Nicole Salomons, Sarah Strohkorb Sebo, Meiying Qin, and Brian Scassellati. 2021. A Minority of One against a Majority of Robots: Robots Cause Normative and Informational Conformity. ACM Trans. Hum.-Robot Interact. 10, 2, Article 15 (February 2021), 22 pages. https://doi.org/10.1145/3442627
Abstract: Studies have shown that people conform their answers to match those of group members even when they believe the group's answer to be wrong [2]. In this experiment, we test whether people conform to groups of robots and whether the robots cause informational conformity (believing the group to be correct), normative conformity (feeling peer pressure), or both. We conducted an experiment in which participants (N = 63) played a subjective game with three robots. We measured humans’ conformity to robots by how many times participants changed their preliminary answers to match the group of robots’ in their final answer. Participants in conditions that were given more information about the robots’ answers conformed significantly more than those who were given less, indicating that informational conformity is present. Participants in conditions where they were aware they were a minority in their answers conformed more than those who were unaware they were a minority. Additionally, they also report feeling more pressure to change their answers from the robots, and the amount of pressure they reported was correlated to the frequency they conformed, indicating normative conformity. Therefore, we conclude that robots can cause both informational and normative conformity in people.
CCS Concepts: • Human-centered computing → Empirical studies in HCI; • Applied computing → Psychology; Additional Key Words and Phrases: Human-robot interaction, peer pressure, informational conformity, normative conformity
DISCUSSION
Previous results have shown that robots are capable of causing adult conformity [24, 36]. In this section, we discuss how robots in our study were causing normative conformity in addition to informational conformity. We also discuss how different tasks affected the conformity rates in human-robot interaction. Last, we discuss some potential future directions.
Informational Conformity
We believe robots were causing informational conformity due to three main reasons. First, when participants were given more information about the robots’ answers, they conformed significantly more. Second, participants viewed the robots as being capable of this task, which is an element that enables informational conformity. And third, the subjective nature of the task increases the willingness to accept information from the robots, which was confirmed by many participants in the open-ended question.
To measure whether robots were causing informational conformity, the current analysis focused on the quantitative, selected, and blind conditions, where the amount of information given to the participant varied. Participants conformed significantly more in the quantitative conditions than they did in the blind and selected conditions. Even though participants in the selected condition did conform a higher number of times on average than the blind condition, this difference was not significant, implying that being aware that at least one robot chose a different answer was not sufficient to sway the participant to change their answer. However, when the participants were aware that all the robots chose a different answer than they did, they more frequently conformed to the robots. The results suggest that having the information that one was in the minority in his/her answer increased the likelihood of accepting information from the majority as the participants had more information from the environment to make decisions [41].
One of the factors that is believed to influence informational conformity is the expertise of the group [10, 44]. Participants viewed the robots as performing well at this particular task: in the questionnaire, participants rated the robots as being similar to themselves at how good the robots were at the game. These results are surprising, considering robots do not usually perform well at high-level tasks such as understanding the meanings of images. Additionally, participants gave similar scores to the robots’ capabilities across the conditions, indicating that they are not viewing the robots as better at the game in one condition compared to another. However, participants conformed more in the quantitative condition compared to the selected and blind conditions. Once participants were given more information (such as how many robots chose each answer), they utilized this information and conformed to the robots. This indicated that participants believed the robots to have the correct answer but only had sufficient information to conform in the conditions with more information.
One of the main factors that influences informational conformity is uncertainty in the answer [11]. Individuals are more likely to copy others when they are uncertain [40]. Therefore, it was likely that the subjective nature of the game increased the participants’ likelihood of accepting information from the robots. Our results are in line with how participants responded to the open-ended question, where they frequently stated they were using the robots’ answers to decide their own final answer. For example, one participant in the quantitative condition wrote: “Yes, from life experiences, majority is usually correct.” Another participant in the quantitative condition wrote: “Yes, because I thought the way they decided was going to be right.” There were also participants in the selected condition who responded that they changed their answer due to information of the robots’ answers: ‘When they highlighted a different option, and then I felt that it was more apt than the one I chose.”
Normative Conformity
Our results show that normative conformity was playing a role in the participants’ decisions to conform to the robots in the staring and quantitative conditions: participants conformed significantly more in critical rounds in the staring and quantitative conditions compared to the other rounds. Participants reported feeling pressure to change because of the robots and acted upon it. Additionally, participants in the selected and blind conditions that did not have the information to conform during critical rounds changed their answers frequently to the answer of at least one of the robots in the next round.
To measure whether robots were causing normative conformity, this analysis was focused on the comparison of the staring, quantitative, and selected conditions (as in the blind condition, participants had no information on the robots’ answers, and therefore normative conformity was highly unlikely). In the quantitative condition, there were significantly more changes in participant answers during critical rounds than in the selected condition, demonstrating that being aware of the number of robots choosing certain answers influenced participant's decisions to conform. Therefore, being aware that one was in the minority in a group of robots increased the likelihood of conforming to them, compared to only knowing that at least one robot chose a different answer.
Participants in the staring, quantitative, and selected conditions were making, on average, a similar number of changes across all the rounds (critical and non-critical). However, participants in the staring and quantitative conditions were making many more of these changes in critical rounds. Providing participants with the information of how many robots chose each card did not increase overall changes but specifically increased the number of changes to the robots in the rounds where they were the minority. This is in line with previous research showing that having a unanimous group increases normative conformity [2].
There were no significant differences in the frequency of conformity between the staring and quantitative conditions. Therefore, adding the staring behavior did not significantly increase conformity. There are multiple possible interpretations of this. The first being that participants did not feel additional peer pressure because of the staring behavior either because it was not very observable or because they did not perceive it as a persuasive behavior. Another interpretation is that the quantitative behavior alone was already causing a large amount of peer pressure, and adding the staring behavior did not increase the frequency of conformity significantly. A previous study has shown that eye contact can actually create resistance to the person who is trying to persuade [9]; therefore, the staring behavior might have caused some participants to conform less. An additional possibility is that the staring behavior is causing psychological reactance in some of the participants towards the robots. Studies have shown that very apparent persuasive behaviors can decrease the amount of compliance [21]. Additional studies should be conducted to determine which social behaviors of robots cause increased peer pressure.
Participants in the staring and quantitative conditions were making most of their changes in critical rounds, whereas participants in the selected and blind conditions were frequently changing their answers in the round right after the critical round. Participants in the selected and blind conditions did not have the necessary information to see they were a minority in the critical rounds in time for them to change their answers. However, when the final answer was shown, the participants observed that all the robots had chosen a different answer than they did. We believe this caused the participants in these conditions to change their answers in the following round, attempting to choose the same answer as the robots. Additionally, the robots getting the answer right or wrong in the critical rounds did not appear to play a role in deciding to change their answer in the subsequent round. Therefore, the main reason they were changing their answer was not necessarily because they thought accuracy would be increased. Instead, we believe this was an indication of normative conformity where participants wanted to be in-group with the robots.
Participants in the staring and quantitative conditions reported higher pressure to change their answers because of the robots than the selected and blind conditions. Additionally, the amount of pressure to change was correlated with the number of critical round changes. This was an indication of normative conformity, where participants were feeling pressure to change and acting upon that pressure. Participants in the staring and quantitative conditions also mentioned feeling peer pressure in the open-ended question. Several participants commented that they changed their answers to match the robots’ answers when they were part of the minority, indicating that participants were changing to become part of the majority. For example, one participant in the staring condition wrote: “Yes, because they'd look at me judgmentally when I had a different answer, so it made me doubt myself.” Another participant in the staring condition commented: “Sometimes when they all chose the rose field, I felt dumb for picking the ballet shoes.” A participant in the quantitative condition wrote: “Yes (I changed) if they outnumbered me on one particular picture.”
Influence of Task in Conformity
Several studies have been conducted attempting conformity with robots, of which some observed conformity [24, 36, 43] and some did not [5, 7, 39]. The main difference between the experiments which observed conformity and those which did not was the task being tested. It is necessary to have a task where the participant is not certain of the correct answer. The robotic studies which failed to show conformity mostly tested Asch's line task, which has a clear, correct answer, whereas the studies which did show conformity with robots had a task in which the answer was not as clear. Our study used a subjective word-card matching task. Hertz and Weise [24] presented the questions to the participants for only 2.5 seconds and the accuracy rate of responding solely was 63% for the analytical task and 68% for the social task. Similarly, the tasks of Williams et al. [43] used socio-conventional and moral questions. Therefore, we believe it is necessary to have a more subjective task to cause conformity. This is in line with human psychology, where more difficult and subjective tasks have higher rates of conformity [3].
Other characteristics of our task that could have influenced the number of times participants conformed were that they were all sat at the same table close together [30] and that the answers were publicly shown on a shared screen [16]. Additionally, the answers of the robots were highlighted on the screen, focusing the participants’ attention on those answers, which could have influenced conformity [10].
Future Work
There are several different areas of potential future work, following our results. The robots used in the study were very simple, but despite their size and simplicity, they caused both informational and normative conformity. Factors that increase informational conformity are the similarity with the group [11] and the expertise of the group [44]. Future studies should analyze whether having robots with increased similarity to humans or with higher appearance of capabilities will also lead to more informational conformity.
Factors that increase normative conformity are group size, the immediacy of the group, and their social importance [30]. Future studies could analyze how changing the perceived social importance of the group, changing the number of robots, and changing how close the robots are will influence the frequency of normative conformity. Another factor that influences normative conformity is whether the other members are considered in-group or out-group [1]. Several studies have shown that in-group robots are rated more anthropomorphic and are favored over out-group members [18, 19]. Therefore it should be studied how group membership and anthropomorphism influence conformity.
Our results indicate that conformity is directly linked with the type of task being tested. Future work could analyze how conformity changes depending on the type of task and to further investigate if conformity to robots can be used in pro-social ways [14]. Last, conformity is influenced by individual characteristics [40]. Culture [6], age [15], gender [6, 17], and other personal factors have been shown to influence the decision to conform in human groups. More studies on different personal factors should be studied to see how they influence conformity to robot groups.
Posts
No comments:
Post a Comment