Article summary of Designing persuasive robots: how robots might persuade people using vocal and nonverbal cues by Chidambaram et. al. - Chapter

What is this article about?

Many researchers think that robots can function as social actors that can improve our motivation and compliance in the areas of health, education and wellbeing. The success of motivating people will rely in a big way on the robot’s ability to persuade. Many people are probably wondering how a robot can persuade individuals and how these types of robots are built.

Research has shown that there are a couple of behavioural attributes that shape an individual’s nonverbal immediacy. This is the degree of perceived psychological and physical closeness between people. Researchers have identified people’s nonverbal immediacy as a huge factor in the persuasion of others. These behaviours include bodily cues, like proximity, gesture, gaze, facial expression, posture, touching and vocal cues. There have also been studies on the role of nonverbal cues during human-robot interaction, but the persuasive ability of the robot in relation to these cues has not been studied. There are only a few studies that have explored how robots can be made into persuasive agents. However, how behavioural attributes of a robot might improve the robot’s persuasiveness has not been studied. In this study, the researchers looked at how manipulations in bodily cues, like proximity, gaze, vocal cues and gestures affect the persuasiveness of a humanlike robot. The researchers designed a set of nonverbal behaviours for a humanlike robot and looked at how manipulations of these behaviours affect the persuasiveness of the robot.

Persuasion can be viewed as an attempt to change or shape behaviours, feelings or thoughts about an issue. It also enables social influence, cooperation and attitude change. The persuasive ability of an individual is associated with a number of factors, like the verbal and nonverbal behaviours of the person, social interaction and psychological factors (personality). Research has also identified a couple of nonverbal cues that shape nonverbal immediacy. The nonverbal immediacy is associated with a person’s persuasive ability, attractiveness and likeability. These nonverbal cues are (as mentioned previously) proximity, posture, gaze, gestures, facial expressions and vocal behaviours. Nonverbal immediacy and individual nonverbal cues affect persuasion. One research showed that 7% of people’s perception of a person is shaped by verbal cues, 38% by the tone of their voice and 55% by the bodily cues. In public presentations, the amount of eye contact the speaker maintains with the audience affects the persuasiveness of his or her message and also affects the degree of compliance from the audience. Closer proximity leads to a higher compliance. More mimicry of the nonverbal behaviour of others also leads to more liking of the individual and the perception of that person’s persuasiveness. Studies that have looked at teacher- student immediacy show that nonverbal immediacy has a positive effect on the teacher’s perceived competence, trustworthiness and caring.

Nonverbal immediacy also has an effect on student motivation, participation and cognitive learning. Some robotics researchers have developed a couple of applications to deliver positive benefits in health management, education and energy conservation. These applications draw on social behaviour. This study will look at how robot design might shape persuasiveness.

Drawing on the findings mentioned above, the authors of the article had three hypotheses. The first one is that participants will find the robots more persuasive and will comply more with its suggestions when the robot displays nonverbal cues compared to when it does not display nonverbal cues to communicate with the participant. The second hypothesis is that the participants will find the robot to be more persuasive and will comply more with its suggestions when the robot displays only bodily cues compared to when it displays only vocal cues. The third and last hypothesis is that women’s perception of the persuasion of the robot and compliance with its suggestions will be higher compared to those those of men in the presence of nonverbal cues.

What is the method of this study?

The writers combined findings from studies on nonverbal behaviour.

• Gaze. They found that gaze cues communicate the social accessibility of an individual. When you look directly at someone, it suggests that you want to start a conversation with that person. The robot was designed to look toward the participant when it spoke and to direct its gaze toward a computer screen when it referred to the items on a certain task the participants had to do (this will be explained later).
• Proximity. Although studies have shown that people’s persuasive ability increases with closer proximity, entering ones personal space (between two and four feet of that person) could evoke feelings of discomfort. Researchers had to find the appropriate balance and they designed the robot’s proximity to vary across a space boundary. When the robot maintained a close proximity, it stood closer to the person than when it maintained a distant proximity.
• Gestures. The authors have found four types of gestures that are common across conversations. They integrated each type of these gestures in the robot’s behaviour. Iconic gestures are the first type. They are associated with the semantic content of speech. So when the robot says ‘you should drink water’, it brings its right hand towards its face. The second type are metaphoric gestures. Metaphoric gestures depict abstract concepts such as knowledge or an idea. When the robot said ‘Do you know that alcohol absorbs water?’ it raised its hands in front of its chest with both hands facing to each other. Deictic gestures are used to direct attention toward concrete entities in the physical environment. This is usually done by pointing the finger. When the robot talked about ‘I’, it pointed toward itself. The last type is the beat gesture. These are beats described as simple up-and-down movements. When the robot explained complex ideas, it used rhythmic up-and-down gestures.
• Vocal cues. These can be vocal rate, pitch and loudness. The pitch of the robot was manipulated, as well its vocal tone. This created two versions of each utterance. These are identical in verbal content, but one of the utterances was monotonic, while the other was highly expressive.

The researchers designed four designs: (a) no nonverbal cues, (b) vocal cues only, (c) bodily cues only and (d) bodily and vocal cues. The authors wanted to see how these manipulations affected participants’ compliance with the robot’s suggestions and what effect it had on their perception of the robot’s persuasiveness. In the no vocal cues condition, the robot talked in a flat tone during its entire speech. The researchers manipulated proximity to be either in or out the participant’s personal space. Their gaze was also manipulated. It could either be static or dynamic. The robot looked straight ahead while talking to the participant in the static condition. In the dynamic condition, the participant divided its gaze between the participant and the computer screen. There was also a gesture and no gesture condition.

The experimenters used the Desert Survival Problem to create a human-robot interaction scenario. In this task, participants have to imagine that their plane crashed and they landed in a desert. They have a couple of items, like a knife, torch and a pair of sunglasses and so on. They then have to rank these items, and items that might increase their chances of survival receive a higher ranking. In the human-robot interaction scene, the participants created rankings for everything. The robot then tried to persuade them to change the fourth, sixth and ninth- ranked items higher. After the robot provided its suggestions, the participants were able to change their rankings.

The first hypothesis was confirmed. The authors predicted that participants would comply more with the robot’s suggestions when the robot applied nonverbal cues than when it would not use any nonverbal cues. The researchers also found that when the robot used only bodily cues, the participants changed their rankings based on the robot’s suggestions more than they did when the robots did not use these bodily cues. Vocal cues alone did not increase the rankings. This also confirmed the second hypothesis: bodily cues alone have a stronger effect on compliance compared to only vocal cues. The third hypothesis, however, was not confirmed. So, there was no interaction found between gender and different cues. However, the researchers did find that women rated the robot as more intelligent than men did and that men rated the robot more intelligent in the bodily cues condition than in the other conditions.

The authors point to two limitations of  their study. The first one is that their set of manipulations was not refined enough and that future research should use a better set of manipulations. The second limitation was that the Desert Survival Task created an interaction between robot and participant in a hypothetical scenario. The persuasive effects of the robot might not have been the same in a real-life scenario.