In this section, we explain why participants might ignore robots based on the CDT, explain the prerequisites for dissonance reduction, and present the results of a preliminary interview study.
3.1 CDT in Admonishment Scenarios
How could CDT explain why people are resistant to robots’ negative feedback (see Figure
2 for an overview)? A feeling of cognitive dissonance constitutes when one’s belief and behavior are incoherent. In the shopping mall scenario, people might think that public regulations and rules ((1) initial beliefs; see Figure
2) are essential for social cohesion, but they sometimes violate these rules. If then confronted with the feedback ((2) understanding) from somebody/something (e.g., human, robot), that their behavior is not aligned with their actual beliefs, it could create a feeling of dissonance and result in an uncomfortable feeling ((3) discomfort). Since having an uncomfortable feeling is mentally unstable, people try to reduce this discomfort. In the best case, people would change their behavior to reduce discomfort. Otherwise, people could reduce the dissonance feeling and justify their behavior by either
•
modifying their cognition (e.g., there is nobody around anyway, so it is okay to use the smartphone);
•
adding cognition (e.g., it is okay to use the phone, because I need to find the location on the map);
•
denying the conflict (e.g., there is no evidence that using the smartphone actually causes incidents);
•
or trivializing (e.g., I just don’t care about the rules or what you say to me).
We assume that people would use different reduction strategies, depending on the agent that approaches the people and admonishes them. However, finding out people’s strategies to reduce dissonance is challenging in an experimental lab setting. Hence, our first stage of this research is to conduct field observations and conduct semi-structured interviews. This first step will clarify which cognitive strategies participants use when confronted with the admonishment of a robot. Primarily, we want to investigate the possible explanation of why participants ignore the robot in an uncomfortable situation. Thus, we look at it from a CDT perspective and gather preliminary hints for which dissonance reduction strategy is active when participants ignore or follow the robot’s admonishment.
3.3 Method
Since our previous observations (see Reference [
24]) showed that not all participants follow the robot’s instructions and some ignore the robot, we started a new investigation and conducted interviews with participants to find out why they either ignore the robot or follow the instructions. The interview should clarify whether participants experience cognitive dissonance when robots admonish them and how they reduce their dissonance. The interview consists of five main questions to probe for the four requirements for cognitive dissonance (initial belief, understanding what the robot said, participants’ discomfort, and the reduction strategies participants had (not) to follow the instructions). Finally, we were interested in whether participants perceive the robot as being able to make moral judgments. In the following, we will explain which question we asked the participants (see Figure
3 for an overview).
Introduction and Understanding. To know whether participants were noticing the robot and could listen to its admonishment, we asked them what the robot was saying to them and whether they noticed the robot. To make sure that the robot is perceivable, we used a loud voice. Additionally, we can validate the statement using video recordings. Saying that they did not understand the robot could also be a cognitive reduction technique.
Q1: Discomfort. Another assumption for dissonance reduction to occur is an uncomfortable feeling. Hence, we asked participants how they felt when the robot was approaching and talking to them.
Q2: Reduction. The question related to the used reduction strategy clarifies the participants’ cognition to justify their behavior. We asked them how they decided to follow or not to follow the instructions of the robot or why they continued to use their smartphone.
Q3: Beliefs. One requirement for cognitive dissonance is that participants have initial beliefs and that their behavior is violating these beliefs. To probe for this requirement, we asked participants what they think about this public regulations and rules in general. This question targets to identify whether they actually hold the belief that using a smartphone while walking is dangerous, which could hint to a potential conflict when the robot confronts them with the fact, that they are not practicing what they preach.
Q4: Moral Judgment. The robot’s task includes to evaluate the participant’s behavior being dangerous or not, thus, right behavior or wrong behavior. It implies that the robot is capable of evaluating participants’ behavior and performing a morally sensitive task. Therefore, we focused in our interview on the perceived morality of the agent and asked: “Is the robot able to decide what is right or wrong?” And “Why?”
Of course, there are various factors (e.g., authority, morality autonomy) that could influence the activation of different reduction strategies. Unfortunately, this interview is highly cost-sensitive, and we can not probe the participants for all possible confounds.
3.3.1 Robot Platform.
A humaniod robot Robovie R3 is deployed for admonishment experiments. The robot is with an omni-directional base for movement. Its movement speed is limited at \(1.2m/s\) and its translational acceleration is also limited at \(1.5m/s^2\) . Its rotational motion is set within the bound of \(1rad/s\) and \(2rad/s^2\) . The robot is with a 32-layer Velodyne LiDAR for self-localization and multiple Hokuyo URG-laser scanners for short-distance obstacle detection. For the sake of admonishment, the robot enables to play prescribed sentences under a kid’s voice. In addition, the robot can exhibit various gestures through motions of its upper body; for instance, to express body language via rotating its head and two arms. The robot is able to get teleoperated via joypad by human operators during the field experiments.
3.3.2 Procedure.
A field observation experiment was run on four days on two consecutive weekdays in a large shopping mall in Japan. We placed the robot in a large hall of the mall from 11 am until the battery was empty or encountered technical difficulties during the experiment. Two experimenters controlled the robot, and two experimenters conducted interviews with the participants. One experimenter was teleoperating the robot, while the other was spotting participants to approach and counted the participants’ behavior.
When a person using a smartphone while walking appears in the hall, a Wizard of Oz operator moves the robot to the person, and once approaching, the robot says two times: “Excuse me. It is dangerous to use your smartphone while walking. Please stop doing it.” We operated the robot to follow the person until the participants stopped using the phone. In case the participants ignore the first round of admonishment, the robot would repeat the admonishment sentences one round again (two times) until the participants finally stopped using the phone or are out of reach for the robot.
In any case, after the admonishment, an interviewer approached the pedestrian and asked her/him for an interview. The interviewer explains that the interview will take around 10 minutes and asks for informed consent to record the interview. After participants gave informed consent, the interviewer always asked whether they understood what the robot was saying. The order of the subsequent questions was not fixed and depending on the answers of the participants. After the interviewer asked all leading questions, the interview ends, and the interviewee receives a monetary compensation of 3000 Yen.
Inclusion Criteria for admonishment. We controlled the robot to admonish participants that were using their smartphone while walking. As a clear inclusion criterion, we defined participants that used their smartphone or were, in general, looking at the smartphone display while walking in the shopping mall. We included approaching couples or groups of participants when either all of them used a smartphone or if one person of the group used a smartphone and the robot could easily approach this person (i.e., when other members of the groups were not blocking the way to the person that used a smartphone while walking).
Exclusion Criteria. We excluded approaching participants that were using headphones, because we could not guarantee that these participants would notice the robot and hear the robot’s admonishment. We also did not approach participants who were only holding their phone while walking and looking in their moving direction. Finally, we excluded to approach participants that obviously looked like non-Japanese tourists, because we can not assure that they can speak Japanese.
Participants. In total, the robot admonished 160 persons; 84 persons followed the instruction and stopped using their smartphone after the first admonishment sentence; 76 participants ignored the robot after the first admonishment sentence. After the final reaction, 88 stopped using their phone (meaning four people additionally stopped using their smartphone) and 72 continued to ignore the robot. A possible reason may be that participants were already away from the robot. Also, they may regard the repeated admonishment not as towards themselves.
We collected 12 interviews from participants that followed the instruction. Nevertheless, only 7 participants that were ignoring the robot’s instruction wanted to give an interview. Participants who ignored the robot were hastening from the scene and thus hard to catch for the interviewers. Since the robot experiments were done during noon/afternoon on workdays, only very few pedestrians agreed to get interviewed during the study. As a response, the major excuse/reason given by the participants was “I’m busy” or “There’s no time.” While many other participants who did not vocally refuse the interview chose to walk away without stopping when the authors spoke to them or chose to give a gesture of refusal (shake their heads or hands) and walked away silently.
3.3.3 Analysis Procedure.
We recorded the interviews and translated them from Japanese to English. We analyzed the answers regarding hints for different strategies participants are using to justify their behavior (i.e., modify, add, deny, trivialize), the feeling they describe (e.g., confused, comfortable, surprised, annoyed, unconcerned, ashamed), the initial beliefs about regulations (whether regulations are important or not). Finally, we looked at whether participants think the robot is capable of judging people (yes/no) and the reasons for why they think it can or can not judge people.
Additionally, we categorized the behavior of the people as stopped and ignored after the admonishment:
•
Participants who were continuing to use their smartphone after the admonishment. People in this category might have had a look at the robot after the admonishment, but continued looking at their phone.
•
Participants who stopped using their smartphone after the admonishment. People in this category typically looked at the robot, stopped looking at their smartphone, and put it in their pocket or in a smartphone case. In some cases, they stopped using their smartphone without looking at the robot.
After an initial screening of the participant’s responses, we identified that we could distinguish the interview responses into a binary coding. For Q1, we choose to either categorize the responses whether participants are focusing on their cognition or their emotion. Regarding Q2, we choose to either categorize the responses based on whether they chose to change their behavior because they felt guilty or ashamed (i.e., self-blame) or whether participants blamed the robot for not being a machine (e.g., not human-like, not polite). Q3 assessed the participants’ belief that rules are important or not important; we coded the responses accordingly. Finally, we either had responses for Q4 falling in the category that the robot is capable of judging the participants’ behavior or not.
3.4 Results
1 shows a count summary in which categories the interview responses fall. In the following, we briefly summarize the main results for each interview question.
3.4.1 Question 1.
The results indicate that there is a subtle distinction in the response behavior between participants that followed the instruction of the robot and participants that were not following it. Regarding question Q1, participants who stopped using their smartphones were more referring to their actual feelings when the robot admonished them (e.g., ID2: “I was scared for a moment”; ID5: “I was a little surprised about the robot recognizing this and scared of the noise”). In contrast, participants in the ignoring group talk more about their cognition and perception (e.g., ID3: “The robot came very slowly, it saw me”; ID18: “I have noticed it, but I didn’t know why it was there”). They highlight that they could see or notice the robot approaching them and their reasoning about it.
3.4.2 Question 2.
The question on their reasons to change their behavior gives insights on participants’ cognitive strategies to cope with the situation. Participants who stopped highlight their feeling of being ashamed and being evaluated by others like their families (e.g., ID1: “When the robot told me, I felt ashamed because of the people around me”). Additionally, they pointed out that the robot was right about what it was saying and that using the smartphone while walking can indeed be dangerous (e.g., ID6: “Because the robot told me, I thought I have to be careful”). Participants who ignored the robot focused their reasoning on the robot, but not on other aspects that could justify using the smartphone while walking and trivialize the robot (e.g., ID17: “The robot is just a machine,” ID19: “The robot wasn’t saying ‘excuse me’ or something human-like before talking to me,” ID12: “I thought it was just a guidance. Robot voice not humanlike enough”).
3.4.3 Question 3.
To find out whether participants could feel a sort of cognitive dissonance because their behavior is not in line with their own beliefs, we made sure that the task that the robot is doing is essential for the participants. The answer to this question revealed that most participants in both cases think that it is, in general, necessary to follow social rules (e.g., ID5: “We should follow rules”; ID19: “I have children, so I have to be a role model.). However, in both conditions, participants also admit that exceptions for rules are legitimate, and it is essential to balance (e.g., ID1: “Public rules are important, but it depends on the circumstances”; ID19: “But if it is urgent a little bit is ok”).
3.4.4 Question 4.
Regarding the last question, whether the robot can judge people, we can not find a clear, distinct pattern between the different behavior outcomes. In both situations, some participants think that robots can judge the behavior of participants (e.g., ID4: “No, I think the robot was programmed. Robots need human support also in the future”) and some participants thought it is technologically too challenging and that the robot must be tele-operated (e.g. ID15: “I think it’s possible, I wonder what the robot will do when it arouses antipathy”).
3.5 Discussion
We conducted a preliminary field observation to investigate whether we could explain why participants follow or ignore a robot’s instruction based on the CDT. This theory could explicate participants’ cognitive dissonance reduction modes to handle an uncomfortable situation. To verify whether a cognitive dissonance applies to this scenario and whether participants try to reduce it, we wanted to probe for the four necessary prerequisites. These could show that participants perceive a dissonant feeling when confronted with the robot’s admonishment and use a cognitive reduction strategy to reduce the dissonance. These four prerequisites are:
(1)
participants should hear what the robot was saying,
(2)
they should hold the belief that following public regulations are essential,
(3)
they should feel discomfort,
(4)
finally, participants should either change their behavior to resolve the dissonance and uncomfortable feeling or they should try to find cognitive reasons that can still justify their behavior.
In the following, we will discuss whether we found evidence for these four prerequisites.
Interpretation: Prerequisite 1. First, we can assume that all participants could understand the robot’s statement and heard what the robot was saying. Thus, we met the essential requirement for our investigation.
Interpretation: Prerequisite 2. Regarding the importance of following public rules, we found evidence that, in general, participants agree on the importance of following public policies. However, participants in both conditions highlight that there can be situations where it is allowed to obey the rules. These answers indicate for one of the modes of cognitive dissonance reduction, adding or modifying the cognition (e.g., a quick look is okay, if there are not so many people). However, this is not a distinct pattern between participants that ignored or followed the robot’s admonishment. Participants that changed their behavior also think that it is sometimes okay to ignore rules, and at the same time, some participants who ignored the robot think that public rules are essential. Thus, adding or modifying cognition seems not to be the critical mode to justify participants’ behavior. Regardless of the modification, we can assume that most of the interview participants believe that public rules are essential.
Interpretation: Prerequisite 3. Concerning the feelings participants had when confronted with the robot’s admonishment, we found an apparent difference between compliant and non-compliant participants. Participants that stopped using their smartphone relate more to their feelings in their answers. Since the robot confronted them with a contradiction in their behavior and beliefs, they did not reduce their negative feelings but instead changed their actual behavior. Accordingly, participants that did not change their behavior and continued to use their smartphone relate more to their perception and cognition. This result highlights that participants might reduce their uncomfortable feeling and thus could not articulate or access their negative emotions. The missing notion of feelings in the ignoring group hints that they suppress their feelings because they did not change their behavior.
Interpretation: Prerequisite 4. While participants who stopped using their smartphones focus on their feeling of being ashamed or the veracity that the robot is correct, the participants who ignored the robot show that they do not take the robot’s instruction seriously but rather think of it as general guidance. Additionally, participants who ignored the robot also highlight technological immaturities like the robotic voice or appearance. These contradicting statements present evidence that participants trivialize the robot and thus justify not ignoring it. Except for one participant, the majority of participants solely used agent trivialization arguments.
Overall, our interview results show a tendency to explain pedestrian’s behavior in light of the CDT. Some participants justified that they did not follow the robot’s instructions by using the mode of adding or modifying their cognition, but most participants used trivializing strategies to reduce their dissonant feeling.
In the next part of our work, we elaborate on how we could exploit this finding to enhance the robot’s effectiveness and persuade more participants to follow the robot’s instructions.