensions. For example, evaluations to develop and implement AI-driven applications to centralize stored electronic data reports revealed lower benefits than expected but indicated that complex interdependencies between socio-technical factors at the working-, organizational-, and the national-level were to be expected (e.g., Greenhalgh et al., 2010).

The 8-dimensional model includes (see Figure 1):

Hardware and Software. This dimension of the model focuses solely on the hardware and software required to run applications. For instance, AI-driven applications used to automate tasks or analyze data for improved decision making have particular hardware and software requirements that depend on the complexity of the tasks involved. This dimension only comprises the physical systems and the software required to keep these systems running. One of the critical aspects of this dimension is that users, for the most part, only become aware of the existence of the infrastructure once it fails (Leveson & Turner, 1993; Sittig & Singh, 2010).

Content. This dimension includes everything on the data-information knowledge continuum stored or used as requirements to train the system (i.e., structured and unstructured data, e.g., textual, numerical, and images) captured from device sensors or other sources; e.g., Bernstam, Smith, and Johnson, 2010). The content elements are used to configure specific software requirements. Examples include for instance, controlled terminology selected and the logic required to generate a specific type of automated action. These elements may also describe aspects of certain conditions (e.g., laboratory test results, discharge summaries, or imaging). Other information content may be used to manage administrative aspects. These data may need to be created, entered, read, modified, or deleted by the user and stored. Content elements, such as those which inform AI-driven decision support interventions, must be managed regularly.

Interface. This dimension includes the hardware and software that operationalize the user interface to enable entities to interact with the system and includes aspects of the system that users can touch, see, or hear. The interface is designed according to human-computer interaction and ergonomic guidelines to match the user’s workflow and reduce task complexity. For instance, when a user wants to change the amount for an order fulfillment, the software requires the user to discontinue the old order and enter a new one, but the user interface should hide this complexity (Sittig & Singh, 2010).

People. This dimension represents the humans (e.g., software developers, system configuration, training personnel, and users) involved in all aspects of designing, developing, implementing, and using AI-driven technology. It also includes how the AI-driven application helps users think and make them feel (Pelau et al., 2021). In addition to the users of these systems, this dimension includes the people who design, develop, implement, and evaluate these systems. For instance, these people must have the proper knowledge, skills, and training to develop safe, effective, and easy-to-use AI-driven applications. This is the first aspect of the model that is purely on the social end of the socio-technical spectrum. In most cases, users will be employees of the system. However, with recent advances in customer-centered applications and the development of personal AI-driven systems (e.g., smart home devices), end-users of all ages are increasingly becoming essential users of AI-driven technologies. End-users may not possess the knowledge or skills to manage AI-driven technologies, which has become a concern.

Workflow and Communication. This is the first portion of the model that acknowledges that people often need to work cohesively with others in the working system to accomplish their tasks. This collaboration requires significant two-way communication, horizontally and vertically. The workflow dimension accounts for the steps needed to ensure that customers receive the service or product they need at the time and to the quality they need it. Often, the AI-driven system does not initially match the actual workflow. In this case, either the workflow must be modified to adapt to the AI-driven system, or the AI-driven system must change to match the identified workflows.

Organizational Policies, Procedures, and Culture.

An organization’s internal structures, policies, and procedures affect every other dimension in this model. For instance, top management is responsible for allocating capital budgets to enable the purchase of hardware and software, and internal policies influence how data backups are accomplished. Those responsible for implementing IT policies and procedures may manage the AI-driven technology life-cycle, from procurement, implementation, use, and monitoring, to evaluation. The AI-driven technology intervention must ensure that the software accurately represents and enforces organizational policies and procedures. Likewise, the user workflow for using these AI-driven technologies must be consistent with policies and procedures. Internal rules and regulations are often created in response to external rules, regulations, and forces.

External Rules, Regulations, and Forces. This dimension accounts for the external forces that facilitate or constrain the design, development, implementation, use, and evaluation of AI-driven technologies. For example, according to Statista, the global market for AI in healthcare was valued at approximately US$11 billion in 2021 and is expected to rise to a huge US$188 billion by 2030 (Elliot, Fox, Hon & Vithlani, 2023). Meanwhile, a host of federal regulations are put in place to better regulate the use of AI-driven technologies as well as the need for procedural safeguards to ensure the use of transparent AI systems that explain the reasoning that the AI follows before arriving at a decision, both to benefit users and oversight authorities. In July 2022, the US American Data Privacy and Protection Act (ADPPA) received approval from the House Energy and Commerce Committee. The bill proposes national standards regarding personal data collected by companies and AI decision-making. Further regulation of AI decision-making is likely to see continued focus from the federal government following the publication of a blueprint for an AI Bill of Rights by the White House Office of Science & Technology Policy (Elliot et al., 2023).

Measurement and Monitoring. This dimension posits that the effects of AI-driven technology must be measured and monitored on a regular basis. An adequate AI system measurement and monitoring program must address not only four issues related to AI-driven technology features and functions but also its ethical issues that still need to be addressed. First is the issue of availability – the extent to which features and functions are available and ready for use (i.e., system availability such as response times and percent uptime). A second measurement objective is determining how users use the various features and functions. Third, measuring and monitoring the system’s effectiveness on task performance to ensure that anticipated outcomes are achieved. Finally, in addition to measuring the expected outcomes of AI-driven technology implementation, it is also necessary to identify and document unintended consequences that manifest from using AI-driven systems (e.g., Ash et al., 2004).

Figure 1.

The 8-dimension of the Socio-Technical Model for AI-driven Technology.

To fully achieve the potential of AI, according to Naik and colleagues (2022) four ethical issues must be addressed, measured, and monitored: (1) informed consent to use data, (2) safety and transparency, (3) algorithmic fairness and biases, and (4) data privacy (Gerke et al., 2020). Therefore, in addition to measuring the use and effectiveness of AI-driven at the local level, we must develop methods to measure and monitor these systems and assess the quality and ethicality of their use on a state, regional, and even national level.

Discussion

The most important result of this conclusion is that hierarchical decomposition (i.e., breaking a complex system, process, or system down into its components, studying them, and then integrating the results in an attempt to understand how the complete system functions) is an improper approach in evaluating technology (Rouse, 2008). Evaluating the AI-driven technology intervention, complex interdependencies between various socio-technical dimensions are to be expected. This model should not be used to evaluate how the eight dimensions interact and depend on one another. The model can be used to evaluate how these multiple interacting components with non-linear, emergent, and dynamic behavior appear and impact each other. Hence taking into account that a small change in one aspect of the system leads to small changes in other parts under certain conditions but large changes at other times.

The model also provides a comprehensive framework for the practical implementation of AI system development. For instance, by considering all eight dimensions of hardware and software, content, interface, people, workflow and communication, organizational policies and culture, external rules and regulations, and measurement and monitoring, AI system designers can ensure that the hardware and software requirements are met, the data used for training is managed effectively, the user interface is intuitive and matches user workflows, the system considers the needs and capabilities of users, and the organizational policies and procedures align with the AI system. Additionally, developers ensure that external regulations are considered to ensure compliance and ethical use of AI-driven technology. Further, it is clearly defined who and how regular measurement and monitoring are performed to assess availability, user behavior, task performance, and unintended consequences while also addressing ethical issues such as informed consent, safety, transparency, fairness, biases, and data privacy. Applying human factors expertise within the holistic socio-technical model to AI system design and evaluation, the human factors field can contribute to developing safe, usable, and ethically sound AI systems, ultimately enhancing the overall user experience and societal impact of AI-driven technologies.

In summary, this model provides professionals and researchers with new avenues of thinking about key socio-technical dimensions and how these can be accounted for in future AI-driven technology implementations and research.

Conclusion

Current approaches lack the integration of socio-technical dimensions into models to guide AI-driven technology interventions for the design, development, implementation, use, and evaluation. This lack of integration allows for inadequate approaches to address the complex relationships between the technical dimensions (e.g., hardware, software, content, interface) and social dimensions (e.g., people). A lack of the dimension measurement and monitoring of AI-driven technologies is also apparent when evaluating current models. This paper conceptualizes a comprehensive socio-technical model to integrate these technological and measurement-based dimensions of AI-driven technology, resulting in eight total dimensions for improving the success of adopting AI-driven technology interventions in complex adaptive organizations. This model allows for a novel way of thinking about and evaluating AI-driven technologies.

References

Ash

J. S.

Berg

Coiera

(2004). Some unintended consequences of information technology in health care: the nature of patient care information system-related errors. Journal of the American Medical Informatics Association, 11(2), 104-112.

Bednar

P. M.

Welch

(2020). Socio-technical perspectives on smart working: Creating meaningful and sustainable systems. Information Systems Frontiers, 22(2), 281-298.

Bernstam

E. V.

Smith

J. W.

Johnson

T. R.

(2010). What is biomedical informatics? Journal of Biomedical Informatics, 43(1), 104-110.

Elliott

Fox

Hon

(2023). AI in 2023: Key Trends and Developments. Dentons.

Gerke

Minssen

Cohen

(2020). Ethical and legal challenges of artificial intelligence-driven healthcare. Artificial Intelligence in Healthcare, 295-336. Academic Press.

Greenhalgh

Stramer

Bratan

Byrne

Russell

Potts

. (2010). Adoption and non-adoption of a shared electronic summary record in England: a mixed-method case study. BMJ

Holden

R. J.

Karsh

B. T.

(2010). The technology acceptance model: its past and its future in health care. Journal of Biomedical Informatics, 43(1), 159–172.

Kijsanayotin

Pannarunothai

Speedie

S. M.

(2009). Factors influencing health information technology adoption in Thailand's community health centers: applying the UTAUT model. International Journal of Medical Information, 78(6), 404–416.

Laudon

K. C.

Laudon

J. P.

(2004). Management Information Systems: Managing The Digital Firm. New Jersey: Pearson Prentice Hall.

10.

Leveson

N. G.

Turner

C. S.

(1993). An investigation of the Therac-25 accidents. Computer, 26(7), 18-41.

11.

Malhotra

Jordan

Shortliffe

Patel

V. L.

(2007). Workflow modeling in critical care: piecing together your own puzzle. Journal of Biomedical Informatics, 40(2), 81-92.

12.

McDonnell

(2018). Saline Lectronics Fast Forward to Industry 5.0 – When Luigi meets Alexa. i40today.com.

13.

McEwan

A. M.

(2016). Smart working: Creating the next wave. London: Routledge.

14.

Mohr

B. J.

van Amelsvoort

(2016). Co-creating humane and innovative organizations: Evolutions in the practice of socio-technical system design. Portland ME: Global STS-D Network Press.

15.

Naik

Hameed

B. M. Z.

Shetty

D. K.

Swain

Shah

Paul

Aggarwal

Ibrahim

Patil

Smriti

Shetty

Rai

B. P.

Chlosta

Somani

B. K.

(2022). Legal and Ethical Consideration in Artificial Intelligence in Healthcare: Who Takes Responsibility? Frontiers in Surgery, 9, 862322.

16.

Norman

(1988). The Psychology of Everyday Things. New York: Basic Books.

17.

Pan

S. L.

Scarbrough

(1998). “A Socio-Technical View of Knowledge-Sharing at Buckman Laboratories”, Journal of Knowledge Management, 2(1), 55-66.

18.

Paulin

Joud

Hennebert

Moëllic

P. A.

Franco-Rondisson

Jayles

(2023). HistoTrust: tracing AI behavior with secure hardware and blockchain technology. Annals of Telecommunications, 1-15.

19.

Pelau

Dabija

D. C.

Ene

(2021). What makes an AI device human-like? The role of interaction quality, empathy and perceived psychological anthropomorphic characteristics in the acceptance of artificial intelligence in the service industry. Computers in Human Behavior, 122, 106855.

20.

Reason

(2000). Human error: models and management. BMJ, 320(7237), 768–770.

21.

Rosenbloom

S. T.

Miller

R. A.

Johnson

K. B.

Elkin

P. L.

Brown

S. H.

(2006). Interface terminologies: facilitating direct entry of clinical data into electronic health record systems. Journal of the American Medical Informatics Association, 13(3), 277-288.

22.

Rouse

W. B.

(2008). Health care as a complex adaptive system: implications for design and management. Bridge-Washington-National Academy of Engineering, 38(1), 17.

23.

Sittig

D. F.

Singh

(2010). A new sociotechnical model for studying health information technology in complex adaptive healthcare systems. Quality & Safety in Health Care, 19(3), 68–74.

24.

Trist

E. L.

(1981). The evolution of socio-technical systems (Vol. 2). Toronto: Ontario Quality of Working Life Centre.

25.

Tuli

Bhardwaj

Kumar

Gautam

(2022). An extensive overview on human-computer interaction (HCI) application. Journal on Software Engineering, 17(1).

26.

Venkatesh

Morris

M. G.

Davis

F. D.

Davis

G. B.

(2003). User Acceptance of Information Technology: Toward a Unified View. MIS Quarterly, 27, 425–478.

27.

Vollmer

(2018). What is Industry 5.0. Sunnyvale, CA, LinkedIn.

28.

Weger

Matsuyama

Zimmerman

Mesmer

Van Bossuyt

Semmens

Eaton

(2022). Insight into User Acceptance and Adoption of Autonomous Systems in Mission Critical Environments. International Journal of Human-Computer Interaction.

29.

Dainoff

M. J.

Gao

(2023). Transitioning to human interaction with AI systems: New challenges and opportunities for HCI professionals to enable human-centered AI. International Journal of Human–Computer Interaction, 39(3), 494-518.