Do You Care for Robots That Care? Exploring the Opinions of Vocational Care Students on the Use of Healthcare Robots

: Background : There has been a rapid increase in the population of senior citizens in many countries. The shortage of caregivers is becoming a pressing concern. Robots are being deployed in an attempt to ﬁll this gap and reduce the workload of caregivers. This study explores how healthcare robots are perceived by trainee care professionals. Methods : A total of 2365 students at different vocational levels completed a questionnaire, rating ethical statements regarding beneﬁcence, maleﬁcence, justice, autonomy, utility, and use intentions with regard to three different types of robots (assistive, monitoring, and companion) along with six control variables: gender, age, school year, technical skills, interest in technology, and enjoying working with computers. The scores were analyzed by MANOVA statistics. Results : In relation to our research questions: All students viewed companion robots as more beneﬁcent than monitoring and assistive robots. Level of education did not lead to any differences in appraisal. Participants rated maleﬁcence lowest and the highest scores were given to autonomy and utility, meaning a positive evaluation of the use of healthcare robots. Surprisingly, all students rated use intentions low, indicating a poor motivation to actually use a robot in the future, although participants stated a ﬁrmer intention for using monitoring devices. Conclusion : Care students ﬁnd robots useful and expect clients to beneﬁt from them, but still are hesitant to use robots in their future practice. This study suggests that it would be wise to enrich the curriculum of intermediate care education with practical classes on the use and ethical implications of care robots, to ensure that this group of trainee care professionals fully understand the possibilities and potential downside of this emerging kind of healthcare technology.


Introduction
The rapid increase in the number of older adults (75 years old and up) is a major concern of the developed countries [1]. In the 1950s, the probability that an 80-year-old would survive to age 90 was 15-16% for women and 12% for men. By 2002, this percentage had increased to 37% and 25%, respectively. Since 1840, female life expectancy has increased by approximately two years per decade, worldwide. This linear growth has yet to stagnate and therefore implies that 'the human life span is not closing in on its limit' [2]. Life expectancy at age 20 is predicted to increase by approximately one year per decade for females and males between now and 2040 [3]. Compared to the working-age population, the proportion of older adults will eventually grow to a point where there are not enough RQ-2: Do perceptions of healthcare robots differ between vocational students at lower and middle levels? Related, do they perceive the various robot types differently? RQ-3: How do prospective care professionals evaluate care robots in terms of utility and possible use intentions? Related, how do these evaluations differ per robot type?

Participants
Students of three different Dutch vocational care programs-Helping Care and Cure, Helping Extramural Care and Nursing-volunteered in a questionnaire study. To acquire as many participants as possible, managers of seven different vocational education and training institutes distributed our online Qualtrics questionnaire 1 by forwarding an invitation and link to their students in care. A total of 2365 eligible students completed the questionnaire. This number represents 3.9% of the population of all registered (61,244) lower and middle vocational care students in the Netherlands [22]. On average, there are 5000 students per institute of which 30% follows an education in care [23,24]. That would make a total possible response of 10,500 participants. Therefore, a rough and conservative estimate of our sample response rate is 23%. All participants remained anonymous. The incentive for participation was one of five gift vouchers of €50 that were raffled among the participants.

Data Collection
Participants received a link to an online questionnaire that was specifically developed for this study. The questionnaire had three versions, one for assistive, one for monitoring, and one for companion robots; each version consisted of 39 questionnaire items. The items per robot type were identical so they could be compared. Demographics were probed using seven additional questions.
After opening the link in Qualtrics (Version 24(892), Provo, UT, USA)a brief introduction and consent form was presented. Upon agreement, the participant could commence the questionnaire. Participants were randomly assigned to one of the three versions. The questionnaire opened with a picture of a healthcare robot ( Figure 1) followed by a brief description of its capabilities and the tasks it could execute (e.g., reading aloud, exercise coaching, or reminding of medicine intake). ipants tudents of three different Dutch vocational care programs-Helping Care and Cure, Hel mural Care and Nursing-volunteered in a questionnaire study. To acquire as m ipants as possible, managers of seven different vocational education and training insti buted our online Qualtrics questionnaire 1 by forwarding an invitation and link to their stud e. A total of 2365 eligible students completed the questionnaire. This number represents e population of all registered (61,244) lower and middle vocational care students in rlands [22]. On average, there are 5000 students per institute of which 30% follows an educ e [23,24]. That would make a total possible response of 10,500 participants. Therefore, a ro onservative estimate of our sample response rate is 23%. All participants remained anonym ncentive for participation was one of five gift vouchers of €50 that were raffled among ipants.
ta Collection articipants received a link to an online questionnaire that was specifically developed for . The questionnaire had three versions, one for assistive, one for monitoring, and on anion robots; each version consisted of 39 questionnaire items. The items per robot type ical so they could be compared. Demographics were probed using seven additional questi fter opening the link in Qualtrics (Version 24(892), Provo, UT, USA)a brief introduction nt form was presented. Upon agreement, the participant could commence the questionn ipants were randomly assigned to one of the three versions. The questionnaire opened w e of a healthcare robot ( Figure 1) followed by a brief description of its capabilities and the ld execute (e.g., reading aloud, exercise coaching, or reminding of medicine intake). asures e measured six theoretical variables, which as a group we will term appraisal domains ived beneficence, maleficence, justice, autonomy, utility, and use intentions), four educat (levels 1-4), and three robot types (assistive, monitoring, and companion) along with ol variables: gender, age, school year, technical skills, interest in technology, and enjo ing with computers. Education Level was an ordinal variable: lower vocational education (l 2) and middle vocational education (levels 3 and 4). The theoretical variables were meas

Measures
We measured six theoretical variables, which as a group we will term appraisal domains (i.e., perceived beneficence, maleficence, justice, autonomy, utility, and use intentions), four educational levels (levels 1-4), and three robot types (assistive, monitoring, and companion) along with six control variables: gender, age, school year, technical skills, interest in technology, and enjoying working with computers. Education Level was an ordinal variable: lower vocational education (levels 1 and 2) and middle vocational education (levels 3 and 4). The theoretical variables were measured at quasi interval level with initially six Likert-type items, each rated on a six-point scale (1 = strongly disagree, Robotics 2019, 8, 22 4 of 12 2 = disagree, 3 = slightly disagree, 4 = slightly agree, 5 = agree, and 6 = strongly agree). Principal Component Analyses and Reliability analyses revealed that several items had to be discarded to form reliable scales. The questionnaire items that were included per appraisal domain can be found in Appendix A. Scale reliabilities were calculated with Cronbach's alpha as reported in Table 1, using 0.7 as the cut-off point to decide whether a scale was sufficiently reliable or not [26]. The Justice scale failed the 0.7 criterion and had to be discarded entirely. Apparently, the way we measured justice did not converge into a solid underlying concept. All other scales performed well (Table 1). After recoding the counter-indicative items, the items on each scale could be summed and averaged to calculate a mean index (Table 1). Although 2365 participants completed the questionnaire, cells were not filled equally. Unfortunately, only 2 participants studied at level 1, which made it necessary to drop that level from our analyses. Then, we had to exclude another 38 participants because they had completed the questionnaire in an unreasonably short period of time (which was recorded by Qualtrics) and/or because they checked the same rating scale answer for each item. That left us with N = 2325 participants in the final analysis. Their characteristics can be found in Table 1. Table 2 provides the means and standard deviations of the appraisal domains. Table 1 shows that participants were evenly distributed over the three robot types (33% each). Table 1 also shows an overrepresentation of females, although in the care professions that is a valid ecological outcome [27]. Unfortunately, however, over 50% of the total sample studied at vocational level 4 and a mere 9.1% at level 2, which seriously jeopardized our questions on differences in perception between educational levels. Almost the entire sample self-reported that they were skilled computer users.

Results
We calculated the grand mean scores of our measurement scales ( Table 2) and ran a GLM Repeated Measures for the 5-leveled within-subjects factor of appraisal domain (beneficence vs. maleficence vs. autonomy vs. utility vs. use intentions) by the between-subjects factors of robot type (assistive vs. monitoring vs. companion) and education level (2 vs. 3 vs. 4).  Although 2365 participants completed the questionnaire, cells were not filled equally. Unfortunately, only 2 participants studied at level 1, which made it necessary to drop that level from our analyses. Then, we had to exclude another 38 participants because they had completed the questionnaire in an unreasonably short period of time (which was recorded by Qualtrics) and/or because they checked the same rating scale answer for each item. That left us with N = 2325 participants in the final analysis. Their characteristics can be found in Table 1. Table 2 provides the means and standard deviations of the appraisal domains. Table 1 shows that participants were evenly distributed over the three robot types (33% each). Table 1 also shows an overrepresentation of females, although in the care professions that is a valid ecological outcome [27]. Unfortunately, however, over 50% of the total sample studied at vocational level 4 and a mere 9.1% at level 2, which seriously jeopardized our questions on differences in perception between educational levels. Almost the entire sample self-reported that they were skilled computer users.

Results
We calculated the grand mean scores of our measurement scales (

Rating of Appraisal Domains through Ethics
The interaction between robot type, education level, and appraisal domain was not significant (p = 0.296) nor was the interaction between robot type and education level (p = 0.672).
However, the main effect of appraisal domain was significant with an intermediate effect size
Participants indicated higher intentions to use monitoring than companion robots (t (1550) = 3.46, p = 0.001). All other comparisons were not significant.

Overall Ratings of Appraisal Domains
In sum, independent of robot type or education level, maleficence and use intentions scored lowest while autonomy and utility scored highest. Independent of education level, participants judged that monitoring robots were more beneficent than assistive robots and assistive robots more beneficent than companion robots. Assistive robots were perceived as more maleficent than monitoring robots and more than companion robots. Monitoring robots had more utility than companion robots and as a trend, assistive robots also had more utility than companion machines. Participants indicated a firmer intention to use monitoring than companion robots. All other comparisons were not significant.
With respect to education level, independent of robot type, level 4 students perceived more beneficence than level 3 students, whereas level 3 saw, as a trend, more maleficence than level 4. Level 4 also assigned more utility to robots than level 3 students, and as a trend, level 4 deemed that robots increased the autonomy of patients, more so than level 2.

Conclusions
The purpose of the current study was to determine which of the four principles of Beauchamp and Childress [18] were most prominent in the estimations of lower and middle vocational care students with regard to working with robots in their future care practice.
Overall, students scored Maleficence the lowest, meaning that care robots in general were not seen as pernicious. However, students of care also rated use intentions low, indicating poor motivation to actually use a robot in the future. Possible beneficent effects of care robots were received with relative neutrality, whereas the potential increase in a client's autonomy was deemed considerable. Highest scores were obtained for the utility of robots, which is surprising in view of the students' reluctance to use them. On the whole, level 4 students were slightly more positive than level 3 or 2 students with respect to beneficence, utility, and the client's autonomy. All students viewed companion robots as more beneficent than monitoring robots and monitoring robots more beneficent than assistive robots. Although regarded as more maleficent, monitoring and assisting machines were also seen as more useful.
In all, these care students saw little harm in robots, found them useful, and expected clients to become more independent because of them; on the other hand, they were quite hesitant about using robots in their future practice. It is possible that the perceived potential effectiveness of robots for their work practice was affected by fear of job loss.

Discussion
One might think that current students of care have already acquired so-called '21st century skills', that is, they would have "information literacy, media literacy, and information, communication and technology literacy" [29]. Additionally, if that is the case, so theory has it, perceived usefulness and a positive attitude toward technology should increase the intention to use technology. Indeed, computer self-efficacy can act as an antecedent for perceived usefulness and positive attitude towards computer use [30]. 93.6% of our survey participants stated they were skilled computer users and 72.3% claimed to enjoy working with computers. Nevertheless, these 21st-century skilled students were not very eager to employ robots in their future work practice.
Admittedly, the vocational care students in this survey did not see much harm in care robots, for the patient that is. However, when it came to their future work practice, they were reluctant to envision employing care machines, in spite of their potential utility. Listed hereafter are a few possible explanations for this.

Possible Explanations for the Hesitation in Using Care Robots
Robots may not have been viewed as maleficent, but neither were they seen as being beneficial or helpful (beneficence was rated 'neutral'). Although useful for the instrumental side of nursing and caring (utility high), it might be that these students feared that robots would weaken the relationship between caregiver and care receiver (autonomy high), commonly considered fundamental to 'good care.' Hence, they were hesitant to work with robots. Lewis and West [31] state that the care relationship is crucial to securing care quality. If 'good care' depends substantially on the quality of the care relationship, then more attention should be paid to the human care workforce. Perhaps this was a principal concern of our sample of care students. Moreover, in covering the instrumental aspects of care, robots may also have been perceived as an occupational threat, making the nurse seem less important (e.g., when the client becomes attached to the robot) or even redundant.
On a societal plane, 70% of Europeans have a positive attitude towards robots in general [32]. However, when it comes to healthcare robots, they are not so positive. Only 22% of Europeans think that robots should be introduced in healthcare [32]. Sixty percent are even resentful of the prospect of robots caring for children and older adults [32]. Perhaps our care students were resonating with a more public trend, vented in the media and discussed at the coffee table.
We did not find great differences between educational levels in the way they perceived care robots. It could be that the educational differences between lower and middle vocational care students are negligible. As an additional exploratory analysis, we combined levels 2 and 3 and compared them with level 4, performing chi-square analysis is (Table 3). We found that level 4 students perceived more beneficence in care robots. However, contrary to our expectations, they do not perceive less maleficence. Higher vocational students also perceived a greater utility in care robots than lower vocational students, however this higher perceived utility does not translate into a higher use intention. All care students expressed low use intentions, the potential utility of robots notwithstanding. As advised by Holloway and Wheeler [33], it may be more worthwhile to turn to qualitative research approaches to resolve matters of change or conflict, particularly in care relations, and explore the behaviours, feelings, and experiences of care students in confrontation with robots on the work floor in more detail.
Trainee care professionals perceived particular types of robots differently in terms of occupational ethics and use. Our research challenge is to ascertain what these care professionals would want from an assistive or monitoring machine that is-in their eyes-potentially very useful and somewhat innocuous. Should a robot assistant have a moral reasoning system that tells it what is 'good care'? Should a monitoring device know what information is private and what should be disclosed to the nurses at the ward? Companion robots may be seen as beneficent, but how can they become more useful? Additionally, how should they behave such that they do not undermine the quality of the care relationship between humans? These are not questions of occupational ethics and utility alone, as their answers will encourage better partnerships between human caretakers and artificial systems of the future. More research is needed, with a more rigorous research methodology to truly obtain the objections raised by (trainee) caregivers to facilitate acceptability. It is stated in the literature [34] that healthcare robots can potentially enhance elderly well-being and decrease the workload on caregivers.

Education could Make a Difference
This study suggests that it would be wise to enrich the curriculum of intermediate care education with practical classes on the use and ethical implications of new care technology, particularly in the