Introduction

Computers, robots, drones, cameras or integrated systems are increasingly becoming part of our working environments and are organized in networks, constituting the Internet of Things (IoT). This unprecedent deployment of interrelated information technologies in the 21st century has improved productivity in many sectors and contributed to major transformations of working conditions, sometimes leading to intensive algorithmic management and sectorial replacements of workers by machines [1–4]. Optimistic authors announce an “age of augmentation”, where humans and machines co-evolve harmoniously in an integrated manner [5]. Whether positive or negative, such technologized work environment is made possible by the deployment of smart devices, called Internet of Things devices (IoTs). They are equipped with information sensors and connected through Internet-based communication networks. They can collect data en masse on work production, individual characteristics, and group dynamics. This data enables the analysis of workflows with unprecedented granularity, leading to reliable predictions and preventive and adaptive measures.

IoTs can also serve occupational safety and health (OSH) purposes. They include portable cameras, augmented or virtual reality tools, wearables, apps, smart clothes, or protective equipment such as exoskeletons [6, 7]. In a proactive manner, these devices are used to assess risks and prevent work-related illnesses or safety incidents. In a reactive manner, they are used to improve the quality of the investigation and reporting of adverse events or to reduce their negative effects [8].

Despite the value of IoTs for OSH purposes, their deployment may involve worrying drawbacks and raise ethical issues that are overlooked by relevant stakeholders. Policymakers and stakeholders implicated in the implementation of IoTs in working contexts are ambivalent regarding ethical issues. Some warn of the risks posed by IoTs, while others are extremely optimistic about the value of the technology [9–11]. For years, scholars in various fields (including the social sciences, law, medical ethics, business ethics, and ethics of new technologies) have pointed out worrying issues, such as unexpected adverse effects on the mental and social well-being of workers, or cases of problematic dual use when IoT for OSH also collect data for surveillance purposes or performance analysis [12–20]. These many relevant discussions are somewhat scattered over various academic fields and interwoven with related but broader topics of discussion, such as the social impact of AI, algorithmic management in a work context, or self-quantification of health [21]. This literature, however, clearly points out the need to assess the technology ahead of its deployment and prevent possible abuses and questionable uses.

At the empirical level, there is also interesting burgeoning work on stakeholders’ evaluation of IoTs. In a survey addressed to 702 European managers, Pütz and colleagues [22] found significant acceptance of IoT devices specifically designed for monitoring mental health. Nonetheless, positive expectations were counterbalanced by privacy concerns. In a qualitative study, Bovens and colleagues [23] investigated personnel’s view on the values relevant to deploying an occupational health management IoT. Their participants allotted importance to the values of security, trust and hierarchy, but did not report a strong need for ethics support. In another study, Le Feber et al. [24] explored the ethical and privacy issues that may occur with the introduction of sensors in the workplace with 32 stakeholders. The stakeholders were open-minded to the use of sensors for applied purposes such as OSH. Participants expressed the following: end users should be involved in the decisions on the purpose and application of IoT, possible negative adverse effects need to be considered, and continuous sensor-based monitoring is too intrusive. While important, these results nevertheless remain limited and difficult to generalise. Notably, the Bovens et al. study involved a very specific category of participants: reservists who had previously accepted use of such a device. It is unclear to what extent participants’ evaluation was influenced by the social values of their profession and by an a priori positive attitude towards IoTs. And to frame the interview discussion, Le Feber et al used an ethical grid that included the values of fairness, proportionality, and intrusiveness. It is therefore unclear whether participants would have come up with the same ethical themes without being explicitly primed.

In this context, we consider that there is still a need for a more encompassing work in the area of the ethics of IoTs. Firstly, multidisciplinary and fragmented ethically relevant information and reflections need to be collated in a comprehensive overview. Secondly, more data is needed on the points of view of stakeholders, including company managers, OSH specialists, and first-hand users (employees). It currently remains unclear which ethical issues are relevant to them, and to what extent they are aware of the relevant benefits and risks associated with the use of these devices in working contexts.

Relevant ethical insights on IoTs can be drawn from academic literature and from stakeholders’ point of view. In the context of a larger research project aiming to provide a comprehensive overview of the ethical opportunities and problematic issues posed by IoTs for OSH purposes, we used a mixed method approach, conducting a scoping literature review (reported in a separate paper [21]), and a complementary empirical investigation involving stakeholders (presented here). This double methodological perspective enables to identify ethical concerns experienced by stakeholders that have received limited or no attention in the existing literature (business ethics, digital ethics and health ethics), and, reversely, to highlight issues that are discussed in the literature, but systematically ignored by stakeholders due either to a lack of awareness or to an underestimation of their significance.

The literature review [21] revealed an unexpectedly detailed and substantial list of issues posed by using IoTs for OSH purposes, including the topics of surveillance and problematic data re-purposing, the risk of suboptimal data management, and the difficulty of informing, consulting, and obtaining consent from employees. The review also revealed an important number of possible unintended and unpredictable adverse effects (e.g. weakening of workers’ competencies and self-confidence, exacerbation of social stigma and inequalities, deresponsibilization of employers), in addition to contextual factors that enhance risks and ethical difficulties (e.g. blurred limits between professional and private life, complexities and lack of competencies, costs of ethical safeguards).

This paper presents the results of the empirical investigation involving one focus group and a series of qualitative interviews. The goal was to investigate the opportunities, challenges, and ethical issues, as viewed by a large variety of stakeholders in possession of firsthand experience with IoTs for OSH purposes (workers, worker representatives, policymakers, managers, OSH specialists, occupational physicians, and critical analysts). For the sake of concision and relevance, we limited our research on IoTs deployed for the purpose of preventing health hazards (proactive approach). These devices are the easiest and most likely to be implemented by large and small-scale companies and have received little attention in the ethics literature. We analysed our data with a grounded theory method and conducted the literature review mentioned above at a later stage to minimize the theoretical influence of academic discussions on the empirical analysis presented here. In the discussion section, we compare and discuss the outputs of both lines of research.

Overall, our results help to structure the debate on the ethical issues at stake. We provide fundamental knowledge for the critical assessment of IoTs deployed for OSH purposes, and for the development of ethical frameworks, guidelines, regulation instruments, safeguards, and accompanying measures to mitigate individual and societal risks. We also highlight ethical issues that are likely to be overlooked either in the academic realm, or among stakeholders and decision-makers. Our results are thus helpful for theoreticians and policymakers.

Materials and method

We conducted one focus group and semi-structured individual interviews with experts (employee with user experience, advocate of employees, company manager, Human Resources manager, occupational physician, medical information physician, cantonal physician, occupational safety and health professional, public administrators, health insurance representative, IT developer, critical analyst, tech journalist, lawyer). We chose this double method because focus groups usually enable debate, discussion, sharing of ideas, and generate more refined lines of thought, while individual interviews provide more space for shyer participants to speak, especially for participants from lower down in the hierarchy. Interviews also allow to adapt the language (French, English) to participants’ ease.

Participant recruitment

To map as many ethical issues as possible and identify different points of view, following a maximum variation sample method [25], our goal was to include a broad range of actors in different domains of activity related to IoTs (see detail in Table 1): we aimed at covering knowledge and competencies related to 1) the development and production of IoTs (e.g. IT developer, manager in a start-up producing IoTs), 2) the deployment and use of IoTs in private companies (e.g. private company manager, Human Resources manager, OSH professional, occupational physician, end-user employees, advocate of employees, health insurance representative), and 3) the assessment, control and regulation of such use (e.g. lawyer, public administrator, cantonal physician, tech journalist, critical analyst working in a university or think tank). Overall, we also aimed at a diversity in gender, hierarchical level on the workplace, socio-cultural regions in Switzerland, and a variety of public and private working sectors (e.g. university, high tech start-up, federal administration, think-tank, media sector, goods-producing company, selling company, politics). Given that Switzerland is a small country and given the specificity of the type of participants we required, we did not use a random sampling method but invited stakeholders based on personal recommendations or online searches of profiles.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of study participants.

https://doi.org/10.1371/journal.pone.0315671.t001

Data analysis

Participants’ answers to the closed questions are compiled in the descriptive Table 2. To leave space for unanticipated information to emerge when analysing participants’ answers to open questions, we employed a method involving elements from grounded theory [26]. First, we transcribed the recording verbatim in the original language with help of the software Word, pack version Microsoft 365, and anonymized the data. Coding was done in English only. After an immersion phase of reading and note taking, we proceeded to the initial coding of the verbatim transcription using software (Atlas.ti). To guide this initial step, we used a “start list” of codes (applied top-down), containing a limited number of broad categories classically used in value maps (autonomy, justice, safety, security, solidarity, etc.), goods (health, well-being), harms (dependency, exploitation), and concerns (for privacy, for well-being, etc.). All other first-level codes were generated bottom-up. While proceeding, items from the start list were modified, expanded, or deleted until we obtained a unified (top-down and bottom-up) list of codes. We then focused on codes only and searched for emerging themes and sub-themes, using word clouds (Atlat.ti function) and mindmaps. We renamed, merged, split or deleted codes whenever necessary, until we obtained a clear branching organization around a low number of overarching categories (final conceptual framework). While proceeding, we systematically checked categories, themes and first order codes matched with the verbatim transcription. All steps were mainly carried out by the first author. In addition, 50% of the initial transcripts were double coded independently by the second and last authors and an independent colleague, then compared and discussed. 80% of the thematic organization of codes was done or reviewed in closed feedback loops during regular meetings between the first and last authors. The second author occasionally checked overall consistency, and matching between categories, themes, and verbatim transcription, to ensure that concepts were clearly defined, appropriately derived from the data, and that codes and themes were being used consistently.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary of participants’ responses to our closed questions.

https://doi.org/10.1371/journal.pone.0315671.t002

Two-step analysis (with and without answers to follow-up questions).

Since we asked follow-up questions designed to attract participants’ attention to relevant ethical issues, overall responses were biased towards the topics of our follow-up questions. To account for this, we separately analysed participants’ responses before the follow-up questions. In a second stage, we added follow-up responses to the data set and observed which additional codes emerged. This enabled us to report participants’ responses with and without additional guiding questions.

Subgroup analyses.

To observe whether different types of participants were more (or less) sensitive to specific ethical issues, in a post-hoc analysis, we created four subgroups of the dataset: women, men, participants with a conflict of interest in favour of the technology (decision-makers, health insurance representatives, IT developers), and participants without a conflict of interest in favour of the technology (employee with user experience, health professionals, occupational safety professionals, critical analysts, public administrators, lawyers). We only kept codes related to discussions of ethical issues (e.g. removed codes related to the benefits of the technology, to the condition of acceptability of the technology, or to descriptions of external factors), and calculated, within each subgroup, the mean proportion of first order codes contained in the five overarching categories. This enabled descriptive between-groups comparison.

Broad scoping literature review.

Near the end of the interview analysis (summer-autumn 2023), we reviewed conceptual and empirical academic papers discussing ethical issues related to the use of IoTs in OSH contexts. Following the PRISMA-ScR checklist [27], our goal was to list and describe the main opportunities and issues as discussed in the academic domain (detailed method and results are submitted in a separate paper). We chose this timing and procedure to minimize the theoretical influence of academic discussions on our empirical analysis. We then compared the two sources of information (interviewed stakeholders and academic literature), and searched for similarities, differences and blind spots. The main findings are commented in the discussion section.

Results

The focus group included 7 participants. Due to a snowstorm, 3 participants joined late and missed the discussion of the connected chairs scenario (Posture-tracker Scen). We conducted 17 further individual online interviews. We reached data saturation before fulfilling our aimed variation sampling (total n = 24): coding of the last 7 interviews generated few new codes (88 out of 2202 in total) and none of these codes provided substantive added content.

Our sample, described in Table 1, includes 5 women and 19 men, which is reflective of the representation of women in the field of the Internet of Things technology [28, 29] with an average age range of 40–50. The focus group discussion lasted 3h30, and the individual interviews lasted between 45 and 80 min (60 min on average).

Answers to main open questions

Overall, while evaluating the advantages and ethical issues related to the three scenarios, participants expressed numerous ideas, generating 2202 codes that we grouped into five overarching categories: goal relevance, adverse side effects, role of employees, data process, and vagueness. The full book of codes is available on the OSF platform of the project: osf.io/e97v2.

The main results are summarised in Tables 3–7. They include information about the number of participants that mentioned the codes (nb-participants), and about the overall number of occurrences of those codes (nb-occurrence). The tables also systematically differentiate between numbers of full results (total including follow-up discussions) and results that came up before we asked follow-up questions (total excluding follow-up discussions). More detailed descriptions of the results are provided in S2 Text.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Goal relevance.

https://doi.org/10.1371/journal.pone.0315671.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Side effects.

https://doi.org/10.1371/journal.pone.0315671.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Role of employees.

https://doi.org/10.1371/journal.pone.0315671.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Data process.

https://doi.org/10.1371/journal.pone.0315671.t006

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Vagueness.

https://doi.org/10.1371/journal.pone.0315671.t007

Sub-group analyses

The subgroup analysis revealed no major differences in participants’ sensitivity to specific ethical issues when stratified by typology of participant. As illustrated in Fig 1, women, men, and participants with or without conflicts of interest gave similar weight to each of the five groups of ethical issues. From the figure, we can see that participants generated more input (number of codes) on topics related to the data process. The vagueness category is artificially small: numerous codes included in the four other categories could also have been classified here.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Subgroup analysis ‐ mean proportion of first order codes (related to ethical issues) contained in the five overarching categories (goal relevance; adverse side effects; role of employees; data process; vagueness) for each four sub-groups of participants: Women (n = 5), men (n = 19), participants with (n = 7) and without (n = 17) a conflict of interest.

Nb codes = total number of codes produced by all participants in the subgroup.

https://doi.org/10.1371/journal.pone.0315671.g001

Discussion

Our study’s findings are rich in content and shed light on how stakeholders react to an emerging realm of the digital workplace. In this section, we discuss our findings, highlighting the importance of one unexpected category (vagueness), link the empirical results of this research with our complementary literature review, and with similar empirical results.

First of all, it must be noted that participants were favourably disposed towards the use of some IoTs for OSH purposes (e.g. posture-tracker chair or step-tracker watch) while rejecting more privacy-invasive devices (e.g. sound-tracker on computer). This result is similar to previous results [22, 24].

Our findings show that participants pointed out an important number of ethical issues, which we grouped into five overarching categories. First, participants reflected on the goal relevance of the device. While acknowledging the most obvious OSH benefits provided by the technology, they expressed strong concerns about effectiveness and lack of clarity regarding employer goals, particularly problematic dual uses or later repurposing of data. Second, they were preoccupied with adverse side effects, for employees (e.g. psychological impact of monitoring, risk of algorithmic discrimination or distorted incentives and responsibilities), and for companies and society (e.g. risk of damaging trust in work relationships, of generating new cheating habits, of making surveillance more common). They emphasised the unpredictability of many adverse effects, making it difficult to set up efficient gatekeeping measures. Third, participants discussed the role of employees (the primary users of the devices), highlighting that their point of view in consultation processes and the quality of their consent were compromised by a number of factors: the complexity of the technology and data process, a lack of technological literacy and relationships of dependency. Participants emphasised that fair information, understanding of relevant issues and freedom of choice were difficult conditions to meet, undermining the relevance of consultation and consent procedures. Fourth, participants shared numerous concerns regarding the various stages of the data process, including data collection, transmission, storage, analysis, and overall governance. Most revolved around security and confidentiality issues; there was repetitively expressed distrust towards HR and frequent complaints about the lack of trustworthy actors, or existing trustworthy actors’, such as occupational physicians, lack of power. They were insecure about their own capacity to identify the relevant facts (about the technology, the data flow, the status of the law, etc.) necessary for sound ethical evaluations.

Our last category, Vagueness, encompasses the aforementioned insecurity issue, as well as many further concerns. Participants expressed recurrent concerns about a lack of background knowledge and the intrinsic difficulty of obtaining relevant information for assessing where ethical problems lie. Vagueness issues occurred repetitively across all the other four topics. Participants expressed their lack of knowledge about multiple aspects that are important to understand in order to produce a sound ethical evaluation, such as how does the technology work; what is truly intended with its deployment; who gets access to the data; who knows (and controls) who gets access to the data; to what extent can we expect end-users to understand the relevant features of the technology and the consequences of its deployment; what are the long-term effects of the technology; what uses are legally acceptable; how can unavoidable conflict of interests be attenuated; what are employers’ obligations; to what extent do IoTs fulfil true OSH goals, and what are the relevant methods and criteria for assessing such technology.

This vagueness issue is worrisome because withholding one’s judgement is a wise attitude in the face of felt incompetence, but it can lead to a kind of ethical resignation, leaving the floor open to all sorts of unregulated social experiments. Despite having been selected as the most knowledgeable people on IoTs, participants often felt unsure about what was is at stake in particular domains. For instance, in their answers to our closed questions, many of them confessed a lack of knowledge to assess the legality of our case scenarios, and those who did assess them expressed opposing opinions. Despite frequently expressing worries about lack of knowledge, the vast majority said that, in practice, they would endorse the deployment of an IoT such as the posture-tracker chair or the step-tracker watch, demonstrating either ethical naivety or resignation. In the domain of IoT for OSH goals, hardly anyone seems to feel fully competent, resulting in suboptimal ethical evaluation and a lack of precautionary or preventive measures. Such a result may indicate that stakeholders need to be supported by regulatory and political bodies that are responsible for supporting the ethical process.

Another interesting finding, linked to the vagueness issue, is the extent to which stakeholders miss relevant ethical thoughts if they are not actively informed of the details of the implementation procedure in real life. Participants in our study produced numerous novel insights once we showed them the data flow picture and asked follow-up questions. Although our follow-up inputs were purely descriptive (for methodological reasons, we did not want to introduce ethical themes during the interviews), they enabled participants to refine their analysis and further explore ethical aspects overlooked in the first place. We collected more thoughts on issues related to data quality, access, and governance, and on the rights, roles, and trustworthiness of stakeholders, including external third-party companies. The follow-up process also revealed divergences of opinion among stakeholders. They disagreed on whether consultation of employees should be done in direct and democratic ways or through discussions with knowledgeable representatives. Some stakeholders objected to all forms of data repurposing, while others were ready to accept some repurposing upon sufficient justification.

Beyond this inquiry in two stages (questions and follow-up questions), participants’ insights present interesting commonalities, differences, and blind spots with the topics discussed in the academic literature [21].

Overall, we found broad agreement among stakeholders, including those with or without conflicts of interest (as illustrated in Fig 1). The ethical issues they highlighted were fully in line with the worries expressed in the academic literature. Such a pattern is unfamiliar in practical ethical debates, often characterised by fierce fights between irreconcilable camps (think, for instance, of abortion, immigration, or environmental debates). This result engenders both hopes and worries. Hopes because effective solutions in communities can only be achieved through agreement on what matters and what should be improved. Worries because it is easy to agree on theoretical matters that have no practical effect whatsoever. An observed agreement may thus be a sign that we are far from any actionable solution to address the ethical issues at stake (laws, regulations, guidelines, safeguard and mitigation measures).

Regarding differences with the academic literature, participants tended to express strong scepticism about the expected efficacy of IoT for OSH purposes and about the good intentions of employers who deploy them. They complained heavily about the lack of powerful trustworthy actors, making it difficult to enforce ethics-by-design procedures and mitigation measures. Participants expressed strong doubts about the applicability of fair consultation and consent procedures and, consequently, awarded them lesser value as an ethical means to protect employees. Such a resigned attitude was less noticeable in the academic literature [with exceptions e.g 16, 30].

Participants highlighted practical issues overlooked in the academic literature, such as the difficulty of establishing efficient safeguards without compromising the adequacy of the devices or the difficulty of defining employers’ legal obligations. Further, they discussed at length the practical difficulties experienced by particular professions (HR, direct managers, OSH professionals, occupational physicians) when juggling a need to please employers while protecting employees, which are seldom thematised in the academic literature.

Reversely however, participants tended to overlook or to express only peripheral interest in numerous issues. The list of unintended and unpredictable adverse effects was strikingly shorter in stakeholders’ reports. For instance, they hardly took seriously the risks of third-party exploitation of sensitive personal data that was a major concern in the literature [e.g 31–34], nor of workers’ excessive trust in employers [e.g 35]. Overall, participants in our study did not take a global and holistic approach and rarely came up with societal and political level considerations. For instance, topics related to cyber criminality [e.g 36, 37], international surveillance [e.g 31–34], nor of workers’ excessive trust in employers [e.g 35]. Overall, participants in our study did not take a global and holistic approach and rarely came up with societal and political level considerations. For instance, topics related to cyber criminality [e.g 36, 37], international surveillance [e.g 38–40], dystopian workplaces [e.g 32, 41], or social moves towards a banalisation of surveillance were missing [e.g 40, 42]. Similarly, participants hardly commented the role (and current limitations) of regulatory bodies and institutional safeguards which is heavily discussed in the literature [e.g 43–49].

Several factors may explain participants’ stronger emphasis on subsets of ethical issues and their capacity to pinpoint specific issues that are overlooked in the academic literature. One is their first-hand experience that allows them to see vividly the problems that they have experienced, and, reversely, to overlook or express peripheral interest in issues that they have not experienced. We think that the risk that stakeholders underestimate some ethical issues because of a lack of personal experience is particularly high in the case of IoTs for the Good (i.e devices deployed for noble goals): the fact that those devices are deployed for OSH purposes tends to numb awareness of lurking ethical issues [e.g 16].

Another explanation of the observable differences between stakeholders’ insights and the academic literature, is the cultural background in which the former are embedded. When discussing ethical issues, participants in our study considered work-context issues, taking place in the horizontal hierarchical relationships that characterise the classical internal organisation of Swiss companies. Except for general complaints about lacking laws or trustworthy actors, participants showed little interest in evaluating legal, administrative and institutional safeguards. The Swiss data protection authority (préposé federal à la protection des données et à la transparence), which is supposed to play a crucial regulatory role, was not even mentioned. Such omission may reveal their low expectations of Swiss administrative and legal authorities. This is the mark of a strongly liberal culture within the working economy and public health.

Lastly, participants’ selective ethical considerations may be partly due to methodological reasons: they were asked to discuss three specific case-scenarios that, for instance, did not include societal considerations.

Finally, our results complement existing research [22, 24, 35]. The method that we used (inclusion of a large variety of stakeholders; use of descriptive scenarios and open questions free from ethical concepts; explicit presentation of the data process in the scenario; open coding focusing on ethical issues and not on ethical solutions) allowed us to obtain a larger range of ethical insights without influencing content. We replicated most previous results: stakeholders are globally open-minded towards IoT for OSH purposes, but less open towards more intrusive devices. We observed a sensitivity to issues related to security, trust, consultation, side effects and intrusiveness. We obtained a richer set of concerns, notably related to the first-hand relevance of the technology, to various issues surrounding the data process, the lack of trustworthy actors and the overall lack of information enabling objective assessment of the technology. In contrast to Bovens, we did not find topics related to hierarchy and our stakeholders repetitively expressed a need for support to objectively evaluate the technology.

Conclusions

Our study provides novel insights into the perspectives of relevant stakeholders regarding the digitalization of the workplace and the implementation of IoTs for OSH purposes. The numerous ethical issues highlighted here and in our connected literature review [21] indicate that even if deployed for OSH purposes, IoTs can easily lead to irresponsible use. Stakeholders’ inputs were partly disenchanted and shed a crude light on the lack of information at their disposal even while in the front line to assess such technology.

Increased knowledge about relevant ethical dimensions is a necessary first step to foster responsible use of this technology in the workplace. Future research should aim to address some of our limitations by including a more culturally diverse sample of participants and varying the case scenarios. Nevertheless, our results provide important ground material, crossing ethical fields (business, health and digital ethics), on which to base necessary and still awaited ethical frameworks for assessing such technology, guidelines and regulation instruments.

Supporting information

Acknowledgments

We thank Marie-Gabrielle Mansour for her contribution to the design and focus group, Dana Naous for her contribution to the design and double coding.