Propensity to trust: comforting touch between trustworthy human and robot partners.

doi:10.21203/rs.3.rs-3738758/v1

Download PDF

Article

Propensity to trust: comforting touch between trustworthy human and robot partners.

https://doi.org/10.21203/rs.3.rs-3738758/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 21 Mar, 2024

Read the published version in Scientific Reports →

You are reading this latest preprint version

Touching a friend to comfort or be comforted is a common prosocial behaviour, firmly based in mutual trust. Emphasising the interactive nature of trust and touch, we suggest that vulnerability, reciprocity and individual differences shape trust and perceptions of touch. We further investigate whether these elements also apply to companion robots. Participants (n = 152) were exposed to four comics depicting human-human or human-robot exchanges. Across conditions, one character was sad, the other initiated touch to comfort them, and the touchee reciprocated the touch. Participants first rated trustworthiness of a certain character (human or robot in a vulnerable or comforting role), then evaluated the two touch phases (initiation and reciprocity) in terms of interaction realism, touch appropriateness and pleasantness, affective state (valence and arousal) attributed to the characters. Results support an interactive account of trust and touch, with humans being equally trustworthy when comforting or showing vulnerability, and reciprocity of touch buffering sadness. Although these phenomena seem unique to humans, propensity to trust technology reduces the gap between how humans and robots are perceived. Two distinct trust systems emerge: one for human interactions and another for social technologies, both necessitating trust as a fundamental prerequisite for meaningful physical contact.

Biological sciences/Psychology/Human behaviour

Biological sciences/Psychology

When we feel vulnerable or distressed, we will often reach out to touch or be touched for comfort. The toucher’s brain activates empathy and reward related regions ^1–4 and the touchee’s distress is reduced at both self-reported and neurophysiological levels ⁵, as touch facilitates emotion regulation ⁶. On the other hand, bodily contact poses inherent risks by reducing self-other boundaries (e.g., COVID-19 social distancing to avoid disease spread). Mutual trust may be a key prerequisite for people to get physically in touch with one another. The COVID-19 pandemic showed us the negative effects of social isolation, particularly on the lack of physical contact with our loved ones ^7,8. It also showed us the dramatic effects of the shortage of medical and health personnel to care for the sick. In this apocalyptic scenario, technologies have attracted particular interest in our society, because of their potential to connect us with others and make basic services such as health and education accessible to geographic areas and population groups that are difficult to reach by local services. Specifically, robots have been looked at as tools with very high potential for support in various functional and affective domains ⁹. When it comes to care robots that can engage in social exchanges, to touch and be touched in affective ways, many questions arise. Under which conditions can we perceive them as trustworthy partners? How does it feel to be touched and comforted by a robot? Are there individual differences with regard to each person's propensity to trust technology?

According to traditional definitions, trust is the willingness to be vulnerable to the actions of another party based on positive expectations about the trustee’s behaviour ¹⁰. Notably, a state of trust is associated with reduced physiological arousal ¹¹ and cognitive monitoring ¹². There are several dimensions of trust. Firstly, people have a general predisposition to trust others ¹³, which we can call propensity to trust. This dispositional aspect of trust is plastic, varying a lot from person to person and also changing across one’s lifespan and with experience. Secondly, situational trust relies on the specific characteristics of the trustee (ability, integrity, benevolence), and is context-dependent ¹⁴. We can more precisely refer to perceived trustworthiness when we delve into the characteristics that make a trustee more or less trustworthy. Some authors suggest that we can further distinguish between cognitive and affective dimensions of trust and perceived trustworthiness ¹⁵. They propose that one’s ability and integrity are strong predictors of cognition-based trustworthiness (expectations grounded in cost-benefit economic reasoning), whereas benevolence and values congruence are stronger predictors of affect-based trustworthiness (built upon positive affective bond between the parties).

It's interesting to note that most studies on trust investigate organisational contexts, work relationships and marketing, which inherently might draw more upon cognitive dimensions of trust. We know much less about the foundations of interpersonal trust and perceived trustworthiness in affective contexts where the sharing of emotions, vulnerability, empathy, and mutual support form the true cornerstone of interactions and relationships. For instance, friends may be perceived as trustworthy because of their benevolence, whereas leaders are trusted because of their integrity¹⁶. Behavioural paradigms for measuring trust and trustworthiness have certainly favoured the cognitive and socio-economic interpretation of trust. In the widely used Trust Game, 2 individuals are assigned the roles of investor and trustee. The investor possesses a designated sum of money and has the option to transfer a portion to the trustee. Upon doing so, the transferred sum is tripled, and the trustee can reciprocate by sending back an amount to the investor. The sum of money the investor chooses to transfer to the trustee serves as an indicator of their willingness to trust the recipient, while the sum returned by the recipient reflects their trustworthiness¹⁷.

Notably, most of the time researchers conceive the trustor as the vulnerable party that depends upon a “strong” trustee. Things in everyday relationships are more complicated than that and we may trust someone not only because they are able to help and comfort us when we are not self-sufficient, but also because they are willing to open up and share their own vulnerabilities. Some authors found that seeing someone in tears can improve trust, resulting in more money given to them as the receiver of a Trust Game ¹⁸. Several socio-affective mechanisms and cues drive interpersonal trust and perceived trustworthiness ¹⁹. A long tradition of studies has focused on the presentation of sequences of neutral and decontextualized faces, suggesting that people automatically judge faces on a trustworthiness dimension ²⁰, with a certain level of agreement among observers, particularly for faces belonging to social in-groups²¹. However, what these approaches overlook is the relational nature of trust and perceived trustworthiness, as well as the crucial role of context. Previous research has especially neglected the role that trust may play in shaping attitudes and behaviours related to social touch. Being willing to trust others and perceiving the partner of a specific interaction as trustworthy may facilitate experiencing a tactile exchange as appropriate and pleasant, especially in situations of emotional vulnerability.

We live in an increasingly technological world, yet we know so little about the cognitive and behavioural mechanisms that shape our interactions with technology. Are we ready to (should we) transfer the mechanisms that govern human interactions to those mediated by technology? In what contexts, under which conditions, and for what purposes can we trust social robots? The very concept of what constitutes a robot can be subject to lengthy debates. Robots are tools designed with specific features and affordances, which determine the ways in which they can be interacted with and their function. In the case of certain types of robots, their primary function is to act as social entities capable of communication and forming relationships. Companion robots, also known as social or assistive robots, are programmed to interact and engage with humans in social settings, providing assistance, companionship, and emotional support (for a review on definitions, see ²²). Companion robots find applications in a wide range of settings, including healthcare and education, especially for older people and children. In healthcare, several types of robots have been integrated in care services ²³, but evidence of effectiveness is often limited ²⁴. In education, companion robots have been utilised to enhance learning experiences²⁵. When a social robot was integrated into a toddler classroom for over five months, the interaction quality between children and the robot improved over time, becoming more and more similar to peer-to-peer relationships²⁶. However, also in this research field and applicative context, there is a lack of solid evidence in support of the specific benefits that robots can bring to children’s lives.

To use companion robots for meaningful interactions, it is crucial to leverage the cognitive and behavioural mechanisms underlying effective social interactions, delving into how these mechanisms operate with technology. For people to be willing to interact with any specific technology, trust is key²⁷. Within the context of the Trust Game, the amount of money transferred by the investor is influenced by the investor's perception of interacting with a human rather than a computer²⁸. A machine’s trustworthiness in terms of ability, benevolence, and integrity increases as the technology is perceived to be more human-like and personified. Conversely, trustworthiness related to system attributes (functionality, usefulness, reliability) hold more relevance for technologies perceived as object-like²⁹. The framework known as Computers as Social Actors (CASA) proposes that machines can possess personalities, prompting people to interact with them as if they were human, even while being aware of their non-human nature³⁰. This goes beyond mere anthropomorphism, as the acceptance and trust in Artificial Intelligence (AI) technologies also require empathy, leading to informative, enjoyable, effortless, and person-centred interactions³¹.

We have discussed how social touch is particularly intriguing as a means for connecting with others, comforting and being comforted. In clinical applications, the possibility to give robots the superpower of social touch is of particular interest to leverage the many therapeutic effects of touch. For instance, pet-like robots that can be interacted with on a tactile level have been found to improve the well-being of people with dementia³² and facilitate anxiety management³³. In education, a social robot has been integrated in a toddler classroom for months, and children's tactile behaviours were particularly informative of the relationship with the robot. Specifically, children touched the robot quantitatively and qualitatively differently from the way they touched a similar inanimate robot or soft toy²⁶. Since social and affective touch can help children in social³⁴ and cognitive³⁵ learning processes, more research is needed to unveil the potential (and limits) of using social robots that can engage in touch interactions with children.

Yet, there is evidence suggesting that touch from a robot is not perceived to be as pleasant as touch from a human, sometimes failing to elicit the same neurophysiological and affective reactions^36,37. We therefore need to better understand which aspects of human-robot social tactile interaction influence its meaning and effects. Firstly, individual differences play a big role in how human-robot touch is perceived. Observers with negative attitudes towards robots, in general, are more likely to perceive a robot touching a human as machine-like rather than human-like³⁸. Moreover, several situational characteristics related to the context in which the interaction takes place and how it feels can modulate social touch perception. For instance, the role each partner plays in terms of sender or receiver is key³⁹. Pet-like robots for therapy have been mainly conceived as touch receivers^32,33. A touch-receiving social robot has been found to reduce pain perception and increase mood in human touchers⁴⁰. In other works social robots were tested as touch senders, with robot touch being associated with reduced physiological stress⁴¹ and increased compliance with a request⁴². Additionally, social touch can be initiated, reciprocated, and become a dynamic, two-way communicative exchange. In a study, participants watched videos featuring a small humanoid robot assisting a human with computer issues. The agents either never touched or engaged in robot-initiated, human-initiated, or reciprocal touch. The robot's proactiveness was manipulated, with it offering help on its own initiative (proactive) or upon request (reactive). Observers perceived the robot as less machine-like when it was proactive and more machine-like when it was reactive, highlighting initiation as a fundamental ability of humans and human-like machines³⁸. However, it is not yet clear whether a social robot is perceived as more or less trustworthy depending on the type of role it assumes (sender or receiver) and the type of touch (human-initiated or robot-initiated, with or without reciprocity). Similarly, individual dispositional characteristics related to attitudes towards social touch have enormous effects on how different users may perceive and use touch with robots⁴³.

On the issue of what situational and dispositional factors influence the effects of human-robot social touch, the concept of trust is finding particular interest due to its dual nature (situational and dispositional) and the plasticity with which it can change with experience and be tuned. Trust in a robot is influenced by many individual human factors and dispositions such as demographics, personality traits, attitudes and implicit cognitive mechanisms⁴⁴. Most research on the situational factors modulating the trust-touch link investigated whether social touch has the power of boosting trust in human-human or human-robot interactions (for a systematic review, see⁴⁵). Provided that the robot-initiated touch is perceived as appropriate (which is not always the case), it has been shown to promote the robot’s perceived trustworthiness⁴⁶. Touch from a robot providing feedback to a person facing a computer issue increased observers' perception of the robot's trustworthiness on both functional and personal dimensions⁴⁷, especially when the robot was more human-like⁴⁸. During 1st -person interaction with a robot providing assistance on a specific task, older adults reported higher trust if the robot-initiated touch compared to no touch conditions, but only if the robot did not make mistakes⁴⁹. In another study, holding hands with a robot while watching a scary video led to higher trust when the robot’s hand was warm compared to cold or in no touch conditions⁵⁰. Overall, social “touch does not increase prosocial behaviours in the absence of meaningful social and psychological connotations”⁵¹. On the other hand, it is rarely investigated whether trust is an essential prerequisite for people to have positive perceptions of human-robot touch and be willing to interact with robots using touch. Our research is situated within this gap in the literature.

The present study

Previous accounts have proposed that touching enhances prosocial behaviours but here we test whether trust promotes positive appraisal of social touch. We ask how people perceive observed interactions between humans or a human and a companion robot that engage in tactile exchanges with a comforting intent. We aim at investigating if people would trust a person or robot when they offer comfort to a vulnerable other, or rather they express vulnerability and receive comfort. We investigate what specific role different dimensions of situational trust (i.e., perceived trustworthiness across the dimensions of ability, benevolence, and integrity) play in people’s perceptions of the different scenarios. Moreover, we explore how the interactions are perceived in terms of realism, appropriateness and pleasantness of the touch, and affective state (valence and arousal) associated with the exchange. Lastly, we question whether people’s perceptions are moderated by individual dispositional characteristics such as propensity to trust other people and technology in general, and attitudes toward interpersonal touch.

We hypothesise that certain factors modulate (and perhaps increase) the perceived trustworthiness of a character. Since social touch is an intimate contact that brings us closer to each other's vulnerabilities, a character might be perceived as trustworthy if they are capable not only of being comforting but also of showing vulnerability themselves. When it comes to interactions that focus on affective content and involve social touch, we expect people to perceive other humans as more trustworthy than robots. However, we wonder whether the ability to express vulnerability would act as a humanising factor and improve the robot’s perceived trustworthiness. We also hypothesise that reciprocity (i.e., observing the touchee touching the toucher back) induces more positive perceptions of the tactile interaction compared to when touch is unidirectional. Whether this is also the case in human-robot interactions will be explored. Lastly, we expect people’s perceptions to be moderated by individual differences, with propensity to trust others being positively associated with perceptions of human-human interactions, propensity to trust technology positively associated with perceptions of human-robot interactions, and touch aversion generally leading to less positive evaluations of observed social touch. Figure 1 depicts the experimental design and theoretical model.

Trustworthiness. The model includes the significant effects of Partner * Role (χ² = 8.5, df = 1, p = .004), Partner * Subscale (χ² = 34.5, df = 2, p < .001), Partner * Propensity to trust others (χ² = 5.5, df = 1, p = .02), Partner * Propensity to trust technology (χ² = 11.6, df = 1, p < .001). Indices of the model goodness of fit are: R² (fixed effects) = 0.23; R² (total) = 0.51. Figure 2 visualises the significant effects predicted by the model.

Interaction

Interaction realism. The model includes the significant effect of Partner * Role (χ² = 21.02, df = 1, p < .001), Propensity to trust others (χ² = 5.86, df = 1, p = .02), Partner * Propensity to trust technology (χ² = 16.26, df = 1, p < .001). Indices of the model goodness of fit are: R² (fixed effects) = 0.31; R² (total) = 0.60. Figure S1 in the Supplementary Information visualises the significant effects predicted by the model. The results suggest that human-to-human interactions are perceived as more realistic than human-robot interactions. The latter is less realistic, especially when the robot expresses vulnerability. Individual differences among participants moderate these effects. In general, social interactions involving comforting touch are perceived as more realistic by those who trust others more. The propensity to trust technology is linked to perceptions of human-robot interactions as more realistic.

Touch appropriateness. The model includes the significant effect of Partner * Role (χ² = 19.96, df = 1, p < .001), Partner * Propensity to trust others (χ² = 4.43, df = 1, p = .04), Partner * Propensity to trust technology (χ² = 29.52, df = 1, p < .001). Indices of the model goodness of fit are: R² (fixed effects) = 0.24; R² (total) = 0.59. Figure 3 visualises the significant effects predicted by the model.

Touch pleasantness. The model includes the significant effect of Partner * Role (χ² = 15.65, df = 1, p < .001), Propensity to trust others (χ² = 3.91, df = 1, p = .05), Partner * Propensity to trust technology (χ² = 39.68, df = 1, p < .001). Indices of the model goodness of fit are: R² (fixed effects) = 0.26; R² (total) = 0.62. Figure S2 in Supplementary Information visualises the significant effects predicted by the model. Results show that human-to-human comforting touch is perceived as more pleasant than human-robot touch. The latter is less pleasant, especially when the robot expresses vulnerability. Individuals’ propensity to trust others is associated with increased pleasantness. Moreover, individuals’ propensity to trust technology is associated with increased perception of robot touch as pleasant, thus reducing the gap between humans and robots.

Characters’ affective state. Since these data are derived from spatial coordinates dependent on the size of the viewing window on each participant's screen, we have retained the data from participants who maintained this window at a constant size throughout the task. Six participants whose screen dimensions changed during the task (e.g., they resized the experiment platform window) were excluded from these analyses. Moreover, 4 participants did not provide a valid response to these questions (i.e., did not click on a point inside the EmojiGrid). Therefore, these analyses are based on n_participants = 142 and n_observations = 1115. Figure 4 visualises the significant effects predicted by the models on valence and arousal.

As for valence, the model includes the significant effect of Partner * Touch phase (χ² = 7.41, df = 1, p = .006). Indices of the model goodness of fit are: R² (fixed effects) = 0.08; R² (total) = 0.53. As for arousal, the model includes the significant effects of Partner * Role (χ² = 9.27, df = 1, p = .002), touch phase (χ² = 13.99, df = 1, p < .001), Partner * Propensity to trust others (χ² = 17.72, df = 1, p < .001), touch aversion (χ² = 5.06, df = 1, p = .02). Indices of the model goodness of fit are: R² (fixed effects) = 0.07; R² (total) = 0.55.

In this study, adult participants from around the world observed and evaluated scenes of social interactions between two humans or a human and a robot, described as peer familiar relationships. In different experimental conditions, one of the characters expressed emotional vulnerability by saying, "I am sad," and the other performed a comforting gesture by touching their arm. In response to this gesture, the character who was touched reciprocated the touch. Participants were asked to assess how trustworthy the character providing or receiving comfort was in terms of ability, benevolence, and integrity. Additionally, observers rated the realism of the interaction, the appropriateness and pleasantness of the touch, the valence and arousal attributed to characters', distinguishing between phases where touch was initiated and reciprocated. We support the idea that trust is an interpersonal bond amplified by our paradigm which uses a social touch exchange to emphasise the interactive nature of the scene. We show that trust promotes positive appraisal of social touch, with the experimental manipulations of the social scenario, mediated by observers’ propensity to trust, resulting in differences in perceived trustworthiness, perceptions of the interaction and associated affective states. In addition, we shed light on the limitations of applying these concepts to companion robots. Nevertheless, propensity to trust is a subjective and potentially plastic trait that can be leveraged to facilitate acceptance of technologies through positive experience. Showing that if we trust, then social touch will be perceived as more appropriate and pleasant, we take a complementary perspective to previous studies that have investigated the reverse relationship (if we touch, we trust).

First and foremost, we did not find differences in the perceived trustworthiness of individuals based on their role in the interaction. People perceive as equally trustworthy someone who comforts another in a moment of vulnerability and someone who expresses their own vulnerability. This finding significantly expands our understanding of trust, which has been mainly conceived as a one-way perception and behaviour from the trustor to the trustee^10,17. Trust is rather an interpersonal, interactive mechanism built upon the willingness to share one's vulnerabilities with the other. Observing how these mechanisms operate in human-robot interactions allows us to understand whether they are more or less specific to human interactions, or rather fundamental principles that can be leveraged to build trust in technologies designed for social presence. We found that the robot is overall perceived as less trustworthy than human interaction partners, especially when it expresses vulnerability. This suggests that people may not desire to interact with a social robot that, like a human being, can express vulnerability and receive comfort. Symmetry in human relationships among peers is fundamental for various social processes, including perspective taking and empathy. Instead, we should perhaps consider robots as partners of asymmetric, more unidirectional relationships, where they need to possess specific social skills to provide emotional support to humans. This clearly imposes limitations on the social relationship with a robot and raises important questions about the foundations of human-robot trust, the design and implementation of companion robots. According to previous literature, robots are perceived as less reliable if designed in a more anthropomorphic way⁵². Anthropomorphism of a robot has been found to be implicitly associated with lower agency and capacity to sense and feel compared to humans⁵³, potentially because of the mismatch between affordances (what I expect the robot to do given its appearance and features) and actual performance. Indeed, our results indicate that, when comforting one another, humans are perceived as trustworthy especially for their abilities to provide support and assistance to another person, whereas robots are perceived as less skilled for social exchanges. Some promising alternatives for robots that can receive touch and comfort are pet robots, which can be used in healthcare to promote patients’ well-being^32,33,54. Notably though, perceived trustworthiness of an agent in a specific situation is influenced by observers’ dispositional attitudes, such as their general propensity to trust. Our data suggest that there are two somewhat distinct systems for trusting other people or technology, which specifically come into play in these two different types of (social) interactions.

Secondly, we see that our manipulations of the social scenario, mediated by propensity to trust, results in differences in how the interaction was perceived, with trust promoting positive appraisal of social touch. In human-to-human scenarios, propensity to trust others is positively associated with perceived character’s trustworthiness, interaction realism, touch appropriateness and pleasantness. In human-robot scenarios, ratings of realism, appropriateness and pleasantness are lower. This is especially evident when the robot assumes the vulnerable role. Nevertheless, individual propensity to trust technology reduces the gap between humans and robots. These insights offer a new perspective in the study of the link between touch and trust, where researchers have primarily investigated the role of social touch in promoting and facilitating interpersonal trust, whether mediated or not by technology (see Valori et al., in press, for a systematic review). Here we look at the other side of this presumably two-way interaction. We propose that trust is a prerequisite for positively perceiving tactile social interactions and that there are two somewhat distinct systems for trusting other people or technology, which specifically influence these two different types of (social) interactions. Additionally, propensity to trust is a subjective and plastic trait with the potential to influence acceptance of technologies through positive experience. It can be hypothesised that with the advancement and widespread use of technology in everyday life, people's overall trust in technologies is likely also to increase. If trust is moderated by familiarity with specific tools^55,56, we may have to wait for companion robots to appear more regularly in our daily contexts to understand whether future humans will be more inclined to trust and interact with them in affective ways. Studies on the development of trust in children show that familiarity is particularly important in novice learners, and that with increasing social experience, discrimination, e.g., of more or less trustworthy informants, is refined to be increasingly driven by the other’s competence, also when it is a robot⁵⁷. Therefore, trust towards others and robots is plastic and understanding individual differences can aid in personalising robot touch behaviours to optimise interactions.

Lastly, we investigated which affective states are associated with the different social scenarios, particularly in terms of valence and arousal, which are key dimensions for understanding social touch^58,59. In our paradigm, social touch is used to amplify the interactive nature of a peer-to-peer comforting exchange. We see that reciprocity of touch influences the affective experience, alleviating feelings of sadness (as shown by less negative valence and reduced arousal). Observers with higher propensity to trust others also attributed less arousal to the characters in the human-to-human scenarios. The power of reciprocal touch and trust is lessened in human-robot interactions, where we see less arousal, especially when the robot assumes the vulnerable role. Previous research found that interpersonal touch is more arousing than object-based touch, suggesting that human-to-human touch is experienced as more intense⁵⁹, and the robot in our study may have been perceived as an object more than a social partner. Such human-robot interaction is therefore perceived as less realistic, appropriate, pleasant, and less emotionally meaningful. We also found that observers with higher aversion towards social touch perceived the scenarios as overall less arousing. If higher touch aversion is associated with higher vigilance to observed social touch (as suggested by the neural responses found by⁶⁰), we could expect the opposite relation between touch aversion and arousal. On the other hand, it is possible that less touchy-feely people are simply less activated by scenarios of vicarious touch, without necessarily showing discomfort or hyper-vigilance. Indeed, valence does not appear to be influenced by individuals’ touch aversion in our data.

It is worth mentioning that this study has some limitations, which open the doors to future research. We focused on the perception of observed social tactile interactions between two humans or a human and a robot. To safeguard the simplicity of experimental design and statistical models, we did not include a control condition in which the interaction did not involve touch. Moreover, we used static pictures instead of animations to avoid confounding aspects such as touch velocity. Comforting touch has well-known optimal velocity ranges in human-to-human interactions⁵. Robots can also be programmed to execute movements with spatio-temporal patterns designed to represent different emotions (e.g., in⁶¹). However, the movements of real robots are still far from the smoothness of human ones, and creating animations in which a robot moves to touch as a human would move can lead us into an uncanny valley. To deepen the role of social touch in human-robot interactions, future studies might not only compare touch and no-touch conditions, but also explore different types of touch. Different combinations of physical parameters of touch, such as velocity, intensity, duration, and contact areas result in different gestures (e.g., stroking, holding, shaking, tapping, squeezing) that convey different emotional meanings, from sadness, to joy, gratitude, and love⁶². It is crucial to disentangle the importance of the robot being able to understand and communicate through touch. To become a socially intelligent partner a robot must be able to capture and classify human touch and respond to this in an appropriate manner, interpreting not only tactile features but also contextual factors⁶³. At the same time, the robot could also be able to touch the human in an affective way, and produce tactile gestures that the human can understand⁶⁴.

The present study is based on an observational task in which participants are exposed to images of social interactions that include touch. Although the participants play the role of simple observers of scenes taking place between two characters, literature suggests that the mere observation of others' touch leads to pleasantness ratings⁶⁵ and brain activity similar to those associated with a first-person experience of touch (e.g., as ⁶⁶ found with monkeys). Therefore, the participants' evaluations of the proposed stimuli can be interpreted as an indicator of how they would perceive the social situation themselves. Nonetheless, future studies would need to conduct lab-based experiments whereby participants interact with robots. This possibility is challenged by the limited skills and capacity for actual interactivity that robots have at the present time especially with regards to exchanges involving social touch^32,63. In terms of the possibilities this set-up would open up, among the most fascinating is surely the integration of neural, physiological, and kinematic measurements to characterise human cognition, perception, and action during social interactions with robots.

Although there has been significant progress in creating more advanced and socially adept robots in recent years, there are concerns that the field is entering a winter phase of disillusionment⁶⁷. Researchers are putting a lot of resources into enhancing the naturalness and authenticity of robot behaviours (e.g., designing robots to display emotions and responses that are as realistic as possible), with the idea that this will foster more genuine and meaningful interactions with humans. For instance, robots are being programmed to recognize touch gestures⁶⁸ and to perform touches with optimal sensorimotor features to be perceived as pleasant and non-intrusive^50,69. However, touch is a communicative signal that takes on various nuances, uses, and interpretations depending on the context and the person giving or receiving it⁷⁰. Our society has yet to establish new social norms for digital social touch, through a dialogue between what is technologically feasible and what is truly desired by and beneficial for the human in the loop^71,72. It is crucial that we understand under which conditions and in what contexts human-robot interactions can benefit from social touch. To address this, it is essential to clearly define the neurocognitive processes that underpin human-robot interactions, employing neuroscience and psychophysiology techniques to uncover the genuine capabilities and limits of social robots⁷³.

In conclusion, perceiving other individuals as trustworthy is crucial in affective exchanges that involve social touch, where barriers between the self and the other are reduced, we share vulnerabilities, offer closeness and comfort. Here we provide evidence that trust is an interpersonal, interactive tango rather than the one-way mechanism from trustor to trustee that has been studied in previous literature. We also show that trust promotes positive appraisal of social touch, offering a complementary perspective to studies that have shown the reverse effect of touch as a trust booster. Looking into the future, we see our lives increasingly intertwined with those of technologies such as robots, which are not only tools but also partners in social exchanges. Yet, we still do not know what social norms apply to these new interactions. The present findings show potential limits to the social power of trust and touch in human-robot interactions, suggesting, however, that leveraging individuals' positive attitudes and trust towards technology can reduce the distance between humans and robots. This will help to shed light on crucial challenges in robot design that we humans could potentially perceive as partners to trust and touch.

Participants

Eligible participants were older than 18 years of age and fluent in English. From the a priori power analysis (see Statistical approach section for details), we aimed at n = 152. We collected data from 153 participants. One participant has been excluded from analyses because they used a mobile device rather than a personal computer, which was required to participate. The final sample is n = 152 (nFemales = 77, nMales = 76; age range = 19:67; meanAge = 29.04; sdAge = 8.81). Despite being given the option to select and specify non-binary gender identities, all participants identified with a female or male gender. Participants come from 33 nationalities across Europe, Africa, America, and Asia. With the intention of representing the general adult population, we did not establish exclusion criteria on the basis of medical or psychological conditions. Self-reported medical or psychological conditions included anxiety and/or depression (n = 9), neurodevelopmental conditions (n = 4) such as ASD (Autism Spectrum Disorder) and/or ADHD (Attention Deficit and Hyperactivity Disorder), medical conditions (n = 5). Participants reported minimal previous experience with robots (mean = .68 on a 0 “none” to 4 “a lot” Likert scale). When given the chance to briefly describe such previous experience, only 13 participants mentioned brief, occasional interaction with robots we could call social. These qualitative data are reported in the Supplementary Information.

Procedure

Participants were recruited via Prolific and compensated 9,63 £ average reward per hour for a median completion time of 12 minutes. Participants were given the chance to read the study general goal, procedure and methods before signing up and being redirected to the Gorilla experimental platform (www.gorilla.sc), where they provided written consent to participate. The experiment consisted of one task and a series of questionnaires, which were created and hosted using Gorilla Task Builder 2 and Questionnaire Builder 2. The study received ethical approval from the Ethics Committee at the Technische Universität Dresden and was carried out in accordance with the approved guidelines and regulations.

Task

The study is based on an observational task in which participants are exposed to images of social interactions that include touch. On commencing the experimental task, participants were given the following introductory information:

In the next screens you will see a series of comics representing everyday interactions between friends living together (two humans or a companion robot and a human). The robot and the human have a friendly relationship and share their daily life. The robot knows and can move around the home environment, engage in joint activities and has communication skills to allow for conversations with the human. Similarly, the human is familiar with the robot and interacts with them on a daily basis.

They were then presented with pictures of 2 female human characters (called Anna and Sarah) and a humanoid robot (called Pepper). Across 4 trials, they were presented with scenes depicting the two humans (H) or a human and a robot (R) sitting one in front of the other at the table of a living room. In a 2x2 design, comics were created by a combination of 2 factors:

Partner: there was a human interacting with either another human (H) or the robot (R)
Role: participants were asked questions about how they perceived the one character that was either vulnerable (V) or comforting the other (C).

Moreover, each scene consists of 3 segments: in the first picture one of the characters expresses emotional vulnerability (i.e., says “I am sad”) (V). In the second picture the other character initiates touch by placing their hand on the other’s arm in a comforting manner (C). In the last picture, the receiver reciprocates the touch (Fig. 1). Trials were presented in 2 blocks (H, R). We randomised blocks presentation order, and trials presentation order within each block.

After the presentation of each scene, participants rated the observed social interaction reporting their agreement (on a 7-point Likert scale ranging from 1 - strongly agree - to 7 - strongly disagree) with statements about the:

Character’s trustworthiness (Trust questionnaire, adapted from¹⁴): 9 items that capture participants’ trust in a certain character (highlighted in Fig. 1 through a yellow circle) considering 3 aspects of trust (i.e., integrity, benevolence, ability). As for the H conditions, participants were always asked about Anna’s trustworthiness (the blond female character in Fig. 1), who was either vulnerable or comforting the other human. As for the R conditions, questions referred to Pepper’s trustworthiness, that was either vulnerable or comforting the human. Table S1 of the Supplementary Information reports the adaptation of the trust questionnaire to assess trustworthiness of either the human or robot character. Three scores are calculated summing up the responses to items grouped by subscale. Higher scores indicate higher trustworthiness of the character.

Afterward, participants were sequentially shown again the first (initiation) and second (reciprocity) touch phases of the current scene. For each phase, they reported their agreement (on a 7-point Likert scale ranging from 1 - strongly agree - to 7 - strongly disagree) with statements about the observed interaction:

Interaction Realism: 1 item “The interaction was realistic”
Touch Appropriateness: 1 item “Touch was appropriate”
Touch Pleasantness: 1 item “Touch was pleasant”

Moreover, they rated the:

Characters’ affective state by clicking on a point of an EmojiGrid⁵⁹ that best represented how the person(s) in the picture felt. The EmojiGrid is a square grid labelled with emoticons expressing different levels of emotional valence (e.g., sad vs. smiling face) on the x axis and arousal (e.g., sleepy vs. excited face) on the y axis. Participants clicked on a single point inside the grid, which represents the combination of valence and arousal they attribute to the scene displayed.

Questionnaires

At the end of the task, participants filled out a series of questionnaires about themselves.

The Social Touch Questionnaire (STQ) is a 20-item scale that measures participants’ aversion towards social situations involving touch⁷⁴. On a 5-point Likert scale ranging from 0 (not at all) to 4 (extremely), participants indicate how characteristic or true each statement is of them. A total STQ score is calculated summing up the responses to all items after reversing those that express positive attitudes towards touch (e.g., “I generally like when people express their affection towards me in a physical way”). Higher scores indicate a participant’s dislike for social touch.

The Propensity to trust scale is a 4-item scale that measures individuals’ dispositional trust in other people⁷⁵. On a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree), participants express to what extent they agree with statements like “I usually trust people until they give me a reason not to trust them”. A total score is calculated summing up the responses to all items. Higher scores indicate higher propensity to trust other people.

The Propensity to trust technology scale is a 3-item scale that measures individuals’ dispositional trust towards technology in general (Trusting Stance—General Technology²⁷). On a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree), participants report to what extent they agree with statements like “My typical approach is to trust new technologies until they prove to me that I shouldn’t trust them”. A total score is calculated summing up the responses to all items. Higher scores indicate higher propensity to trust technology.

They also reported their previous experience with robots with 1 item on a 5-point Likert scale from 1 (none) to 5 (a lot): “How much experience have you had with robots?”. An optional open question gave them the possibility to briefly describe such previous experiences with robots.

Statistical approach

Sample size specification. To establish the sample size, we run a priori power analysis (using GPower 3.1) on the main effects of interest. From previous literature, we can expect a main effect size of touch on trust towards a robot to be around Cohen’s d = .23⁴⁶. Robot-related characteristics have been found to be moderately associated with trust in human-robot interaction, with r̄ = +0.24 (according to a meta-analysis from¹⁹). Individual (e.g., gender) differences on touch pleasantness ratings previously showed effect sizes around g = .25 (according to a meta-analysis from⁷⁶). We therefore run a power analysis for F tests, with repeated measures and within-subjects design, effect size f = .115 (conversion from Cohen’s d = .23); alpha error probability = .05; power = .80, 4 conditions (2 Partners * 2 Roles) and 2 measurements (Touch phases) per condition, resulting in a required sample size of n = 152.

Variables of interest. Below is a description of the variables included in the statistical models. Descriptive statistics (means, standard deviations) of the Dependent Variables (DVs) by relevant experimental conditions are reported in the Supplementary Information.

Dependent variables (DVs - continuous variables from self-reported perceptions of the task stimuli): characters’ Trustworthiness, interaction Realism, touch Appropriateness, touch Pleasantness, characters’ affective state in terms of Valence and Arousal.
Independent variables (IVs − 2-level categorical factors representing the experimental conditions): Partner (human vs robot), Role (vulnerable vs comforting), Touch phase (initiation vs reciprocity).
Moderators (self-reports filled out at the end of the task, which are hypothesised to moderate the effect of the IVs on the DVs): Propensity to trust others, Propensity to trust technology.
Covariate (self-report filled out at the end of the task, which is hypothesised to have a direct main effect on the DVs): Touch aversion (total STQ).
Control variables: Gender (female vs male - no participants reported non-binary gender identities), Participant (random effect of individual variability that accounts for the repeated measure, within-subjects design of the experiment).

Pre-processing. As for the EmojiGrid data, if participants clicked more than once, the last click was considered the definitive answer for analysis (as per the instructions displayed upon presentation of the EmojiGrid). Each response is encoded by coordinates on the x-axis (valence) and y-axis (arousal), which are then analysed separately (as in⁵⁹). Since the size of the participants’ screens varies, the coordinates were normalised by dividing the x-value by the width of each participant's grid size and the y-value by its height. Because of the way the grid is positioned on the Gorilla Task Builder screen, raw values on the x-axis (valence) range from 0 (left) to the maximum (right). On the other hand, the raw values on the y-axis range from the maximum (bottom) to zero (top), and have therefore been reversed. Moreover, the EmojiGrid scale is conceptualised as a matrix where the neutral affective state (namely, the “true” 0,0 position) is located in the centre of the grid. Therefore, we rescaled the normalised response coordinates so that both valence and arousal range from − 50 to + 50. Thus, in our statistical analyses, higher values of valence and arousal indicate greater valence and arousal. Negative values on the valence dimension indicate responses on the left side of the EmojiGrid, and negative values on the arousal dimension indicate responses on the bottom side of the grid.

Generalised linear mixed-effects models. Statistical analyses have been run with R, version 4.3.0. Generalised linear mixed-effects models were employed to account for the repeated measures design of the experiment (i.e., trials nested within participants, which has been included as a random effect in the analyses). We specified the research hypotheses on the link between each dependent variable and the predictors of interest as statistical models. Analysis of deviance (Type III Wald chi-square test, r package ‘car’⁷⁷) was used for assessing the effect of individual predictors and interactions included in the models. As an index of goodness of prediction, conditional R2 (the ratio of variance explained by fixed and random effects over total variance) and marginal R2 (the ratio of variance explained by fixed effects over total variance) were calculated to quantify the variance explained by the whole model (including the contribution of individual variability) or the fixed effects only (excluding the contribution of individual variability)⁷⁸. Higher percentages of explained variance indicate a stronger association between the dependent variable and the predictors, with the model making better predictions.

Model specification

Trustworthiness. With a generalised linear mixed effect model, we tested how perceived trustworthiness was predicted by the 2-way interaction between Partner (Human or Robot) and Role (Comforting or Vulnerable), including touch aversion as a covariate, controlling for gender and individual variability. Moreover, the model tested whether the effect of Partner was moderated by individuals’ propensity to trust others and technology, and the subscales of the trustworthiness measure (ability, benevolence, integrity). Below, the formula is reported.

Trustworthiness ~ Partner * Role + Partner * Propensity to trust others + Partner * Propensity to trust technology + Partner * TrustSubscale + Touch aversion + Gender + (1|Id)

Interaction. With several generalised linear mixed effect models, we tested how each DV (interaction realism, touch appropriateness, touch pleasantness, valence, arousal) was predicted by the 2-way interactions between Partner and Role, and Partner and touch phase (initiation or reciprocity), controlling for gender and individual variability. Moreover, the model tested whether the effect of Partner was moderated by individuals’ propensity to trust others and technology, and whether the DV was covarying with the individuals’ touch aversion. Below, the formula is reported.

DV ~ Partner * Role + Partner * Touch_phase + Partner * Propensity to trust others + Partner * Propensity to trust technology + Touch aversion + Gender + (1|id)

Acknowledgements

Funded by the German Research Foundation (DFG, Deutsche Forschungsgemeinschaft) as part of Germany’s Excellence Strategy – EXC 2050/1 – Project ID 390696704 – Cluster of Excellence “Centre for Tactile Internet with Human-in-the-Loop” (CeTI) of Technische Universität Dresden. The authors acknowledge the financial support by the Federal Ministry of Education and Research of Germany in the programme of “Souverän. Digital. Vernetzt.”. Joint project 6G-life, project identification number: 16KISK001K.

We thank Wenhan Sun for his support in handling the EmojiGrid data.

Informed consent

Informedwas obtained from all subjects and/or their legal guardian(s).

Competing interests

The authors declare no competing interests.

Data availability

The original dataset and analysis script are available from the OSF public repository at the following URL: https://osf.io/7bf2m/?view_only=b2c801bfbc4d4467b461a22c492dac31

Author contributions

I.V., M.M.J, and M.T.F. conceived the experiment, Y.F. designed the stimuli, I.V. implemented the experiment, collected and analysed the results, I.V., M.M.J, and M.T.F. interpreted the data and wrote the manuscript. All authors reviewed and edited the manuscript and approved the submitted version.

Corresponding author

Correspondence should be addressed to I.V.

Goldstein, P., Weissman-Fogel, I., Dumas, G. & Shamay-Tsoory, S. G. Brain-to-brain coupling during handholding is associated with pain reduction. Proc. Natl. Acad. Sci. 115, E2528–E2537 (2018).
Peled-Avron, L., Goldstein, P., Yellinek, S., Weissman-Fogel, I. & Shamay-Tsoory, S. Empathy during consoling touch is modulated by mu-rhythm: An EEG study. Neuropsychologia 116, 68–74 (2018).
Masten, C. L., Morelli, S. A. & Eisenberger, N. I. An fMRI investigation of empathy for ‘social pain’ and subsequent prosocial behavior. NeuroImage 55, 381–388 (2011).
Mathur, V. A., Harada, T., Lipke, T. & Chiao, J. Y. Neural basis of extraordinary empathy and altruistic motivation. NeuroImage 51, 1468–1475 (2010).
Shamay-Tsoory, S. G. & Eisenberger, N. I. Getting in touch: A neural model of comforting touch. Neurosci. Biobehav. Rev. 130, 263–273 (2021).
Korisky, A., Eisenberger, N. I., Nevat, M., Weissman-Fogel, I. & Shamay-Tsoory, S. G. A dual-brain approach for understanding the neuralmechanisms that underlie the comforting effects of social touch. Cortex 127, 333–346 (2020).
von Mohr, M., Kirsch, L. P. & Fotopoulou, A. Social touch deprivation during COVID-19: effects on psychological wellbeing and craving interpersonal touch. R. Soc. Open Sci. 8, 210287 (2021).
Meijer, L. L. et al. Affective touch perception and longing for touch during the COVID-19 pandemic. Sci. Rep. 12, 3887 (2022).
Shen, Y. et al. Robots Under COVID-19 Pandemic: A Comprehensive Survey. IEEE Access 9, 1590–1615 (2021).
Mayer, R. C., Davis, J. H. & Schoorman, F. D. An Integrative Model of Organizational Trust. Acad. Manage. Rev. 20, 709–734 (1995).
Merrill, N. & Cheshire, C. Trust Your Heart: Assessing Cooperation and Trust with Biosignals in Computer-Mediated Interactions. in Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing 2–12 (Association for Computing Machinery, 2017). doi:10.1145/2998181.2998286.
Perello-March, J. R., Burns, C. G., Woodman, R., Elliott, M. T. & Birrell, S. A. Using fNIRS to Verify Trust in Highly Automated Driving. IEEE Trans. Intell. Transp. Syst. 24, 739–751 (2023).
Patent, V. & Searle, R. H. Qualitative meta-analysis of propensity to trust measurement. J. Trust Res. 9, 136–163 (2019).
Gefen, D. Reflections on the dimensions of trust and trustworthiness among online consumers. ACM SIGMIS Database DATABASE Adv. Inf. Syst. 33, 38–53 (2002).
Tomlinson, E. C., Schnackenberg, A. K., Dawley, D. & Ash, S. R. Revisiting the trustworthiness–trust relationship: Exploring the differential predictors of cognition- and affect-based trust. J. Organ. Behav. 41, 535–550 (2020).
Moore, A. K., Lewis, J., Levine, E. E. & Schweitzer, M. E. Benevolent friends and high integrity leaders: How preferences for benevolence and integrity change across relationships. Organ. Behav. Hum. Decis. Process. 177, 104252 (2023).
Camerer, C. F. Strategizing in the brain. Science 300, 1673–1675 (2003).
Reed, L. I., Matari, Y., Wu, M. & Janaswamy, R. Emotional Tears: An Honest Signal of Trustworthiness Increasing Prosocial Behavior? Evol. Psychol. 17, 1474704919872421 (2019).
Hancock, P. A. et al. How and why humans trust: A meta-analysis and elaborated model. Front. Psychol. 14, 1081086 (2023).
Todorov, A., Said, C. P., Engell, A. D. & Oosterhof, N. N. Understanding evaluation of faces on social dimensions. Trends Cogn. Sci. 12, 455–460 (2008).
Silvestri, V., Arioli, M., Baccolo, E. & Macchi Cassia, V. Sensitivity to trustworthiness cues in own-and other-race faces: The role of spatial frequency information. Plos One 17, e0272256 (2022).
Rogge, A. Defining, Designing and Distinguishing Artificial Companions: A Systematic Literature Review. Int. J. Soc. Robot. 15, 1557–1579 (2023).
Kyrarini, M. et al. A Survey of Robots in Healthcare. Technologies 9, 8 (2021).
Broekens, J., Heerink, M. & Rosendal, H. Assistive social robots in elderly care: a review. Gerontechnology 8, 94–103 (2009).
Brink, K. A. & Wellman, H. M. Robot teachers for children? Young children trust robots depending on their perceived accuracy and agency. Dev. Psychol. 56, 1268–1277 (2020).
Tanaka, F., Cicourel, A. & Movellan, J. R. Socialization between toddlers and robots at an early childhood education center. Proc. Natl. Acad. Sci. 104, 17954–17958 (2007).
Mcknight, D. H., Carter, M., Thatcher, J. B. & Clay, P. F. Trust in a specific technology: An investigation of its components and measures. ACM Trans. Manag. Inf. Syst. TMIS 2, 1–25 (2011).
Johnson, N. D. & Mislin, A. A. Trust games: A meta-analysis. J. Econ. Psychol. 32, 865–889 (2011).
Lankton, N. K., McKnight, D. H. & Tripp, J. Technology, humanness, and trust: Rethinking trust in technology. J. Assoc. Inf. Syst. 16, 1 (2015).
Nass, C., Moon, Y., Fogg, B. J., Reeves, B. & Dryer, D. C. Can computer personalities be human personalities? Int. J. Hum.-Comput. Stud. 43, 223–239 (1995).
Pelau, C., Dabija, D.-C. & Ene, I. What makes an AI device human-like? The role of interaction quality, empathy and perceived psychological anthropomorphic characteristics in the acceptance of artificial intelligence in the service industry. Comput. Hum. Behav. 122, 106855 (2021).
Jung, M. M., van der Leij, L. & Kelders, S. M. An Exploration of the Benefits of an Animallike Robot Companion with More Advanced Touch Interaction Capabilities for Dementia Care. Front. ICT 4, (2017).
Sefidgar, Y. S. et al. Design and Evaluation of a Touch-Centered Calming Interaction with a Social Robot. IEEE Trans. Affect. Comput. 7, 108–121 (2016).
Della Longa, L., Gliga, T. & Farroni, T. Tune to touch: Affective touch enhances learning of face identity in 4-month-old infants. Dev. Cogn. Neurosci. 35, 42–46 (2019).
Farroni, T., Della Longa, L. & Valori, I. The self-regulatory affective touch: a speculative framework for the development of executive functioning. Curr. Opin. Behav. Sci. 43, 167–173 (2022).
Willemse, C. J., Toet, A. & Van Erp, J. B. Affective and behavioral responses to robot-initiated social touch: toward understanding the opportunities and limitations of physical contact in human–robot interaction. Front. ICT 4, 12 (2017).
Willemse, C. J. A. M., Huisman, G., Jung, M. M., van Erp, J. B. F. & Heylen, D. K. J. Observing Touch from Video: The Influence of Social Cues on Pleasantness Perceptions. in Haptics: Perception, Devices, Control, and Applications (eds. Bello, F., Kajimoto, H. & Visell, Y.) 196–205 (Springer International Publishing, 2016). doi:10.1007/978-3-319-42324-1_20.
Cramer, H. S. M., Kemper, N. A., Amin, A. & Evers, V. The effects of robot touch and proactive behaviour on perceptions of human-robot interactions. in Proceedings of the 4th ACM/IEEE international conference on Human robot interaction 275–276 (Association for Computing Machinery, 2009). doi:10.1145/1514095.1514173.
Fairhurst, M. T., McGlone, F. & Croy, I. Affective touch: a communication channel for social exchange. Curr. Opin. Behav. Sci. 43, 54–61 (2022).
Geva, N., Uzefovsky, F. & Levy-Tzedek, S. Touching the social robot PARO reduces pain perception and salivary oxytocin levels. Sci. Rep. 10, 9814 (2020).
Willemse, C. J. A. M. & van Erp, J. B. F. Social Touch in Human–Robot Interaction: Robot-Initiated Touches can Induce Positive Responses without Extensive Prior Bonding. Int. J. Soc. Robot. 11, 285–304 (2019).
Hoffmann, L. & Krämer, N. C. The persuasive power of robot touch. Behavioral and evaluative consequences of non-functional touch from a robot. Plos One 16, e0249554 (2021).
Erp, J. B. F. V. & Toet, A. How to Touch Humans: Guidelines for Social Agents and Robots That Can Touch. in 2013 Humaine Association Conference on Affective Computing and Intelligent Interaction 780–785 (2013). doi:10.1109/ACII.2013.145.
Sanders, T. L. Individual Differences in Trust Towards Robotic Assistants. Dr. Thesis Univ. Cent. Fla. Orlando Fla. (2016).
Valori, I., Jung, M. & Fairhurst, M. Social touch to build trust: a systematic review of technology-mediated and unmediated interactions. Comput. Hum. Behav. (In press).
Law, T., Malle, B. F. & Scheutz, M. A touching connection: how observing robotic touch can affect human trust in a robot. Int. J. Soc. Robot. 1–17 (2021).
Arnold, T. & Scheutz, M. Observing robot touch in context: How does touch and attitude affect perceptions of a robot’s social qualities? in 352–360 (2018).
Mazursky, A., DeVoe, M. & Sebo, S. Physical Touch from a Robot Caregiver: Examining Factors that Shape Patient Experience. in 2022 31st IEEE International Conference on Robot and Human Interactive Communication (RO-MAN) 1578–1585 (2022). doi:10.1109/RO-MAN53752.2022.9900549.
Giorgi, I. et al. Friendly but faulty: A pilot study on the perceived trust of older adults in a social robot. IEEE Access 10, 92084–92096 (2022).
Nie, J., Pak, M., Marin, A. L. & Sundar, S. S. Can you hold my hand? physical warmth in human-robot interaction. in Proceedings of the seventh annual ACM/IEEE international conference on Human-Robot Interaction 201–202 (Association for Computing Machinery, 2012). doi:10.1145/2157689.2157755.
Rosenberger, L. A., Ree, A., Eisenegger, C. & Sailer, U. Slow touch targeting CT-fibres does not increase prosocial behaviour in economic laboratory tasks. Sci. Rep. 8, 7700 (2018).
Roesler, E., Onnasch, L. & Majer, J. I. The effect of anthropomorphism and failure comprehensibility on human-robot trust. in vol. 64 107–111 (SAGE Publications Sage CA: Los Angeles, CA, 2020).
Li, Z., Terfurth, L., Woller, J. P. & Wiese, E. Mind the Machines: Applying Implicit Measures of Mind Perception to Social Robotics. in 2022 17th ACM/IEEE International Conference on Human-Robot Interaction (HRI) 236–245 (2022). doi:10.1109/HRI53351.2022.9889356.
Bucci, P. et al. Sketching CuddleBits: Coupled Prototyping of Body and Behaviour for an Affective Robot Pet. in Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems 3681–3692 (Association for Computing Machinery, 2017). doi:10.1145/3025453.3025774.
Komiak, S. Y. X. & Benbasat, I. The Effects of Personalization and Familiarity on Trust and Adoption of Recommendation Agents. MIS Q. 30, 941–960 (2006).
Idemudia, E. & Raisinghani, M. The Influence of Cognitive Trust and Familiarity on Adoption and Continued Use of Smartphones: An Empirical Analysis. J. Int. Technol. Inf. Manag. 23, (2014).
Baumann, A.-E., Goldman, E. J., Meltzer, A. & Poulin-Dubois, D. People Do Not Always Know Best: Preschoolers’ Trust in Social Robots. J. Cogn. Dev. 24, 535–562 (2023).
Lee Masson, H. & Op de Beeck, H. Socio-affective touch expression database. PloS One 13, e0190921 (2018).
Toet, A. & Erp, J. B. F. van. The EmojiGrid as a rating tool for the affective appraisal of touch. PLOS ONE 15, e0237873 (2020).
Peled-Avron, L. & Shamay-Tsoory, S. G. Don’t touch me! autistic traits modulate early and late ERP components during visual perception of social touch. Autism Res. 10, 1141–1154 (2017).
Suguitan, M., Gomez, R. & Hoffman, G. MoveAE: modifying affective robot movements using classifying variational autoencoders. in 481–489 (2020).
McIntyre, S. et al. The language of social touch is intuitive and quantifiable. Psychol. Sci. 33, 1477–1494 (2022).
Jung, M. M., Poel, M., Reidsma, D. & Heylen, D. K. J. A First Step toward the Automatic Understanding of Social Touch for Naturalistic Human–Robot Interaction. Front. ICT 4, (2017).
Teyssier, M., Bailly, G., Pelachaud, C. & Lecolinet, E. Conveying Emotions Through Device-Initiated Touch. IEEE Trans. Affect. Comput. 13, 1477–1488 (2022).
Haggarty, C. J., Makdani, A. & McGlone, F. Affective Touch: Psychophysics, Physiology and Vicarious Touch Perception. in Somatosensory Research Methods (ed. Holmes, N. P.) 109–128 (Springer US, 2023). doi:10.1007/978-1-0716-3068-6_6.
Sharma, S., Fiave, P. A. & Nelissen, K. Functional MRI Responses to Passive, Active, and Observed Touch in Somatosensory and Insular Cortices of the Macaque Monkey. J. Neurosci. 38, 3689–3707 (2018).
Henschel, A., Hortensius, R. & Cross, E. S. Social Cognition in the Age of Human–Robot Interaction. Trends Neurosci. 43, 373–384 (2020).
Jung, M. M., Poel, M., Poppe, R. & Heylen, D. K. J. Automatic recognition of touch gestures in the corpus of social touch. J. Multimodal User Interfaces 11, 81–96 (2017).
Triscoli, C., Olausson, H., Sailer, U., Ignell, H. & Croy, I. CT-optimized skin stroking delivered by hand or robot is comparable. Front. Behav. Neurosci. 7, (2013).
Saarinen, A., Harjunen, V., Jasinskaja-Lahti, I., Jääskeläinen, I. P. & Ravaja, N. Social touch experience in different contexts: A review. Neurosci. Biobehav. Rev. 131, 360–372 (2021).
Fairhurst, M. T. & Valori, I. A functional framework for multisensory and interactive mediated social touch experiences. IMX ’23 Proc. 2023 ACM Int. Conf. Interact. Media Exp. (2023).
Jewitt, C. et al. Manifesto for digital social touch in crisis. Front. Comput. Sci. 3, 754050 (2021).
Wiese, E., Metta, G. & Wykowska, A. Robots As Intentional Agents: Using Neuroscientific Methods to Make Robots Appear More Social. Front. Psychol. 8, (2017).
Wilhelm, F. H., Kochar, A. S., Roth, W. T. & Gross, J. J. Social anxiety and response to touch: incongruence between self-evaluative and physiological reactions. Biol. Psychol. 58, 181–202 (2001).
Frazier, M. L., Johnson, P. D. & Fainshmidt, S. Development and validation of a propensity to trust scale. J. Trust Res. 3, 76–97 (2013).
Russo, V., Ottaviani, C. & Spitoni, G. F. Affective touch: A meta-analysis on sex differences. Neurosci. Biobehav. Rev. 108, 445–452 (2020).
Fox, J. & Weisberg, S. An R Companion to Applied Regression. (Sage, 2019).
Nakagawa, S. & Schielzeth, H. A general and simple method for obtaining R2 from generalized linear mixed-effects models. Methods Ecol. Evol. 4, 133–142 (2013).

No competing interests reported.

SupplementaryInformation.pdf

Download PDF

Journal Publication

published 21 Mar, 2024

Read the published version in Scientific Reports →

Editorial decision: Revision requested
03 Mar, 2024
Reviews received at journal
11 Jan, 2024
Reviewers agreed at journal
03 Jan, 2024
Reviewers invited by journal
28 Dec, 2023
Editor assigned by journal
21 Dec, 2023
Editor invited by journal
13 Dec, 2023
Submission checks completed at journal
13 Dec, 2023
First submitted to journal
11 Dec, 2023

You are reading this latest preprint version

Propensity to trust: comforting touch between trustworthy human and robot partners.

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

The present study

Results

Interaction

Discussion

Methods

Participants

Procedure

Task

Questionnaires

Statistical approach

Model specification

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1