The Less You Look the More You Get: Designing Gaze Avoidance in Human-Robot Interaction

2.1. Gaze Cues in Human-Robot Interaction

Gaze is usually recognized as an essential component of human-robot communication. As communicating with a social partner with gaze can make positive impressions, many studies have been conducted on the topic of gaze with robots in the field of HRI (Imai et al., 2002; Sidner et al., 2004). For example, Imai et al. (2002) found that people can precisely interpret a robot’s attention using cues from its gaze. Sidner et al. (2004) proposed that a robot’s use of gaze and gestures increased people’s engagement with them. Although these studies provide some evidence that robot gaze affects people’s perception, a study on the effect of the gaze avoidance is lacking.

2.2. Gaze Avoidance in Human-Human Interaction

Gaze avoidance can be used differently depending on the situation. While gaze avoidance is recognized as an impolite and inadequate attitude during a normal conversation, it is perceived as a communicational intention during an embarrassing situation (Keltner & Buswell, 1997). One of the embarrassing situations people experience is when asked to a question that invades their privacy (Keltner & Buswell, 1997). Consistent with this study, nonverbal cue, such as gaze avoidance could be used as a cue to reveal intention and information to others about the mental and emotional states of an individual (Ekman & Friesen, 1969; Zuckerman et al., 1981). Therefore, we assume that if a robot is asked an embarrassing question, gaze avoidance of the robot can be used as a channel to express the robot’s intention to escape from that moment. According to a study by Sohn et al. (2011), a robot with intention can be perceived as more sociable and intelligent.

These analyses led to the following hypotheses:

H1. A robot with gaze avoidance is perceived as more sociable than a robot with gaze in an embarrassing situation while a robot with gaze is perceived as more sociable than a robot with gaze avoidance in a normal situation.

H2. A robot with gaze avoidance is perceived as more intelligent than a robot with gaze in an embarrassing situation while a robot with gaze is perceived as more intelligent than a robot with gaze avoidance in a normal situation.

H3. A robot with gaze avoidance is perceived as having intention than a robot with gaze in an embarrassing situation while a robot with gaze is perceived as having intention than a robot with gaze avoidance in a normal situation.

3.1. Participants

As most of university students are familiar with high technology, we recruited twenty-four undergraduate students as participants. Participants’ ages are ranged from 22 to 27, and gender was balanced across conditions (Male:11, Female:13). They were fairly well educated on average with the college level. Participants were given a $1 gift certificate.

3.2. Materials

The robot we used in the experiment was the Eye-robot shown in Fig. 1. We constructed four different scenarios. A woman asked the robot either a normal question, “Have you ever had a pet?” or an embarrassing question, “Have you ever lied?” At the end of each question, the robot answered using gaze or gaze avoidance.

Figure 1

Conditions by the Gaze Type (a: gaze, b: gaze avoidance)

3.3. Procedure

Participants were welcomed to the lab and watched a short introductory video about the function of the robot and about the experiment. They were then shown four videos with different robots in a random order. After the participants saw each video, a questionnaire regarding each stimulus was administered.

3.4. Measures

The post-experimental survey was composed of ten seven-point Likert type items, which were combined into two scales.

(1) Sociability Measures

Sociability was an index of five items (Cronbach’s α= .95) which were drawn from Parise et al.’s (1998) research. The five items were “cheerful, friendly, optimistic, warm, and happy.”

(2) Intelligence Measures

Intelligence was an index of four items (Cronbach’s α= .94), which were drawn from Parise et al.’s (1998) research. The four items were “knowledgeable, responsible, intelligent, and sensible.”

(3) Intentionality Measures

Intentionality was an index of one item (Short et al., 2010). Participants were asked whether gaze or gaze avoidance was interpreted as having intention to escape from a normal or an embarrassing situation.

3.5. Experimental Manipulations

The independent variable question type had two levels: normal question and embarrassing question. Participants watched either the video a woman asked the robot a normal question, “Have you ever had a pet?” or an embarrassing question, “Have you ever lied?” In the questionnaire, participants were asked to indicate whether the question type was either a normal or an embarrassing question using seven-point Likert scale. The differences by the question type were significant (t=-32.95, df=94, p=0.000, one-tailed). Participants who were in a normal situation (M=1.65) rated the question type as a normal question and those who were in an embarrassing situation (M=6.27) rated the question type as an embarrassing question.

The independent variable gaze type had two levels: gaze and gaze avoidance. Participants watched either a robot with gaze or gaze avoidance. In the questionnaire, participants were asked to indicate whether the robot gazed or not using seven-point Likert scale. The differences by gaze conditions were significant (t=-51.29, df=94, p=0.000, one-tailed). About the degree of gaze, participants who were shown a robot with gaze rated higher (M=6.54) than a robot with gaze avoidance (M=1.13).

4.1. Sociability

In the case of sociability, as predicted by H1, there was a significant difference depending on the question type and the gaze type (F(1,23)=124.19, p<.001). Participants evaluated a robot with gaze avoidance in an embarrassing situation (M=5.66, SD=0.83) as the most sociable compared to a robot with gaze in a normal situation (M=3.98, SD=0.58) and a robot with gaze in an embarrassing situation (M=3.32, SD=0.70). The robot with gaze avoidance in a normal situation was evaluated as least sociable (M=2.48, SD=0.72). Table 1 and Fig. 2 show the effect of the question type and the gaze type on the perceived sociability of a robot.

Table 1

The Effect of the Question Type and the Gaze Type on the Perceived Sociability of a Robot

4.2. Intelligence

In the case of intelligence, as predicted by H2, there was a significant difference according to the question type and the gaze type (F(1,23)=131.23, p<.001). A robot with gaze avoidance in an embarrassing situation (M=5.68, SD=0.74) was evaluated as more intelligent than a robot with gaze in a normal situation (M=3.89, SD=0.41) and a robot with gaze avoidance in a normal situation (M=3.22, SD=0.61). Participants evaluated a robot with gaze in an embarrassing situation as least intelligent (M=3.06, SD=0.74). Table 2 and Fig. 3 show the effect of the question type and the gaze type on the perceived intelligence of a robot.

Table 2

The Effect of the Question Type and the Gaze Type on the Perceived Intelligence of a Robot

4.3. Intentionality

In the case of intentionality, as predicted by H3, there was a significant difference according to the question type and the gaze type (F(1,23)=279.88, p<.001). A robot with gaze avoidance in an embarrassing situation (M=5.83, SD=0.92) was evaluated as having more intention than a robot with gaze in a normal situation (M=3.08, SD=0.78) and a robot with gaze avoidance in a normal situation (M=2.46, SD=0.59). Participants evaluated a robot with gaze in an embarrassing situation as having least intention (M=1.71, SD=0.86). Table 3 and Fig. 4 show the effect of the question type and the gaze type on the perceived intentionality of a robot.

Table 3

The Effect of the Question Type and the Gaze Type on the Perceived Intelligence of a Robot

5.1. Summary and Interpretations of Results

All three hypotheses were supported by the data. As predicted by H1, people perceived the robot with gaze avoidance in an embarrassing situation as sociable unlike the robot with gaze avoidance in a normal situation. Gaze avoidance was interpreted as a positive attitude in an embarrassing question while as a negative attitude in a normal question. This indicates that gaze avoidance can be used differently as a social attitude depending on the situation.

As predicted by H2, people evaluated the robot with gaze avoidance in an embarrassing situation as most intelligent while the robot with gaze in an embarrassing situation as least intelligent. This implies that gaze avoidance had a positive effect on robot intelligence in an embarrassing situation. In addition, while people perceived the robot with gaze avoidance in a normal situation as less sociable than the robot with gaze in an embarrassing situation, the opposite result was revealed in regards to the perceived intelligence of the robot. This informs that even though gaze avoidance could be interpreted as having less sociability, it could be perceived as having intelligence in a normal situation.

Consistent with H3, the robot with gaze avoidance in an embarrassing situation was interpreted as having intention, while a robot with gaze in an embarrassing situation was interpreted as having no intention. This finding indicates that gaze avoidance could be effectively applied to endow a robot with intention.

5.2. Implications for Design

The results of this study have implications for the interaction design of robots. Designing social cue considering the social context is important as the same social cue, such as gaze avoidance is perceived differently by the social context. Gaze avoidance can positively affect robot’s sociability and intelligence in an embarrassing situation. On the other hand, in a normal situation, gaze avoidance can be interpreted as having less sociability. Thus, gaze avoidance is desirable when a robot is in an embarrassing situation while it needs to be avoided in a normal situation. Like gaze avoidance, gaze could be differently interpreted by the social context. In a normal situation, gaze can be effectively used as a social cue to increase a perceived sociability and intelligence of a robot. On the contrary, in an embarrassing situation, gaze could be interpreted as having less intelligence and sociability. These findings inform our understanding of how people perceive and interact differently with a robot depending on the situation with the same social cue. These results suggest that designers of robots should consider not only the gaze type of a robot but also the social context.

In addition, this study showed that gaze avoidance could express intentionality of a robot. A robot makes mistakes due to the limitation of technology. Alleviating these inconveniences is important to sustain users’ acceptance of a robot. When a robot is likely to have mistakes, gaze avoidance could be used as a strategic way to reveal the robot intention by minimizing people’s dissatisfaction of a robot.

5.3. Limitations

There are several limitations to this study. First, our participant pool was limited to university students. Replicating this study with people of different ages, backgrounds, and cultures is an important next step. Second, the robot in this study was one type of robots. Future studies need to be done using various types of robots. Third, we examined interactions between humans and robots in an experimental room with short-term study. As interactions in more natural settings featuring different tasks may produce different results, future studies need to examine long-term experience in natural settings. While this study provides evidence that gaze and gaze avoidance could be used as social cues for robots, further work with wider social contexts is required.