Most studies of clinical applications of robots have focused on health outcomes for the patient (see [1]) or on robot acceptance by patients (see [17]). These studies are mainly concerned with the interaction between the patient and the robot, and only very few studies have investigated the effects the robot has on the patient’s interactions with other people.

Several studies have looked at the social effects of PARO, a robot with the appearance of a baby seal that is responsive to touch, sound, temperature, and posture. By serving as the focus of the interaction between people (participants in these studies often interact in pairs or small groups with the robot), PARO was shown to have positive effects on social life such as increasing the density of social networks in a care home for the elderly [8], increasing social activity [18], and increasing social engagement of elderly nursing home residents with varying levels of dementia [19]. PARO was also shown to increase the reported quality of the interaction among people from a nonclinical population [20]. A couple of studies with robots used in therapy for children with autism have also been shown to have effects on the interaction between the child and others. In 1 study, the robot played a mediator role between the therapist and the child by allowing the child to express positive emotion in playing with the therapist [21]. In another study, the robot successfully served as a focus of free play between 2 children [22].

In nonclinical contexts, the social mediation effect of robots has been studied, among others, for the following purposes: conflict resolution with children [23] or adults [24,25], active listening [26], teaching EI through interactive storytelling [27], and enhancing cooperation [28]. Although these results represent a promising start, more research is needed to better understand both the ways in which robots can support social interactions among people and how they might inadvertently influence human-human interactions in perhaps undesirable ways. In clinical contexts, different aspects of human-human interactions, for example, how patients are perceived by others, can have important implications for health care.

A number of studies have documented the effects of a patient’s affect or perceived personality traits on care decisions; in short, doctors appear willing to prescribe more care for more positive or likable patients. Despite only ranking emotional state as an important consideration in decisions about Intensive Care Unit admission 6% of the time, doctors were almost 3 times more likely to recommend admission of a hypothetical patient if given a vignette that described the patient as upbeat and courageous rather than sad and discouraged [15]. Similarly, another set of doctors recommended more follow-up visits and calls for simulated likable and competent patients and more staff time spent educating simulated likable patients [16].

Other research has indicated that doctors have more positive feelings toward patients who appear happy [29,30] or toward those who express both positive and negative affect over the course of a visit [31]. On the contrary, some primary care physicians report that their challenging patients often become favorites over time and that favorite patients likely benefit from the extra effort that physicians feel inclined to spend on these patients’ care [32].

Doctors are not the only ones who are influenced in their behavior toward patients by the patient’s perceived attributes. A study found that participants (who were not specifically medical professionals) tend to dislike patients who appear distraught but are slightly more willing to aid patients displaying negative affect than those displaying positive affect, offering the least aid to those who show little affect [33].

Given how consequential people’s perceptions of patient psychological attributes are to the patients receiving optimal care, roboticists should be mindful of how they design robots so as not to negatively affect relationships between patients and health care providers or other people. This could also be seen as an opportunity for health care robotics: Social robots could be used to enhance the perception of positive attributes and promote good relationships among patients and doctors, caregivers, or others.

How robots influence relationships in the context of health care, whether in a desirable or undesirable way, will of course depend on their design. More specifically, it will depend on what approach the robot will take for providing care to the patient, and what social capabilities the robot will have.

Patient-centered care [34] is an influential approach to health care that has formulated desirable features for the relationship of the patient with others. This approach emphasizes respect for the patient’s preferences, information, education and communication with the patient, coordination of care, emotional support for the patient, physical comfort, involvement of family, continuity and transition, and access to care [35]. It has been proposed that success in providing patient-centered care may depend on social capabilities such as EI [36].

EI has been linked to positive social relationships in multiple contexts, including health care. EI comprises abilities such as perceiving, understanding, and managing one’s own emotions and the emotions of others. EI has been linked to benefits for both the person possessing these abilities, such as enhanced job performance and stress management [13,37] and better educational outcomes [38], and also for the social group that one is embedded in: EI has been associated with improved teamwork and conflict resolution [39], higher leadership ratings [40], and successful social interactions [41].

In the health care context, EI in medicine has been linked to positive doctor-patient relationships, increased empathy, better teamwork and communication skills, stress management, and organizational commitment and leadership [14]. EI concepts are also central to nursing practice, with implications for nursing students’ learning, ethical decision making, critical thinking, evidence, and knowledge use in practice (for a review see [42]). We thus proposed that EI is needed in social robots that are to operate in the health care setting. To our knowledge, only a couple of studies have so far investigated EI in robots [43,44] and have found that people do expect and detect differences in EI in robot agents and that these differences influence how much robots are trusted [43]. It is thus important to understand people’s perceptions of robots’ EI in the context of patient care.

Patient care, however, is not the only context in which robots’ influence on relationships between people can have an impact on health and well-being. In the domain of public health, robots could be meaningfully used to mitigate problems such as epidemic obesity and to enhance the relationship between clients and dieticians or weight loss support groups. In the field of education, robots could help promote individualized learning plans and student-centered approaches as well as enhance relationships between students and teachers. In the industry, robots could optimize mentor-trainee interactions and assist in retraining people for new jobs, which has been increasingly needed as technological advances are reshaping the labor market. We have focused the first step of this study on how robots affect relationships in the health care setting, and then we broadened the scope of our inquiry in the second step of the study to other contexts that are more indirectly linked to people’s health and well-being.

The aims of this study were to investigate the effects of robot behavior in the health care context and to probe whether similar effects extend to other contexts relevant for well-being and quality of life. In the first step of the study, we investigated whether the robot behavior, patient-centered or task-centered, has an impact on (a) how the robot is perceived, in terms of EI, trustworthiness, and acceptability and (b) the impression that people form of the patient assisted by the robot. We hypothesized that (1) a robot that acts in a patient-centered way will be perceived as having higher EI, (2) inspired by the findings of [43], a robot that acts in a patient-centered way will be trusted more, and (3) accepted more, and we also hypothesized that (4) by acting in a patient-centered way and respecting the patient’s agency, the robot will cause others to think more highly of the patient.

In the second step of the study we further investigated the extent to which the effects observed in the health care context can be replicated and extended to other contexts relevant for people’s health and well-being, contexts in which social robots can provide assistance by monitoring, keeping record, and informing a professional about the user’s progress.

The experiment used a between-group design and a text vignette methodology (see the Materials section). For the first step of the study, investigating the effects of robot behavior in the health care context, we conducted a series of one-way analyses of variance (ANOVA). Our dependent variables were perceptions of robot EI, trust in the robot, robot acceptance, and impression of the patient cared for by the robot (see the Measures section). Our main independent variable was the condition based on which the vignette described the robot’s behavior as being either patient-centered or task-centered (see the Materials section for further details). We also verified for effects of gender as an independent variable and age as a covariate in a series of analyses covariance (ANCOVA).

For the second step of the study, investigating the effects of robot behavior in other contexts relevant for well-being, we used a 2 × 3 between-group experimental design. In addition to the 2 conditions (person-centered or task-centered robot behavior), we also designed vignettes that varied in context (weight loss or learning or job training), which we used as an additional independent variable. As for the health care context, scores from our 4 different questionnaires were used as dependent variables. Gender and participant age were added as variables in the models in a further step.

Our participants were recruited on the Amazon Mechanical Turk (AMT) platform. This made it possible to reach participants from more diverse ethnic backgrounds and spanning a wider age range than what is typical for in-laboratory studies. To ensure language comprehension and a similar compensation incentive, we only recruited participants based in the United States who were fluent in English.

For the first step of the study, in which we investigated the effect of robot behavior in the health care context, a total of 199 participants completed the experiment through the AMT. A total of 11 participants failed to pass our attention checks and were excluded from the analyses. According to standard practice for AMT studies, our attention check consisted of a reading comprehension question about the topic of the vignette meant to assess whether participants read it attentively. The 188 participants with usable data ranged in age from 20 to 72 (mean 34.58, SD 10.04) years, 77 were female, and 2 identified as Other. The ethnic composition of the sample was white or Caucasian (74.5%, n=140), Asian (5.8%, n=11), African American (9%, n=17), Hispanic (6.9%, n=13), and Other (3.7%, n=7). In total, 95 participants saw the person-centered vignette and 93 participants saw the task-centered one.

The second step of the study, in which we investigated the effects of robot behavior in other contexts related to well-being, was completed by 299 participants, out of which 254 passed our attention checks, and thus provided usable data. Participants’ ages ranged from 20 to 71 (mean 36.48, SD 10.54) years, and 111 of the participants were female and 5 of the participants identified as Other. The ethnic composition of the sample was as follows: white or Caucasian (79.1%, n=201), Asian (3.1%, n=8), African American (7.8%, n=20), Hispanic (6.7%, n=17), and Other (3.1%, n=8). A total of 93 participants saw the weight loss vignette, 92 saw the learning vignette, and 69 saw the job training vignette. The procedure was identical to that in step 1 but with more vignettes. Each participant saw one vignette only, either in the person-centered or task-centered form.

For this study, we used the text vignette methodology to evaluate people’s perceptions, attitudes, and impressions toward assistive robots, their behavior, and the patients they care for. In text vignettes, hypothetical situations are described to which participants are asked to respond. This is a common methodology in psychology and sociology experiments and has been used successfully in human-robot interaction research (eg, [43]) as an initial step of investigating and informing possible robot design choices.

For the first step of the study that focuses on robot behavior in the health care context we used a text vignette featuring 3 characters: a health care provider, a patient, and an assistive robot. The scenario starts with the health care provider noticing that the patient is using an assistive robot for treatment monitoring and asks permission to get a report from the robot about treatment progress, to which the patient responds affirmatively. In one condition, the robot gives a patient-centered report that focuses on the needs and choices of the patient with regard to the treatment plan (patient-centered condition). In the other condition, the robot gives a task-centered report focused on how faithfully the treatment plan was adhered to (task-centered condition; see Figure 1). The treatment plan refers to a medication schedule and a physical exercise routine.

The vignette was presented in a video format. The dialogue was parsed into chunks usually containing one dialogue turn or, in the case of the robot, which speaks for longer into sentences that conveyed one idea. Each chunk was displayed on a blank screen. The text appeared across the screen as if it was being typed (using the typewriter animation effect) with a speed similar to what it would take to say the words out loud. This gave the impression of a dynamic conversation. The speaker was clearly indicated at the beginning of each dialogue turn or chunk of text (eg, “Robot: The patient...”).

The vignette characters were referred to as health care provider, patient, and robot, and no personal characteristics were specified about the human characters (eg, no gender was suggested). In addition, no description of the robot’s capabilities or appearance (eg, whether it was humanoid or not) was provided; rather, this was left to the participants’ imagination. From the vignette, it could be inferred that the robot had the capacity to track treatment progress and produce a spoken report.

For the second step of the study in which we sought to generalize our findings to other contexts, 3 additional vignettes (with 2 conditions each) were used, modeled closely after the vignette in step 1 in terms of (a) structure (ie, the same number of characters and the same conversation progression were kept), (b) theme (ie, the robot monitored progress with regard to a body-oriented task such as dieting and exercising, dancing, and training for a physically active job), (c) conditions (ie, an assistant robot gave a report in a person-centered or task-centered manner), (d) display (ie, the same procedures for chunking and displaying text were used), and (e) amount of information conveyed (ie, no personal characteristics of the humans and no description of the robot were given). The vignettes differed in the context described and the characters involved.

In the first vignette (the dieting vignette), the robot assistant was used for monitoring weight loss progress and the interaction happened between a client, a weight loss coach, and the assistant robot. The weight loss plan involved a meal plan and a physical exercise routine. The second vignette (the learning vignette) featured a robot assistant that was used to keep track of progress in learning how to dance. The interaction took place between a student, a dance instructor, and the assistant robot. The learning goals involved a dance practice schedule and a strength and conditioning routine. Finally, in the third vignette (training vignette), a robot assistant was used for monitoring training progress for volunteer firefighters. The interaction involved a volunteer firefighter, a lieutenant, and the assistant robot. The training goals involved a physical exercise training plan and Search and Rescue drill training. Given the diversity of contexts and characters in this second step, we have renamed the patient-centered condition from step 1 as the person-centered condition, person here referring to the weight loss client, the student, or the volunteer firefighter.

Participants accessed the experiment through the AMT. After reading and agreeing to the consent form, participants filled out a demographic questionnaire (age, gender, and ethnicity). They then followed instructions for watching the video. Participants were randomly assigned to one condition: person-centered or task-centered. After watching the video, participants completed the measures in the following order: perception of robot’s EI, trust in the robot, robot acceptance, and impressions of the patient. After concluding the experiment, participants were able to collect their US $1 compensation. All procedures were reviewed and approved by the Institutional Review Board at our university.

This is the first study, to our knowledge, to investigate the influence of robots’ behavior on the impressions that people form about the person the robot is assisting. We found that people perceived the robot as having higher EI when it was behaving in a person-centric way and in some contexts as being more trustworthy and having a higher acceptance rate. Most importantly, robots were able to influence people’s impressions of patients, coaching clients, students, and volunteers by modeling behavior that was respectful of these people’s agency and choices with regard to medical treatment, weight loss plans, learning, and job training. The most immediate implications for human-robot interaction research are that (1) it is important to be aware of the social assistive robots’ influence on the relationships between the people assisted and others and (2) the ability to positively impact human relationships opens up new exciting opportunities for the use of robots in pursuing health and well-being.