Human-robot interaction: the impact of robotic aesthetics on anticipated human trust

Introduction

In a world where robotics is becoming more prominent, our ability to trust them has never been so important. With a robot’s physical appearance drastically influencing our perceptions of trust, a greater awareness of how design elements and their aesthetic effect may trigger what affective processes are imperative. Robots have an exceptional potential to benefit humans within a team, yet a lack of trust in the robot could result in underutilising or not using the robot at all (Floyd, Drinkwater & Aha, 2014). As Barnes & Jentsch (2010) identified, the key to a successful relationship between man and machines is in how well they can work and adapt to each other. This can develop through the form and structure of the robot that in turn helps establish social expectations. In addition, a robot’s morphology can have an effect on its accessibility and desirability (Fong, Nourbakhsh & Dautenhahn, 2003). The research presented in this paper explores how robot aesthetics can heighten participants ability to trust robots. Participants were introduced to an array of robot visualisations (face and chest) and asked to note their impressions towards each visualisation and whether or not they trusted the robot. This enabled the researchers to investigate how design elements and their combined aesthetic arrangement can act as emotional stimuli influencing the ability to trust each robot. In detail, by using various design elements (i.e., colour, blurriness, and tone), we were interested in better understanding how we design for the fundamental principles of aesthetic order in human–robotic interaction. We anticipate that uncertainties in and between the visualisations will greatly influence a participant’s willingness to accept the robot (i.e., The cohesion of messages, positive and balanced stimuli, non-invasive colours, etc.). This paper highlights not only the impact of risks and uncertainties created by the visualisations on the human–robot interaction but also the potential of robot aesthetics to commence a trusting relationship.

Literature Review

Materials and Methods

This study was conducted at Cardiff Metropolitan University from the 31st of March 2020 to the 15th of April 2020 and was designed to capture the perception of participants feelings and attitudes towards trusting robots. The study was conducted using the powerful online survey software: Qualtrics. Participants were selected through stratified random sampling to target both participants with past robotic experience and those without. Through distributing the questionnaire on social media, special robotic interest groups, and online forums, the authors were able to obtain participants from a diverse participant pool. A total of seventy-four participants from the age of 16 plus years (50 female & 24 male) completed the study from a varied demographic. Participants resided globally (i.e., Europe, Africa, Asia, Australia, North America, and South America) and captured an assortment of participants. The questionnaire took approximately thirty minutes in duration. All graphics were generated using Adobe Photoshop, and the study and questions asked had a strong aesthetic visual component.

The study mainly consisted of quantitative questions in order to provide summaries through descriptive statistics. Additionally, an assortment of questions required participants to engage in qualitative questions, which then enabled analysis to enrich interpretations and uncover similarities. The questions were separated into two categories to target both the general acceptance of robotics and specific questions relating to the Canbot U03 robot. In order to not influence a participant’s feelings and past experience with robots, the Canbot U03 was not shown during the first block of questions. Participants were provided with a brief definition of trust at the start of the questionnaire “To Believe that someone is good and honest and will not harm you, or that something is safe and reliable” (Cambridge Dictionary, 2021, trust entry).

Once participants had concluded the initial preparatory questions, they were introduced to the opening visual of the Canbot U03 robot. Participants were presented to a Canbot U03 (see Fig. 1) with no visual modification and asked whether or not they would trust this robot based on its visual appearance (i.e., only based on the design features). To address the concepts of a participant’s ability to trust the Canbot U03 robot, participants were asked to envisage situations in their everyday life where they may encounter a robot. A short list of possible situations and jobs roles were provided to participants (i.e., Teacher, doctor, receptions, bus driver, etc.).”

Participants were then presented with a series of questions with different aesthetic modifications throughout. The first modifications came with the Canbot U03 robot presented with a series of cartoon facial expressions portraying different emotions. Participants were prompted to identify the robot’s emotion and whether or not they felt the robot was more or less trusting than before. To detail, questions such as the following were asked to participants: How trustworthy is this robot’s appearance?, What emotion do you think the robot is feeling?, Does this visual change affect your ability to trust the robot?, How does the robot make you feel with this appearance? Participants were also asked to provide descriptions on the following questions: What characteristics do you believe only robots should have? How do you design a robot that people would trust?

The following block of questions prompted participants to consider the anthropomorphic characteristics of the robot (see Fig. 2). Participants were introduced to a series of robots that related to having human features; these questions probed participants for their feelings towards these powerful visual modifications.

The next section of questions was related to how the design element colour influenced the participant’s opinions and description of the robot. This required participants to associate words (i.e., dangerous, happiest, trusting, unpredictable, and unrealistic) with an array of Canbot U03 robots with different colours hues. Participants were presented with eight robot visualisations (see Fig. 3), all with varying colour hues (i.e., Pink, orange, blue, yellow, etc.) and prompted to associate the expressive wording with an individual Canbot, no Canbot or all Canbots.

Participants were also introduced to a range of visualisations with contrasting images such as conflicting facial expressions and chest screen imagery (i.e., Happy facial expression + Danger symbol on the chest). Participants were asked a series of questions such including: Which Canbot would you describe as most uncertain?, What impact did the cohesion of screens have on your decision? and Does the facial expression overrule the icon on the chest screen when considering the Canbot’s emotions? These questions aimed to understand how the level of cohesion between the chest and facial screens can influence a participant’s willingness to trust the robot.

Finally, to further probe the concept of risk, participants were presented with mathematical problems that would be too complex for human calculation (i.e., 887 × 974 & 997 × 1,066). Participants were then be asked to identify which Canbot (A–H) displayed the correct solution upon their chest screen. This question required participants to determine the answer they deemed correct based solely on trusting the robot’s physical appearance. Optional text boxes were provided throughout the questions to allow participants to expand and express opinions on the robot’s appearances.

The Ethics Board at Cardiff Metropolitan University approved the study (CST_2020_Staff_0002), and participants involved were all provided and signed an online consent form to participate in the study and for the academic use of the non-identifiable data.

Results

The observations indicate that a participant’s willingness to trust a robot was heavily impacted by the aesthetic elements they were exposed to, and whether or not the participant had past experiences with robots. When asked about Fig. 1, fifty per cent of participants said they would trust this robot, twenty-eight per cent were unsure, and the remaining twenty-two per cent recorded that they would not trust the robot. Interestingly, anthropomorphism did not encourage more to trust the robot. Figure 2 (Robot B) shows how the introduction of the face impacted participants who first trusted the robot, twenty of the thirty-seven (fifty-four per cent) of participants who first trusted were now non-trusting or uncertain about trusting the robot. However, anthropomorphism did positively influence those unsure of trusting the first robot introduced, with fifty-two per cent changing their opinion from ‘unsure’ to ’yes’ to trust (see Fig. 4). In the human-like visualisations, it seemed participants had different opinions on how robots should be designed for trust. One participant (P72) said, ’Less human-like as this makes them feel more deceptive’ while another described human features as ’creepy’ and ’People may become intimidated by implementing human behaviours into a machine’.

When probed further into how designing for trust, participants said, ‘Give them their own personality that isn’t based on human expression ‘and that ’human features make the model ‘creepy’. One participant notes that the introduction of a realistic human face ‘makes people uneasy’. When adding human eyes to the robot visualisation (see Fig. 2) participants were asked their feelings on the realistic eyes. 80 per cent of participants expressed their dislike of this appearance, making them feel ‘confused, scared, worried and surprised’. One participant noted ‘the need for distinction between human and robot’ and ‘the inclusion of human likeness may be intimidating’.

When asked Would you trust this robot? and What do you think this robot feels?

In the blurry face visualisations (Fig. 5), it appeared participants were more apprehensive about trusting the robot. The findings show that half of the participants were able to correctly identify the robot’s emotional cue as ‘happy’ despite the introduction of blurriness. In contrast, the other half of the participants were torn between ‘confused, angry, uncertain, uneasy, and uncomfortable’ for the robot’s emotion. The introduction of the dissimilar stimuli of the happy facial expression and the blurriness presented participants with uncertainties through the contrasting messages each present (i.e., Happy face – trust, blurriness – uncertain). The results were clearer when prompting participants away from identifying which emotion the robot depicted to how these changes made them feel. The participants concerns were expressed when asked about how the Canbot made them feel, with the majority of responses including terms such as uncertain, uneasy, and confused.

In addition, the findings show different impressions towards facial features when faced with the decision to trust (i.e., What robot is providing you with the correct information?). Interestingly sixty-six per cent of participants selected robot B (Fig. 6) as the most trusting, despite the introduction of a hybrid robot (Robot D - Fig. 6). Robot five was the next most accepted (fifteen per cent), yet on closer inspection the participant’s speed to answer this question was significantly higher (fifty percent increase) than other responses, thus indicating the potential use of a calculator to determine the correct answer to the equation.

Similar results were seen in Fig. 7, with the alterations to the chest screen affording uncertainty to trust the robot to provide the correct answer. We asked participants to indicate which of the six robots posed the correct answer to the 997*1066 by clicking on the chosen robot. Figure 7 displays the frequency distribution of clicks over the six distinct robot images. Sixty-four per cent of participants selected robot B (the robot with limited visual modifications) as most trusting despite providing incorrect information. Interestingly, robot B presented the incorrect answer to the mathematical equation.

Moreover, participants felt that in order for a robot to be trusting, there is a need for ‘a screen that clearly shows the message that is being transmitted’ and that ‘I would expect the screen display to match with any expressions’. In terms of harmony between face and chest screen, one participant highlighted that ‘It would be difficult to trust a robot with a face and another image within the robot screen. I would trust better with just one option.’ In particular, when exposed to Fig. 8, participants felt that the facial expressions produced a contradicting message to the one upon the chest screen. With sixty per cent of participants declaring the robot as untrustworthy and a further thirty-eight per cent unsure whether or not to trust the robot. One participant could not trust the robot as ‘I could not take anything this bot says seriously with that expression’. This highlights the true impact of the misaligned messages on participants ability to trust.

Discussion

In this study, we investigated the impact of the aesthetic order of facial and chest visualisations on participants willingness to trust robots. In particular, it considered the potential risks and uncertainty afforded by certain aesthetic orders to the human–robotic trusting relationship. Our results show the clear influence that past experience had on a participant’s willingness to trust the original robot. Particularly, the visualisation with no modifications was found to have a substantially higher percentage of trust in those with past experience. Participants with no past experiences were relying solely on the visual appearance to determine their level of trust. These findings are in line with what Sanders et al. (2017) hypothesised and discovered, in detail, how those participants with past robotics experience would lead to a higher trust of robots and a better positive attitude towards them.

Interestingly, we found that a blurred facial expression significantly influenced whether participants trusted a robot. The blurred facial and chest screen visualisations afforded uncertainty and resulted in a participant’s unwillingness to trust a robot.

Figure 7 displayed the extent that physical appearance had on the decision participants made to trust a robot. With a participant’s ability to roughly estimate the correct answer not largely adopted by participants, we can only conclude that the physical appearance was the determining factor in the decisions. Interestingly, over half the participants selected robot B, which presented the incorrect answer to the mathematical equation.

Based on previous research that shows colours can influence various moods (Kurt & Osueke, 2014), we predicted similarly that the aesthetic element colour could initiate different affective responses when applied to a robot. We tested that hypothesis by introducing participants to an array of robot visualisations that applied an assortment of distinct colour changes. We found that comparably participants were following known psychology of colour associations when selecting what feelings and terms they associated to the robots with the assortment of colours. For example, Fig. 9 displays the words participants associated with the array of colours and other visual modifications. As we hypothesized, certain colours had followed the known associations of related words, such as when participants were promoted to associate the red coloured robot to a particular word. Following the commonly known western culture word associations with the colour red (i.e., dangerous, excitement, festive, etc.) (Cousins, 2012), we evaluated its affect while present on a robots outer shell and found a similar result of red being associated with the term dangerous.

However, it is important to consider how cultural beliefs and geographical regions may also have an influence on a person’s perceptions of colour. A particular colour hue can have multiple meanings and interpretations to people in different regions of the world (Kurt & Osueke, 2014). It is critical that when designing a robot to afford trust that these cultural backgrounds, geographical location, and beliefs are carefully considered when selecting a robot’s hue to be fit for purpose. Additionally, it is important that this same level of consideration is taken for other design elements, in order to evaluate how the different designs are perceived in different regions, backgrounds, and faiths.

The research has also highlighted the importance of cohesion between the facial screen and chest screen. In the question prompting participants to consider the information on the chest screen (see Fig. 7), the participants were never asked whether or not they trusted the robot as a whole, only if they trusted the information on the screen. However, the negative stimuli released by the facial expression demonstrated that most participants declared the robot as not trustworthy. Moving forward, when designing a robot that can be trusted, it is important to consider all elements, as stimuli from other visual outputs can potentially influence an independent communication channel.

Conclusion and future work

This research has shown that robots have the unique ability to create an emotional connection with humans through the use of facial expressions and aesthetics. As documented, we have seen the introduction of anthropomorphism which creates a fine line between increasing trustworthiness and becoming ‘scary’. Nevertheless, the non-physical humanlike anthropomorphic designs (cartoon designs) encouraged participants to trust the robots further, showing the unique ability to improve their facilities for empathy. Moreover, this research has shown that the face is not the sole visual aesthetic that can be utilised to initiate affective processes. The chest screen provides an additional entity to further enrich the potential to provide an engaging experience. Ultimately, the cohesion between the multiple screens is an important consideration for designing socially acceptable robots. As is the design elements and principles to understand how their aesthetic order can play such an important role in initiating a trusting robotic experience.

Going forth, we feel there may be interest in replicating the study but utilising actual robots. We acknowledge there is still a substantial amount of research required to fully understand how we form trusting relationships between human and robot. However, we feel this study paves the way for future studies that involve aesthetic physicalisation, where further sensory cues can be tested to evaluate their influence on our trusting ability of robots. Additionally, this research touched upon how design elements may influence different participants from different cultural backgrounds, geographical locations, and beliefs. We feel it would be of interest to further explore the potential to develop culturally appropriate robots.

Moreover, it would be interesting to further expand on the use of aesthetic designs to evaluate how further modifications (i.e., different colour tones, design elements, design principles, etc.) can affect and in some cases, increase a participant’s willingness to trust a robot.

Finally, we believe there would be value in understanding how the trusting relationship between human and robot may develop over time. Whilst this study provides details on the initial engagement/interaction, there may be interest to explore aesthetic designs in different situations and time scales.

Throughout this research, we have explored how we can build trusting relationships with robotics through aesthetic designs. In future work, better consideration of human-centered design perspectives must be explored when considering building trust. The research explores participants not trusting robotics as injudicious when the reason not to trust is still a valid and acceptable response in certain situations.

Supplemental Information