1 Introduction

The Internet of Things (IoT) aims for the widespread diffusion of connected devices (sensors and actuators) that can interact with each other and cooperate to achieve a digital representation of the physical world with little human involvement (Atzori et al. 2010). IoT technologies are among the transformable technologies entering humans’ physical world. They are becoming increasingly ubiquitous and intrusive, quickly penetrating every aspect of human lives: IoT sensors nestle themselves within us (i.e., body area networks and intrabody networks (Celik et al. 2021)), between us (i.e., social-internet-of-things (Atzori et al. 2012)), in our environment to realize smart environments (e.g., smart buildings and industrial IoT (Butun et al. 2020)). They also empower Intelligent Personal Assistants (IPAs) so that they may control and decide on human users’ behalf, and continuously learn to act like us (i.e., supporting the realization of the human Digital Twins — DT). As a result, IoT technologies bring extraordinary promises and possibilities. However, they may also introduce significant risks and have considerable implications, both at the individual and societal level. Along with the increasing developments in this field, there is also a growing need to understand the nature, effect, and impact of IoT technologies. This should not just happen from a technological perspective but also concerning existing socio-technical systems and IoT technologies’ capacity to support key human values, and collective ability to solve humanity’s problems and thrive (Sawyer and Jarrahi 2014).

In this article, we understand human-centric IoT as an approach that can meet the needs of communities served by IoT technology and that helps to find out what technically works best for humans and society first and foremost. To date, IoT-related human-centered design has predominantly had a technical approach, concentrating for example on adding computing to everyday objects, managing connectivity issues for things with dissimilar constraints, technology that allows things being able to communicate to each other, and understanding how data can affect people’s sense-making of things (Koreshoff et al. 2013a). Traditionally, IoT systems have lacked appropriate approaches for managing human-centric usability engineering, with many unresolved research questions touching upon multiple disciplines (Nunes et al. 2015). For instance, issues such as how people could interact with automated and Internet-connected technical systems, how they support empowerment, and how human agency is configured remain largely unaddressed (Wagner 2019).

While human aspects are to a certain extent taken into account, the technology’s capabilities are limited in profoundly reflecting human values, needs and behaviors, ensuring that human judgment and discretion naturally flow into the system requirements (Wagner 2019). IoT implies a delegation of actions and decisions by human beings to objects, where the system does not ask but “adapts” to humans and the environment. Humans are therefore not represented as autonomous agents, with minds of their own, or as a particular kind of psychology with different susceptibilities (Pereira et al. 2013). Instead, current IoT network infrastructures consist of distributed devices communicating with each other, sometimes on behalf of humans, inside a connected physical environment. The pervasive network’s logic aims to fit the human to the perceived or observed intelligent environment, not the other way around (Ystgaard and De Moor 2021).

To delineate the potential implications of these developments, one should also consider market and policy forces. Next-Generation networks have been developed to accelerate technology innovation and uptake, primarily to drive commercial and technological advantages for nations, corporations, and society (Huigen and Cave 2008). At the same time, however, the dominance and self-regulation of digital corporates have led to a context in which global actors are dictating the terms for the digital infrastructure, technology, and economy, towards a pervasive surveillance-society, propelled forward by platform and surveillance capitalism (Cammaerts and Mansell 2020). Other critical scholars have problematized that the logic that commands the technology development is based on economic and technology incentives to achieve an even more pervasive and intelligent IoT network (Zuboff 2019). IoT devices often dictate how users interact with their personal data. Once the IoT network is intelligent, it can learn, predict, and modify human behavior, using algorithms trained to manipulate for commercial gain. The danger presents itself as citizens’ facing a complete loss of privacy and autonomy essential for participation in a democracy, along with unjust algorithmic discrimination exacerbating inequalities and operating for others’ ends (Castells 2009).

In this regard, legal frameworks such as the General Data Regulation Protection (GDPR) (The European Parliament and the Council of the European Union 2016) and very recent initiatives towards a European Data Act (European Commission 2022) aim to ensure privacy protection, fairness, and compliance with ethical imperatives in the sharing and use of different types of data. However, they have important limitations, such as not being available worldwide. For instance, in the case of GDPR, it represents risks and tensions between conflicting interests (Padden and Ajehag Pettersson 2021), and such frameworks have also been criticized for unclear responsibilities in more complex scenarios (Wong 2021) and for offering limited protection in other cases (Royakkers et al. 2018).

At the same time, researchers in the fields of Human-Computer Interaction (HCI), privacy legislation, digital ethics, and humanized computing are calling for new technological infrastructures that support free and democratic societies and that have the technical ability to empower people (Hasselbalch 2021; Kazim and Koshiyama 2021). There is a need for insights and increased efforts to build a genuinely human-centric IoT, where restoring trust in the infrastructure is essential. Furthermore, underlying systemic issues in the existing infrastructure need to be solved, with various interventions targeted across all layers of the Internet technology stack (Bego and Brynskov 2020).

To effectively shape a human-centric IoT that establishes human potential in a societal and ethical way, we argue in this paper that the scientific community, engineers, and designers must first develop a shared understanding of what human-centredness conceptually translates to from a socio-technical perspective. Then, a shared framework is necessary to adjust the appropriate principles, requirements, and technical implementations in IoT. Scientifically, the understanding of the socio-technical embedding of IoT systems is still limited to date — for instance, the impacts they may have, and the effects society has on them, and their operation, among others. Yet, the importance of identifying and investigating socio-technical gaps ahead of the technology design is strongly emphasized in the literature (Mumford and Weir 1979). In addition, there is a need to re-discuss how IoT systems may help realize higher-level policy goals and challenges and avoid undesired consequences and implications.

We argue that a literature review is needed to understand the current scope of how human-centered design in IoT is addressed (conceptual definitions, methods, techniques, proposed solutions) and to create a roadmap for future research directions aligned with humanity-centered policy goals.

The following research questions guide our work:

  • Which disciplines and types of expertise are involved in research on human-centric approaches in IoT?

  • What are the principal underlying theories and conceptual definitions for human-centered design, and how are they interpreted in the context of the Internet of Things and sensor network technologies?

  • In which ways and to which extent is human-centredness operationalized and embedded in the design, method, and implementation of Internet of Things technologies?

  • What gaps should be addressed and prioritized in future research?

This research represents one of the first attempts to bring structure and clarity in the emerging work towards user/human-centredness in an IoT context. To this end, it adopts a socio-technical and critical perspective. There are to date no other systematic reviews that have investigated human-centric design theory, principles, and requirements of IoT. The most similar reviews have focused on specific user-centric issues in Internet of Things, such as privacy (Kounoudes and Kapitsaki 2020), human-in-the-loop (Nunes et al. 2015), and security requirements (Sicari et al. 2015; Rao and Deebak 2022), without widening the design perspective towards incorporating human and social values, human involvement, and impact on human and society (Atzori et al. 2010; Koreshoff et al. 2013b). In this work, we therefore undertook a systematic literature review of the human-centric Internet of Things to capture its breadth across multi-disciplinary databases and academic literature. Further, this review aims to bridge the divide between human-centric/humanity-centered theory and practice in the study of IoT networking.

The remainder of this paper is organized as follows: Sect. 2 introduces key elements of human and humanity-centred design and points to key socio-technical design aspects to be considered in the context of human-centric IoT. Section 3 presents the methodological approach that was adopted to conduct the systematic review, building upon the introduced elements from Sect. 2. Next, the main findings are presented in Sect. 4 and further discussed in Sect. 5. Finally, implications of this work for future research and overall conclusions are shared in Sect. 6.

2 Background

To achieve a model of Next-Generation Internet of Things that is truly human- and humanity-centered, one must consider a complex interplay between technology, humans, and the environmental context (Soegaard and Dam 2012). The assumed decline in trust towards digital technologies also in relation to Internet of Things technologies, due to algorithmic discrimination, inequality, and exploitative design practices represents an additional complexity in this respect. It has been argued that this decline continues to grow, despite initiatives worldwide in increased regulation on user and data privacy (Voas et al. 2018). The IoT paradigm can greatly benefit from boosting the involvement and diversity of human beings in the development of optimized services powered by technology, and hence minimize reluctance (Kim et al. 2017).

Therefore, design approaches that focus on the human and user concern primarily, such as humanity/human-centred approaches, are highly relevant and extremely important in the context of the forthcoming intelligent network environments (Stephanidis et al. 2019). To achieve this, this papers’ research design focus was on the design structure, practice, and link to the technical contribution. Hence, this section briefly introduces key principles underlying human- and humanity-centered design. Further, it zooms in on a set of aspects to consider in the IoT context, which informed the framework underlying the conducted systematic literature review (see Sect. 3).

2.1 Human- and humanity-centred design

Traditional human-centered approaches in HCI and IoT focus on the system’s usability, where the design problem is informed by the various participants’ practices, as opposed to addressing a specific technical problem to an information-processing solution. The term “human-centered” rather than “user-centered” refers to the technology impact on a broader group of participants instead of those typically considered users, so that designers can develop a more humanized view of their responsibilities to the people they design for (International Standard 2010). The humanized approach applies a holistic framework that understands what is valued by a system’s stakeholders and that is supporting them in delivering this value (Cockton 2004). The underlying principles for a humanity-centered experience can be summarized as (Wright and McCarthy 2010; Graeff 2018; International Standard 2010):

  • Protection of human potential and humanity first, rather than exploiting vulnerabilities.

  • Valuing the person behind “the user”.

  • Viewing the designer and human user as co-producers of experience (resulting in active human involvement).

  • Viewing the human user as a part of a network of social (i.e., the self-interacting with connected environments) through which experience is co-constructed (resulting in the appropriate allocation of function between between humans and the technological environment).

  • Viewing the person as an acting, self-directed agent with the ability to imagine possibilities and make creative choices.

These principles align with the European declaration of digital rights, which put people and their rights at the center of digital transformation (The European Parliament 2022). The application of humanity-centered design principles tries to accommodate the needs of communities served by technology to find what technically works best for humans and society (Shneiderman 2020). It has important implications in terms of who should be involved in the design process, how, when, and on which grounds.

2.2 Human-centric approaches in IoT

Moving to a broader networking and IoT context, human-centered technology approaches are traditionally grounded in the interests and needs of the individual user, aiming to make products easy to use and understandable (Koreshoff et al. 2013a) and enabling smooth user experiences. This understanding is rooted in HCI developments in the area of IoT during the period 2012-2014, where users increasingly were included and considered (Chin et al. 2019). One of the related concepts originating from the Telecommunications domain that has been introduced in the literature in this respect is Quality of Experience (QoE) (Raake and Egger 2014). The existing network infrastructure aims for optimal technological performance and good Quality of Service (service perspective), as well as a high Quality of Experience (QoE) and perceived quality (user perspective). QoE metrics should ensure that the system meets users’ expectations in any given context or situation (user perspective) and thus allows them to use services and applications easily and according to their needs and preferences (Suryanegara et al. 2019). However, the role of users in this view is still predominantly passive (for a detailed analysis, see (Wechsung and De Moor 2014)).

From the technical point of view, a human-centric IoT design is intended as the process that considers the requirements from the users’ point of view in terms of needed functionalities, interaction modalities, and which sensors to adopt to monitor the user status (Koreshoff et al. 2013a). In this respect, the interaction between IoT technologies and humans is, at the bare minimum, designed to be trustworthy, secure, and deliver a consistent network performance (Kounelis et al. 2014). The introduction of the use of machine learning and big data analytics in IoT, during 2015-2017, spearheaded personalization as a way to adapt to specific and dynamic user needs, often referred to as a human-centric mechanism (Konstan and Riedl 2012; Chin et al. 2019). Personalized IoT services build the user’s profile and monitor the varying context to adapt the system accordingly (Perera et al. 2013). In other cases, still from the machine learning and big data point-of-view, human-centric IoT refers to a connected network of sensors, objects, and machines with more “human-like” capacities and behaviors. This human-centric view is achieved by applying mathematical and algorithmic modeling to explain, replicate, or replace human action (Chin et al. 2019).

However, given the recent technological developments in the IoT domain and as illustrated in Sect. 1, one can argue that the original interpretations of human-centered design in an IoT context have important limitations and have become outdated. More worryingly, there are a range of human, societal and ethical implications that need to be systematically and profoundly considered due to the ubiquity and enhanced intelligent characteristics of the IoT domain (Bannon 2011). Based on similar developments, human-centered approaches in machine learning have begun to rethink how human-centric goals, context, and practises can make machine learning technologies work better for humans and society primarily (Ramos et al. 2010).

As such, future developments in HCI and intelligent environments identify the need to cater for meaningful human control, human safety, and ethics, in support of human health and well-being, learning and creativity, and social organization and democracy (Stephanidis et al. 2019). When approached from a socio-technical perspective, human-centered research in IoT should incorporate human involvement via methods used during the system technology design process (Breve et al. 2021; Ghiani et al. 2017). Individuals or a group of human stakeholders must have a voice through participation in order to balance requirements between various stakeholders’ intention (situated rationality), the dynamic qualities of human activities, an understanding of the contextual environment, and the technical approach of rule-based, codified procedures managed by technical measures and performance indicators (Sawyer and Jarrahi 2014; Gasson 2003; Shin and Park 2017). Achieving this successfully is exemplified in the study presented by Karni et al. (2022), where they build a partnership with the human users and communities that are directly impacted by the technology. In this case, the design targets patient empowerment, based on the identification of clinically meaningful targets for modification instead of a pure technology-driven selection of measurement end points. Therefore, a key question — and part of our focus — is to what extent these aspects and considerations are already embedded in existing practices, and whether there are human-centric technical contributions in IoT that follow a common methodological framework.

2.3 Socio-technical design considerations in IoT

The complexity of building networked systems at the global scale that works and can deliver value to humans, an updated Internet, without amplifying inequities and jeopardizing human rights, introduces a range of design considerations (Giaccardi and Redström 2020). With the propagation of Cyber-Physical-Systems and IoT into every aspect of humans’ everyday life and societal systems, the significance of socio-technical design dimensions in IoT increases (Streitz et al. 2019): design aspects, such as theoretical approaches, expertise, method for involvement, and technical translation, impact the development of a humanity-centered IoT. We argue in this work that they need to be acknowledged and better understood in order to identify blind spots and to reorient current practices and processes in such a way that key issues can be addressed. As a result, the above design aspects are relevant to consider in the analyses presented in this paper.

2.3.1 Design theories and approaches

Prior research on the role of human-centered approaches in IoT argued that there is a gap between human-centered design frameworks and technology design frameworks in IoT (Koreshoff et al. 2013a) and that a more profound focus on human-centered concerns is needed. To achieve this, the definition of the human role in IoT needs to be clear and incorporate human involvement in the design. To clarify the role humans play in IoT, a shared understanding of human-centredness can contribute to increased trust and collaboration between engineers, human users, communities, private industry, and governmental actors (Kounelis et al. 2014). However, to achieve a shared understanding, bridging the world of telecommunication, informatics, electronics, and social science requires additional effort (Atzori et al. 2010).

Previous attempts to establish a human-centered theoretical approach include Koreshoff et al. (2013a)’s design framework. Koreshoff et al. (2013a) presented a tool for an HCI audience to think through the design of IoT technologies that consider things as “the type of information it would produce” and how people would make sense of it. Koreshoff et al. (2013a)’s design tool considers co-design and “build with, not for” methodologies as a way to achieve human-centric technology solutions. However, this design framework is still predominately technical. It does not address deeper human-centered issues, such as social justice, human empowerment, and equality for all, where priorities and needs address the protection of user rights, also on a societal and community level (Koreshoff et al. 2013a). Accordingly, mapping the use of design frameworks and tools that account for socio-technical and human requirements in the research of human-centric IoT can help unravel more profound human-centered practices and challenges (Hochheiser and Lazar 2007).

As an illustration, human-centered, socio-technical design approaches in medical IoT is particularly important when developing Internet of Things (IoT) systems. Human-centric medical IoT offers opportunities to build safe environments for patients and caregivers to provide them with human autonomy and well-being (Tiersen et al. 2021). However, these systems also represent threats with a potentially significant impact on disadvantaged communities. For instance, ethical and social risks are substantial barriers to more widespread adoption of intelligent assistive technologies (IATs) serving people with dementia. Concerns about preserving autonomy are the greatest in the literature, along with issues surrounding beneficence, justice, independence, and privacy (Ienca et al. 2018).

2.3.2 Team composition, design mindset, and practices

The decisions that underpin IoT development have primarily been driven by experts with technical interest and competencies, also those relating to the human-centered perspective (Cherry et al. 2017; Sovacool and Del Rio 2020). When the expert-handled network technology and services no longer simply deliver connectivity to devices, but actively shape humans’ daily experiences, also in their physical environments, a partnership with research communities that deal with human psychology, humanities, and social science is crucial (Calvo and Peters 2014). Furthermore, technology is never neutral, and the built-in intelligence is neither (Miller 2021). Therefore, the scientists, engineers, and designers’ design mindset and associate practices are highly relevant when addressing the challenge of creating a human-centric IoT that responds to humans’ fundamental needs with the aim for empowerment and inclusion.

The design mindset here refers to distinguishing between an expert mindset and a participatory mindset. When an expert mindset is applied, humans are referred to as users, nodes, or subjects, with no or little direct involvement in decision-making (Stead et al. 2019). Such a mindset is traditionally more associated with technology-oriented disciplines. However, with a participatory mindset, the designer and users operate as co-producers of the design, resulting in active human participation (Schneider et al. 2018).

Furthermore, interdisciplinary fields and research traditions, such as Human-Computer Interaction, science and technology studies, digital ethics, and digital humanities, have a long history of examining the development, use, and impact of technology on humans, society, and the environment, in their social, cultural, and historical context (Fuller 2009). Perspectives from various academic disciplines are better equipped to apply holistic theoretical frameworks, methodologies, factors, and techniques. As a result, diverse perspectives allow for creating a ubiquitously connected reality where human well-being, positive social impact, and sustainability thrive (Calvo and Peters 2014).

2.3.3 Policy goals and underlying value systems

What the envisioned Next-Generation IoT will look like will depend on various factors, e.g., what is promoted by policymakers and industry, as well as designers’ value systems, aims, principles, and technical tools (Cockton 2004). In this respect, from a policy point of view, different countries and regions have suggested distinct and not necessarily compatible value systems and governance approaches (Lee 2021).

While the European Union has explicitly presented a vision, the Next-Generation-Internet initiative, where “Human-centredness is key for a human and sustainable development of IoT” (Brynskov et al. 2019), the United States of America (USA) have historically put forward policy rationales with an emphasis on minimal regulation in technological innovation and development. The recent US policy focus, however, introduces regulatory measures at the State level that protect consumers and citizens from harms associated with undermining public values (Jørgensen and Desai 2017). China has strategically adopted a top-down industry policy to IoT, with sustained financial investments, that influence the international technical standards of next-generation network technology (Lee 2021). For example, the Supreme Court of the People’s Republic of China has become a relevant player in enforcing constraints and influencing the level of privacy in IoT, as in the example of new legislation restricting the use of face recognition technology (Qin 2021). Given the growing interest in strategically and politically exerting the national state’s influence, one of the major challenges is to reach agreements on norms, governance, and operational goals, between the different state systems (Choucri 2012). As a result, the geographical dimension of human-centric approaches in IoT is also relevant to consider in this work.

2.3.4 Intended design outcomes

Naturally, policy perspectives strongly relate to the underlying intended outcomes and visions. Being human-centered also means being aware of the intended and the potentially unintended outcomes and the associated consequences of technology design. There is an increasing realization that design goals that are focused on profit, performance, and efficiency determine satisfactory results for engineers, designers, and employers. However, these goals can have harmful outcomes for human users, communities, or society as a whole (Cammaerts and Mansell 2020).

For a free, democratic society to function, the human user needs protection from harmful outcomes, which in the world of ubiquitous and intelligent sensor networks translates into an unwavering commitment to the privacy protection of human users (Zuboff 2019; Cheryl et al. 2021). Most often, this is technically translated into the protection of users’ privacy and their data (Kounoudes and Kapitsaki 2020). However, the most effective safeguarding mechanism would be to integrate human-rights protection into the internet infrastructure, where user agency is achieved by technical design (Wagner 2019). More recently, however, the focus on human-centric IoT technology is increasingly geared towards and understood as maximizing human well-being and positive social impact (Calvo and Peters 2014). To account for both positive and negative outcomes, IoT technology should be designed to serve and protect public goals and values at a holistic level. To address the design logic holistically, there are trade-offs that need to be considered; a) between human control and automation and b) between privacy and intelligence (Streitz 2021). In this context, human-empowering IoT design outcomes are increasingly targeted and focus on non-instrumental goals such as protecting human privacy (Mähler 2019; Kounoudes and Kapitsaki 2020), agency (Wagner 2019; Cruickshank and Trivedi 2017a), empowerment (Royakkers et al. 2018; Streitz 2019), user control (Stephanidis et al. 2019; Nunes et al. 2015), but also trustworthiness (Hesselman et al. 2020; Kounelis et al. 2014) and civic aspirations (Graeff 2018; Winter 2015) related to the Next Generation IoT (Brynskov et al. 2019).

To achieve a genuinely human-centric IoT, the EU has prioritized three research and innovation areas: (1) privacy and safety (i.e., data protection), (2) humans and citizens, and (3) security (including cyber-security). In this respect, the policy challenges of ensuring privacy and security (including the protection of personal data); opposing disinformation online; guaranteeing access and freedom of choice for humans and citizens; respecting fundamental human rights and empowerment, personalized and enhanced support to increase well-being, and enforcing ethics and sustainability by design are explicitly highlighted by the EU’s Next-Generation-Internet initiative (Brynskov et al. 2019).

2.3.5 Technical mechanisms for human-centric IoT

Combining the practical aspects of engineering with the human and social concerns to influence the design (e.g., intended outcomes as exemplified above) requires a translation and operationalization into technical requirements and mechanisms that work in practice and can be evaluated and validated. However, issues such as reliable IoT systems are not simply technical features that are objective or neutral. Instead, they are inherently socio-technical artifacts, produced in and through the actions of people and systems in the course of the working day (Oudshoorn and Pinch 2003; Shin and Park 2017) and implicitly or explicitly based on their assumptions and underlying value system. This raises the question of how exactly one can achieve human-centered IoT technically.

The technical translation of human-centered machine learning and artificial intelligence, include three solution areas: 1) Privacy and data ownership; 2) Transparency and accountability; 3) Fairness in AI-driven decision-making processes (Lepri et al. 2021). Similarly, human-centered technical solutions developed so far in the IoT domain are about making transparent, with or without automated decision support, the right next step for human users (Royakkers et al. 2018). However, the degree of involvement and control exercised by actual human users, who are involved and thereby represented in the design, may vary significantly. The goal of finding the correct technical translation combined with the suitable degree of human involvement, meets several challenges when entering a territory of system design according to human value frames, means, and ends (Bannon 2011). Mechanisms that allow for safeguarding humans’ own individual objectives keep humans actively involved (Wagner 2019). The system includes human input in the process, method, technique, or solution (Adamo 2019). In this regard, interfaces that let humans observe and interact with the IoT system are used to provide transparency, understandability and accountability features to help humans interact with persuasive IoT technology (Turunen et al. 2015; Stephanidis et al. 2019). These include mechanisms that maintain “veillance” about the connected world, or “quantified selves”, and that help mediate between humans and machines (Van Kranenburg and Bassi 2012). Recent examples, such as augmented reality applications in IoT have the potential to deliver richer, more immersive and timely interventions that lead to effective support for human users (Scargill et al. 2022). AR interfaces use AR visualization for instance to contextualize data disclosure and to improve users’ perceptions of threats (Bermejo et al. 2021).

In this context, the term human-in-the-loop has gained momentum as a mechanism to determine the allocation given to machines and humans. Human-in-the-loop mechanisms provide persons, society, or other stakeholders with the opportunity to control both of one’s actions and, through them, of events in the external world (Stephanidis et al. 2019). Here, human permission is required to have the system carry out an action (Wagner 2019; Adamo 2019). However, it can be argued that other (existing) mechanisms are not designed to enable human self-control or serve deeper human-centric purposes but instead aim to find technical solutions that target the underlying communication infrastructure (Hesselman et al. 2020). These solutions are designed to defer to humans passively, for safety, correction, adjustment, input, or simply be switched off. Well-known examples are the essential building blocks, such as blockchain technologies, along with other cryptography techniques, that put events into public view, boosting transparency, reducing red tape, and the likelihood for errors. Most importantly, such mechanisms provide accountability by ensuring that the responsibilities of actions are immutable and identifiable (Kounelis et al. 2014). Other examples, technically integrate uncertainty, as opposed to hard-coded measurements, to allow for multiple and debatable scenarios and outcomes (Russell 2019).

To summarize, the shift towards human-centric IoT comes with grand visions and opportunities, as well as a set of considerations and challenges to address. The work presented in this paper and underlying search explained in detail in Sect. 3, builds upon the above considerations and concepts to improve our understanding of how human-centredness is currently interpreted and embedded in IoT, and the intended outcomes it is targeting or associated with and what mechanisms are being proposed to do so.

3 Methodology

To answer the research questions introduced in Sect. 1, we carried out a systematic literature review to synthesize peer-reviewed conference and journal publications on user/human-centric Internet of Things (IoT). We chose this methodology to identify a complete census of relevant literature and investigate how various approaches tie the human element closer to the design and function of Next-Generation Internet of Things. To conduct the systematic literature review, we followed the methodological framework proposed by Levy and Ellis (2006), a mixed-method systematic framework that is commonly applied in the field of computer and information science. The Levy and Ellis (2006) framework consist of three stages:

  • Stage 1: Planning the input; involved identifying the appropriate body of literature to analyze via a rigorous and iterative process of defining the key terms and search string.

  • Stage 2: Processing the input; conducting the review, included screening, article selection, review of relevant studies, extraction, and analyses of data.

  • Stage 3: Analyses, reporting and disseminating; involved reporting and disseminating the output, where this article’s development took place. In this section, we describe each of these stages.

3.1 Stage 1: planning the input

To explore the various interpretations of human-centredness in Internet of Things, and building upon introduced concepts and considerations introduced in Sects. 1 and 2, the studies referred to in this review were found through combinations of the search terms; “Internet of Things”, “IoT”, “next-generation internet”, and “sensor technology” combined with expressions of “human-centric”. Other variations of “human-centric”, which can be intertwined in the same meaning, are often expressed as user-centric, people-centric, value-centric, humane, or human-aware/user-aware.

The keyword search for the human-centric Internet of Things was framed using four topics from the related literature. The four topics were chosen to capture the relevant methodological frameworks and technical contributions that targeted human-empowering design goals and mechanisms. The final keywords were defined and further delineated based on an iterative process. Applying the four-part division in the search string overcame focusing on keywords specific to niche technologies. It captured the breadth of how human-centric is defined, conceptualized, and operationalized in network technologies. At the same time, the iterative process ahead of defining the final keywords ensured that disciplinary differences in the use of terms and concepts was accounted for and that more niche or novel terms considered relevant for the search were included (i.e., “citizen-centric” and “civic”). The keyword selections are explained in Table 1. Keywords included in the search string are presented in Fig. 1 (see appendix for full search syntax in Table 9).

Table 1 Definitions of search terms
Full size table
Fig. 1
figure 1

Search query

Full size image

3.2 Stage 2: processing the input

The scientific literature supporting this article was found in the digital libraries; Web of Science, Scopus, IEEE, and ACM Digital Library. The review expanded the search beyond the realm of information science to capture the large distribution of quality literature in a number of multi-disciplinary databases. To guarantee the quality of the inputs, we only considered peer-reviewed journal articles and conference proceedings published between 2010 and 2020. Furthermore, the search was originally limited to between 2010 to 2020 due to finding few articles focused on human-centered IoT earlier than 2010. Internet-of-things became more popular as a technology concept after 2010 (Kounoudes and Kapitsaki 2020). As part of the review process, the search was updated for the years 2021 and 2022, to include the most recent literature.

The different steps and phases in the selection process are visualized in Fig. 2. To ensure a complete census of relevant literature, we first conducted a keyword search, followed by backward searching, and finished with forward searching. The database searches were undertaken on September 9th and October 6th, 2020, while the backward/forward search was conducted on January 20th, 2021. The literature search measured the studies’ keywords, title, and the abstract categories. The total number of articles obtained following database searching and removing duplicates was 389. The screening was completed by reading the title, the abstract, and the keywords of each article. The article pool was reduced to 181 after a second duplicate removal (October 2020) and downloaded into Rayyan, a web-tool designed for the screening and selection processing of meta-analysis/review studies. Two reviewers screened the article pool for relevance based on the selection criteria. A third reviewer completed the screening for articles with selection disagreement. 122 articles were accepted for further screening of the full-text. One author applied the selection criteria to every paper, with another author verifying the decision of inclusion or exclusion.

Fig. 2
figure 2

Flow chart of the different steps in the selection process

Full size image

After reading the full version of every article, a significant number of articles were excluded because they did not directly consider user/humans’ role in-depth or were demo papers with incomplete contributions. Instead, these sources dealt mainly with improving efficiency or performance-related aspects of a specific technical solution without explicitly discussing a link to use/users/user outcomes or only mentioning “human-centredness” as a buzzword.

The final selection from database searches resulted in a list of 63 papers. In addition, a backward and forward search process complemented the article pool (January 2020), searching for articles that either used the selected articles as a reference or were referenced by them (Levy and Ellis 2006). From the backward and forward search, we obtained 35 additional articles, and these were screened with the selected criteria, ending up with an addition of 5 articles. A total of 68 studies were included in the qualitative synthesis.

An updated database search was completed on July 13th, 2022. Still, the search strategy and scope was considered relevant, and remained unchanged. After running the updated search for the years 2021 and 2022, a total number of 125 articles was obtained. When duplicates were removed, a total of 88 articles were selected for screening based on pre-existing selection guidelines (inclusion and exclusion criteria), where one author screened first title, keywords and abstract, then two authors screened full-text versions, resulting in an addition of 16 new articles. These articles were integrated into the quantitative analyses, ending up with a total of 84 articles in the updated dataset.

3.3 Stage 3: analyses, reporting and disseminating

This process aimed to gain an overview of the existing knowledge of human-centric approaches in the context of Internet of Things and sensor network technology. Each article was coded in Excel, using an analytical framework summarised in the themes: “general characteristics”, “theory and definitions”, “human involvement”, “user outcomes and mechanisms”, and “future research gaps”. In addition, we used the statistical software program SPSS to investigate potential dependencies between key variables (e.g., discipline, approach, expertise).

4 Results

4.1 General characteristics

A summary of the reviewed approaches used in the current literature can be found in the Tables 10, 11, 12 and 13. These tables contain information about the origin of the study, the team discipline and perspective, level of involvement, the domain with examples of use, intended outcome, and a general description of the solution provided. As outlined in Table 2, we categorized the articles according to their general characteristics such as year, geographical origin, type of publication, academic discipline, and team expertise. Figure 3 shows the frequency of articles per year and application domain. The majority of studies, 74.9 %, were published in the last 6 years (since 2017), which illustrates how research towards the human-centric side of IoT has increased in level of attention in recent years. Nearly a third, 28.6 %, of the included articles were coded as addressing a generic IoT domain, 27.4 % of the included articles were coded as addressing the application domain Smart home, closely followed by 26.2 % addressing the Smart city domain.

Table 2 General characteristics
Full size table

Geographical origin shows where the research team behind the article is located. 69.0 % of all included studies originated from Europe, while 31.0 % originated from the rest of the world. When only considering included papers from 2016 (i.e., the year that GDPR came into force) on-wards, 70.3 % originated from Europe. Around 60 % of the publications were sourced from conference proceedings, where the most common venues were from IEEE and ACM. A smaller proportion originated from domain-specific venues such as SECITC (Security for Information Technology and Communications, HAS (Human Aspects of Information Security, Privacy and Trust) and ICOST (Smart Homes and Health Telematics). Among the journals, most originated from technology-oriented publications; Ambient Intelligence and Humanized Computing, IEEE Communications, and Pervasive and mobile computing. Fewer originated from journals in other fields, some examples; Digital policy, Regulation and Government, and The Design journal.

The subject matter that the authors specialized in was coded as Academic discipline. The discipline was inferred based on the authors’ affiliation. We looked at the academic teams’ expertise in terms of being a research unit from one academic discipline or a research team with a variety of experts and thus a more multi-disciplinary composition. Of the studies reviewed, 64.3 % of the literature was produced by single-discipline teams. A statistical analysis using the Pearson Chi-square test showed that single-discipline teams in our sample are more likely to stem from technology-oriented fields (\({\chi }^2\)(1)= 17.850, \(p<.001\)). More specifically, they primarily consist of researchers with an expertise in engineering, computer science, and informatics. Correspondingly, 35.7 % of the articles come from teams with academic multi-disciplinary knowledge. On the whole, three-quarters of the selected literature stemmed from technology fields, while 14.7 % originated from authors from social science and 8.8 % from humanities. The articles classified as social science included sociology, psychology, and media/communication, while those classified as humanities were from art, law, and philosophy. In terms of mindset (as defined in Sect. 2), a participatory mindset was — as expected — over-represented among contributions stemming from humanities and social science and less common in papers arising from technology-oriented fields, while the opposite holds for an expert mindset (\({\chi }^2\)(1)= 7.067, \(p<.05\)).

Fig. 3
figure 3

Overview of included articles by year and application domain

Full size image

4.2 Theoretical foundations and conceptual definitions

Table 3 Human-centered perspectives in IoT: Theoretical and conceptual definitions
Full size table

We extracted and analyzed the human-centric theoretical concepts and conceptual definitions from the included studies. The theoretical concepts were found by looking for an explicitly stated system of ideas that explained how the human-centered design worked and which theoretical assumptions it was based on. We also extracted the explicitly stated definition conceptualized for human-centric in an IoT context (see Table 3).

The theoretical concepts naturally fit into overarching perspectives; a human-centric computing, a human-centred or user-centred design, a system thinking design, and a human–machine interaction perspective. The human-centric computing perspective represents 51.2 % of the literature. This perspective treats the design primarily as an information and technical design problem, where computing resources are enhanced with technical capabilities extracted from human resources, represented as sensors, devices, data points, or algorithms (Conti and Passarella 2018; Nasir et al. 2012; Valenzuela-Valdes et al. 2017). Then, 35.7 % of the article pool comprises theoretical concepts from human or user-centered design traditions. Here, the design problem is primarily treated as a socio-technical, human, or user relation, ensuring that boundaries between the user, human and social requirements, and technical requirements are balanced. A smaller group of articles, 6.0 %, treat systemic or materialist theories as foundational design principles instead of human and social. Finally, 7.1 % of the articles are in the realm of the human–machine interaction technologies where humans are an integral part of the system/tool/technology that facilitate mediation for/with/or on behalf of humans, as a service, control, or feedback mechanism/tool (Kong et al. 2019; Dustdar and Scekic 2018; Sen et al. 2019; Adamo 2019).

Within the human-centric computing perspective, most of the conceptual definitions of human-centredness focus on describing components or technical resources of IoT using human, social or ethical labelling; examples include Social-internet-of-things, Trusted IoT, Internet-of-people, user-aware IoT or social-aware IoT. A large group of articles defined user-centric as enabling users with tools to protect their data and security. To illustrate, Nawaz et al. (2016) say; “Incorporating user-centricity in ubiquitous IoT systems, is defined as security (integrity and confidentiality of the recorded image data), and privacy protection (altering the image to make it less recognizable)”.

The human/user-centered design perspective conceptually defines human-centredness in IoT as a socio-technical or user-friendly system (Chibaudel et al. 2018; Shin and Park 2017; Worthy et al. 2016; Feth et al. 2017; Kor et al. 2016). A set of articles included here defines human-centredness using more civic and holistic concepts; Ethical and human-centred (Winter 2015; Grundy 2020); value-driven, human-centric collaborative system (Dustdar and Scekic 2018). Here, human-centric IoT technologies support the cooperation between humans and smart devices for the public good. For example, Dustdar and Scekic (2018) state that the fundamental principle for human-centered design in IoT is an architecture of values that supports humans in performing their cognitive, creative, collective, and social activities.

4.3 Outcomes and mechanisms

Next, we captured the intended design outcomes as indicated in the included articles and mapped them to the related policy challenges explicitly specified by the EU, as addressed in Sect. 2.

Table 4 Mapping of policy goals to human-centric design outcomes
Full size table

In Table 4, the intended design outcomes were ranked according to the number of mentions and grouped according to related policy challenges. The first challenge is the most commonly addressed, with 36 % of the literature looking to guarantee security and privacy (i.e., data protection), describing the intended human-centric design outcomes in terms of security, privacy, and user control. The second group of authors, 20 % of the literature, addressed the challenge of developing personalized, mediated technology to enable ease of use and quality of life/experience. Third, 15 % of the literature addresses the challenge of “guaranteeing universal access”, primarily for the intended design outcomes of transparency and cooperation. Finally, 14 % directed towards incorporating ethics and sustainability principles into the system, with outcomes such as civic participation and human agency, as well as 14 % respecting fundamental rights aiming to help achieve human empowerment. All the papers analyzed, had addressed one or more of the outcomes outlined in Table 4.

Table 5 Major technical contributions of technical proposals of human-centered solutions (note that some works have been associated to more than one contribution)
Full size table

Table 5 lists the key technical contributions of most of the proposed human-centered IoT solutions and the corresponding major intended outcomes. The table also highlights the approaches that are most often adopted. Very frequently, the focus is a new Design framework that is intended as a set of principles and procedures to be followed in the design phase. Very often, the human-in-the-loop approach is adopted which considers the human as an active player in the IoT framework. The social IoT model is also a major approach in this context, which gives the objects the capabilities to understand and exploit the humans’ social interactions. Also, the participatory model is advocated as an effective methodology to encourage the users to actively participate in the generation and sharing of data in the relevant platforms. Some works are more concrete and propose novel architectures by adopting a human/citizen/user-centric perspective rather than the commonly used service-centric one.

Privacy, security, control of the shared data, and trust are often targeted in this context. The privacy by design approach is the one that is exploited very frequently, which results in adding security controls and relevant rules to almost any architectural layer. Several deployments of the proposed architectures are presented in the literature, especially with reference to the smart cities and smart housing scenarios. However, there are only a few times where there is a real assessment of whether the user-centric requirements have been attained; indeed, the performance analyses often focuses on technical indicators far from the user perceived experience, such as system memory occupancy and percentage of satisfied service requests.

The user interface is one of the platforms’ components that is commonly the focus of technical contributions. Most of the time, the objective is to instill trust in the user and be empathetic to augment the relevant involvement in the deployed IoT applications. These design objectives may however contrast with each other, as analyzed in Gracanin et al. (2019), where empathy, privacy and ethics are considered all together in the design of the user-IoT-interaction modalities; for instance, what is a privacy-preserving action may not be appropriate from the emphatic point of view. The user interface should also stimulate the user towards collaborating with the other entities involved in the IoT applications; towards this objective appropriate incentives to the users are introduced, which need to be personalized to avoid poor behavior and preserve privacy (Dustdar and Scekic 2018). In the last few years, a complete personal assistant has been proposed as a crucial part of IoT platforms, which is intended to support the user to exploit the full potentials of the IoT platform and make interaction easier (see e.g., (Santos et al. 2018)).

The user monitoring is another important architectural component. The objective is to learn about user habits and create a user profile to personalize the system functionalities. To this end, different sources of information are used (video cameras, localization sensors, collection of data from the users personal devices, and others) and Machine Learning (ML)-based tools together with edge and cloud computing resources.

4.4 Human/user involvement

Finally, we have also considered the study design/contribution type and whether, how, and in what stages users were involved in research under a human-centric IoT umbrella. As observed from the literature and pointed out in Sect. 2, human-centric design can mean that humans or users are either actively or passively involved in the IoT technology design. The nature and degree of human involvement can vary significantly in this respect. A summary of the nature and degree of human involvement within the reviewed approaches can be seen in Table 6.

Table 6 Human-user involvement
Full size table

First of all, we considered the study stage, i.e., whether a paper presented a conceptual work, case study, method, empirical study, or if it had been implemented. We also considered the methodological approach towards involvement, referring to whether the design took a holistic or a more traditional approach, as introduced in Sect. 2. Here, we found that 34.5 % of human-centric approaches were in the conceptual stage of product/system design. 27.4 % were empirical studies, 9.5 % methods analyses, 8.3 % case studies, while 20.2 % were solutions that were being implemented (including prototypes that were tested or evaluated). 58.3 % of the studies reviewed had a traditional methodological approach towards user involvement, while 41.7 % had taken a more holistic approach. We coded traditional when the design focused mainly on a specific IoT technology/solution, such as in the case of Wickramasinghe and Reinhardt (2019) who investigate users’ interacting with a particular privacy-preserving solution dedicated to smart home environments to tailor the experience to their needs, rather than any wider human-human or human-service interaction. The code holistic was assigned when the methodological approach considered the whole IoT technology system as interconnected and evolving with factors in broader human dimensions. Follow-up analyses indicated that holistic approaches are over-represented among contributions stemming from the social sciences and humanities, as opposed to those deriving from technology-oriented fields (\({\chi }^2\)(1)= 9.035, \(p<.01\)). The latter is more likely to be based on a traditional approach.

Finally, Table 6 also refers to the degree of user involvement. High was assigned when a collaborative design approach brought additional stakeholders into the system development process - such as researchers, designers, users, or potential users working together. Medium involvement was assigned when end-users had a direct influence in terms of specific user input or feedback, but indirect methods were applied during most of the design process. Low was assigned when the design only indirectly involves users. For example, passive data collection or observation is used to infer user needs, behaviors, or preferences to adjust the IoT technology to the individual user.

As can be observed in Table 6, 35.7 % of the total article pool was classified as high, 46.4 % as medium, and 19.4 % as low. For example, Feth et al. (2017) involved users with high participatory involvement throughout the design process. Early involvement happened via a usability walk-through. They created personas for all user groups, including bystanders. Then, scenario use-cases were created to understand system context and goals for the users. They also brought in users to evaluate the performance of tasks. An example of medium involvement is given by Giannetsos et al. (2011) where quantitative models describe their users’ individual and social behavior in the form of mathematical models or algorithmic descriptions. These behavioral models are used in devices to replicate the same behaviors of their users. Via their mobile device, humans/users provide input and data. The goal is to develop policies that complement technology designs and individual participant decisions. The users directly involved are the primary owners of IoT devices and solutions. Finally, in one example of low involvement, Dhelim et al. (2018), use sensors to map human emotions, behaviors, social context, physical context, personality, and thoughts passively to offer personalized service, without any direct involvement from users.

Further analyses confirmed (see Sect. 2) that the type of user involvement is also associated with the disciplinary background of the involved team: high user involvement is significantly over-represented and thus more common in the included articles stemming from the social sciences and humanities as opposed to those stemming from technology-oriented disciplines (\({\chi }^2\)(2)= 7.839, \(p<.05\)).

Table 7 ISO Human-centred design phase
Full size table
Table 8 Empirical studies: HCD phases and methods
Full size table

The ISO 9241 requirements for human-centred design outlines five phases of design activities, as shown in Table 7 (International Standard 2010). The 24 articles that report on conducting empirical studies are mapped to these phases and the associated methods in Table 8. While most empirical studies described user involvement across the core design phases (A2-A5), only one of the papers provided an overview of an entire human-centered design process. Furthermore, the majority of studies applied qualitative methods, where the primary users were involved in understanding and specifying technology requirements. A few empirical studies considered secondary user requirements as well. The domain the empirical studies applied to was mostly in smart homes.

5 Discussion

5.1 Current status

In response to the policy focus and corresponding call for more democratic and humanity-centric (empowering) network technology design, the human-centric terminology and goals have found their way into scientific research development in the field of IoT.

5.1.1 Human-centric focus in IoT

Our study indicates that especially after 2015, the number of articles addressing human-centric IoT increased considerably (see Fig. 3). The majority of these originate from Europe-based research teams. Theoretical or conceptual foundations are explicitly stated by 82.1 % of the articles. Table 3 illustrates that the underlying theoretical perspectives remain somewhat fragmented and correspond to various understandings and human-centered perspectives. Furthermore, they are situated at different levels, while an overall and commonly referred to human-centric framework is lacking in the examined papers. As a result, existing research efforts under the heading of human-centric IoT appear to be fragmented rather than systematically addressing a shared vision. This lack of clear common goals and a consensual research agenda based on a shared understanding of what should be the underlying principles, may strongly reduce the potential impact of existing “human-centric” efforts.

5.1.2 Weak link between human-centred theory and practise

This recent rise of the notion of “human-centric” in the context of IoT technologies has informed approaches that conceptually define what human-centred means depending on who and what they are designing for: 1) the human or the social/ecological system, 2) the commercial, legal, and technological capabilities. Ideally, these two approaches should be intertwined, thus calling for inherently inter-and multidisciplinary approaches. However, the link between them (and the associated communities) appears to be at least partly missing or tends to be unclear when IoT solutions are implemented and evaluated in practice. Our analysis confirmed the predominance of technology-oriented fields in human-centric IoT research. While slightly more than 3 out of 10 of the analyzed articles stem from an interdisciplinary team, the majority is single discipline-based, with technology-oriented fields being over-represented in the latter category. The weak link between the above approaches and the lack of a common understanding can also be seen in the light of the under-representation of traditionally more human- and society-oriented fields (i.e., social science, humanities). In turn, this under-representation also has consequences for the adopted practices and the methodological approaches in human-centric IoT research that have been utilized.

5.1.3 Human concepts as technical formalisations

In this respect, the dominant notion of “human-centredness” primarily appears to have been developed from a “technology” approach, which aims to create emotional, social, or ethical network systems to act as decision advisors via a range of user-IoT-interaction modalities. The underlying assumption here is that it is possible to create a merged human user-IoT system with interfaces that fully act in the interests of the human and society. However, while these interests and intended outcomes are explicitly referred to in the majority of the analyzed papers, the question of how they are understood, transformed into technical requirements and mechanisms, and to which extent these mechanisms contribute to safeguarding these interests, seems to be less explicitly addressed. Furthermore, the primary involvement mechanism of the more “technology-driven approach” is either to more passively observe, collect, and analyze data to map human behavior, or to stimulate or coerce humans to participate by providing data, devices, or input (Table 5). In this regard, while slightly over half of the articles were coded as adopting an expert mindset, 58.3% were characterized as having a traditional approach towards user involvement. Within this group of articles, the involvement of different user groups and other stakeholders was less common. As a results, highlighted by Table 5, most of the technical contributions focused on novel frameworks or architecture, which, however, are not able to address some key outcomes such as agency and civic participation. Furthermore, as was shown in Tables 6 and 8, only slightly over one-third of the analyzed articles were based on high user involvement and only one of the empirical studies, being the study of Karni et al. (2022) reported on an entire human-centered design process. This points towards a discrepancy between how contributions are positioned and framed (“human-centric”), the practices and assumptions which they are based on, and their actual (potential) impact and ability to orient existing efforts towards genuine human-centric outcomes and key values.

5.1.4 A technical status hierarchy

The above mismatch may have important implications in terms of inherent power relations. It has been shown that when human-centred practises/processes are not adopted, and thus not impacting the decision-making of the technologists, the decisions that dictate codified procedures managed by technical measures and performance indicators are not factoring in who defines and enforces what the intended outcome is, and for whom, and from what position of power they are realized (Sloane 2019). In this respect, when technical experts and commercial actors mostly build the IoT/smart ecosystems, the primary measure of success will most likely be driven by the benefits of commercial development, legal requirements, cost-savings, and performance management/control over technical environments (Chin et al. 2019; Sovacool and Del Rio 2020). Commercial and business interests are likely to result in a (technology) push for “creating” needs instead of a pull for overcoming existing and future human and societal needs. An important assumption here is that the interests of those in control will also count for protecting the interests for all who are served by the technical system. Therefore, the matter of which stakeholders’ needs are specified as part of the requirements, and when they are considered in the design process, determines the human-centric success measurements (Cockton 2004). Then, to be truly empowering, the selected success measurements should be evaluated against a shared beneficial outcome that improves based on the actual needs of humans and society, instead of metrics created by those in control of the IoT system (Graeff 2018).

5.1.5 Holistic design frameworks

In the analyzed article set, the requirements for safeguarding humans’ interests are primarily translated as security, privacy, and control constructs, technically represented by protection from, e.g., unauthorized disclosure and mining of personal data or access to restricted resources. However, there is still room for improvement in terms of involving human users from their viewpoint in this translation process, actively and openly, allowing for choice and various interpretations and in line with the fundamental principles introduced in Sect. 2. The main mechanisms focused on universal access and protection of human rights are those that maintain knowledge and authority to the human users of the system. They can examine how to balance equal power between the constellation of actors (developers, users, government, technology), along with protection of the self-understanding as human agents (Feng et al. 2021). These are important advancements; however, they need to be integrated into a holistic framework that re-engineers design practices, based on principles that put people and their rights at the center of the development.

5.1.6 Participatory approaches where everyone is valuable

A humane and holistic design framework for IoT entails considerations that would work towards safeguarding user rights by regular consultations with diverse audiences in civil society, technical features to flag, prevent or mitigate misuse, and audits where the fault lines of existing power structures are addressed, to ensure equitable and empowering outcomes for all. In addition, assessments of the effect on humans’ daily lives, incorporating socio-technical processes and measures to perform evaluations, can reduce certain forms of negative outcomes, such as bias, discrimination, and inequality (Hasselbalch 2021). The development of these assessments and guidance can be done via focused studies, monitoring, and analysis, which requires engagement with interdisciplinary groups, including computer scientists, social scientists, psychologists, economists, and lawyers (Stephanidis et al. 2019). While the literature search in our study yielded a number of recent examples of participatory approaches in this respect, e.g., the study by Tiersen et al. (2021) focusing on people with dementia and IoT smart home solutions or that by Karni et al. (2022) addressing the potential of IoT to empower people with Parkinson’s disease, such examples are still scarce in the literature.

5.2 Towards a future research agenda

Based on the analyzed papers, the following aspects need to be explicitly considered in future efforts towards building a more human-centric and empowering IoT.

5.2.1 The role of non-human actors

In designing for new IoT technologies, the role of non-humans in new IoT/smart environments should be clear and well-understood, especially with regards to defining the relationship with humans. Furthermore, conceptual models of knowledge and action shared between humans and non-humans need to be developed (Cruickshank and Trivedi 2017b). To better understand the resulting tension between the two, a critical analysis of non-human versus human agency, by addressing their theoretical underpinnings (i.e., from the field of Science and Technology Studies / Actor-Network Theory), can help determine the appropriate design and deployment that will safeguard user-rights (Sloane 2019). From the human perspective, more research into how to provide agency to human users is needed, with the development of mechanisms that are at a standard whereby each human actor is able to understand the situation for themselves (Lindley et al. 2018). This calls for the design of novel frameworks and user interface components to provide humans with a more transparent and informed view of the role of the provided technologies and the impact at large of decisions taken (Benhamida et al. 2021). Additionally, transparency should also focus on the deployment and management phases, as the system configuration of even well-designed solutions may significantly impact the users’ agency. From a technical perspective, there is ongoing research into the design for collaboration and information exchange between non-human actors based on human-like features and qualities (Defiebre and Germanakos 2019). The social IoT is tentatively able to smooth the interactions. However, the development of effective social behaviors still needs further development, especially in terms of the impact on the attitude of artificial social peers towards humans.

5.2.2 Diverse user involvement throughout the design process

To have a high level of trust, which is crucial for user adoption in IoT, humane, ethical, and legal considerations must be considered throughout the design process (Shin and Park 2017). Only a few of the analyzed contributions consider early user involvement. The literature to date identifies a need to facilitate the active involvement of various user groups with diverse needs, which requires a movement away from a single user perspective towards a community based or society level of involvement. Balancing multiple interests also requires systems design to address the tension between growing individual needs, private interest, and the collective social responsibility towards the common good (Semeria 2016). These processes should be humanity-based from the start, include regular consultations with multiple stakeholders and robust transparency reporting on activities that may have an adverse impact on human rights (Winter 2015). Finally, future IoT environments need to facilitate massive user involvement, in multiple environments, with new contexts (Gracanin et al. 2019). This may include recommendations for technical design frameworks where multiple stakeholder interests are balanced with system monitoring, and the design is governed throughout the development process.

5.2.3 New methodological approaches

Moreover, there is a mismatch and an overall lack of evaluation of the targeted outcomes to address and manage the effects that IoT systems may have on human users, communities, and society (Memedi et al. 2018; Shin and Park 2017). As such, the technical aspects should be more strongly linked with methods that are more suitable for evaluating the potential human and social impact in the context of IoT. Until now, the evaluation of the network systems’ ability in meeting the user expectation has been quantified using QoE metrics. The scope for improvements into QoE provides an insight into future research to define meaningful human-centric metrics (Geerts et al. 2010). For most human-centric IoT contributions, real-world evaluations are scarce and limited, although future studies plan to conduct systematic assessments of the IoT technology’s effectiveness, efficiency, comprehensiveness, and neutrality (Feng et al. 2021).

5.2.4 Discrete and unobstructed technology/merged machine-human interactions

In the subsequent development of human-centric IoT solutions, new research into tools that can equip users to obtain information, knowledge, and control of discreet IoT technology hidden in the background, often where the most intimate/private data is collected, is essential. In the past, where users only assisted the network in providing a better individual service, it should now evolve into future user equipment that is merged into the IoT architecture to form a deep-fused human–machine intelligent system (Adamo 2019). The gap between technology artifacts and the human experience, with thin and rigid distinctions separating humans and machines, and how these boundaries are established in the connected life compared to the non-connected life, are research topics that will help bring this vision to life. The research challenge of implementing trade-off procedures and conscious user choice remain, when guiding design principles are not adopted (Streitz 2019).

5.2.5 Multi-disciplinary expertise

Future research agendas would benefit from a significant effort to bring together expertise from different communities (e.g., computer science, physiology, cognitive science, communications, technology ethics, and user experience design) to jointly design IoT systems that are able to deliver empathetic communications and instill trust into users (Kaluarachchi et al. 2021). Even so, the challenges related to establishing a clear link between more human-centered aspects and the technology implementation relate to guidelines geared towards legal or design, not technical domains, developers passing the responsibility onto experts, and a lack of access to multi-disciplinary skills (Aljeraisy et al. 2021).

5.3 Limitations

We acknowledge that the methodology may have limitations due to keywords, selection, and eligibility criteria restrictions as described in section 3. However, we have addressed these by applying a rigorous search strategy upfront, grounded in the existing literature. Moreover, we have developed a standardised coding framework for the analyses supported by specific definitions that had to be explicit in the text. However, some articles did not explicitly address the variables of interest, and in such cases, we rigorously considered them to assign the most appropriate category correctly. For variables such as the geographical origins of the team, institutional information was used to assign the paper to the right category. However, due to this, the applied coding does not take into account that researchers are mobile and that for instance, teams located in Europe may have a more diverse composition. In other research papers, we were confronted by a need for shared terminology and definition of certain concepts/words. For example, when categorizing targeted user outcomes codes, such as privacy, agency, and empowerment, the grouping of the annotations and codes was first based on an explicit mention, then mapped to the most relevant goal that was expressed based on an interpretation. Despite this, we were able to define most other variables quantitatively with objective indicators. Based on these observations, we recommend that future research projects consider the discrepancy in understandings and explore the possibilities of clarifying and unifying the terminology used to achieve human-centric design outcomes and practices (Law and Van Schaik 2010), especially when aiming for a more inclusive, democratic, and empowering design.

6 Conclusion

This survey paper has covered the theoretical ground, conceptual definitions, methodologies, and the associated solutions to human-centric IoT technology. In particular, our aim has been to review the literature from a socio-technical lens to assess the re-alignment of IoT network technology towards human-centredness and sustainability. Furthermore, we have sought to address the divide between human-centered/humanity-centred theory and practices in IoT by detailing the existing scope for what human-centered translates into and how it can be practically implemented in the context of IoT.

A key observation is that, despite the recent increase in research on humane perspectives for IoT, “human-centredness” often still seems to be used primarily as a label and overarching paradigm, without leading to a profound and necessary change to underlying practices. This is also reflected in the under-representation of more human-oriented fields and disciplines and the associated approaches and practices, which still holds, also for the most recent years. When considering key principles of human- and humanity-centric design and their implications also reflected in the European declaration of digital rights, the human-centric label, therefore, seems to be misrepresented to a certain extent. This observed mismatch between higher-level intentions and reality calls for renewal and re-alignment at several levels.

We argue that increased multi-disciplinary research, such as the work presented in this paper, can help to shape how future human-centric IoT roadmaps may re-align towards one that benefits humans and humanity first, leading to a successful adoption on a societal scale. Future multi-disciplinary efforts need to be guided by a shared understanding of the fundamental underlying principles with the development of a common research agenda, and to have a shared vision of how the intended outcomes can be best addressed and evaluated.

We further call for more critical theory-based research on human-centric IoT. Research that challenges the “inevitability” of an IoT where the path represents a ubiquitous/pervasive presence of IoT devices in everyone’s living environment is scarce. However, without exercising active human choice and direct participation in the development, the design equality of pervasive IoT environments needs to be challenged. This requires critical socio-technical values, factors, and principles to be considered more extensively and explicitly to actively enable human potential when everything is connected. Furthermore, the infusion of IoT with human users’ everyday life and practices, requires that human users’ are actively motivated to use it and for them to hold authority over the use of system (Kim and Gupta 2014). In order to be genuinely human-centered, IoT technologies require a clear-eyed model of user empowerment, inclusivity, human control, and involvement. Therefore, future work in this domain must ensure that human-centered IoT is at the core of societal, technological, economic, and political agendas.