Effects of Agents’ Embodiment and Robot Anxiety Scale on Social

Priming

Tomoko Koda

and Kensuke Kikuzawa

Department of Information Science and Technology, Osaka Institute of Technology, Osaka, Japan

Keywords: Virtual Agent, Conversational Agent, Humanoid Robot, Social Priming, Priming Effect, Social Presence,

Embodiment, Robot Anxiety Scale, RAS.

Abstract: In this study, we focus on the priming effect that would affect the social relationships among the agent, robot,

and human because the opportunities to converse with robots and agents are increasing. This research

investigated the effect of embodiment of the priming agent on the perception of social presence of the primed

agent. The preliminary results did not support our hypothesis that "the social presence of the primed agent

becomes higher when the embodied robot primes than when the virtual agent primes." However, the results

indicated that there is a dichotomy in the perceived social presence between the participants' groups when we

divide them according to their anxiety level toward robots. This indicates that the priming effect on the social

presence of the primed agent is different depending on the embodiment of the priming agent and people's

anxiety toward robots.

1 INTRODUCTION

In recent years, our daily lives widely include

humanoid robots and virtual agents equipped with

conversational ability. Robots and agents help

humans as receptionists, museum guides, and even as

job interviewers. Because of their ability to converse

with humans, they can affect the social relationship

between the humans, agents, and robots. In this study,

we focus on the priming effect that would affect the

social relationships among the agent, robot, and

human. Priming effect is a phenomenon in which

giving prior information influences a person's later

judgment (Bargh, et al. 1996). Priming effect occurs

among humans in everyday life. Because the

opportunities to converse with robots and agents are

increasing, we believe it is important to investigate

the effect of priming on the social relationships

among humans, humanoid robots, and agents.

As agents and robots have different embodiments,

the amount of potency (i.e., perceived extroversion,

perceived self-confidence, and presence) from them

is different. Furthermore, their different embodiments

may affect not only their potency but also their social

presence. In other words, robots and agents would

affect our behavior or relationship with other humans

http://www.oit.ac.jp/is/~koda/hiserver01/profile-e.html

(de Greef et al. 2001). Daher et al. tackled this issue

by using two agents, namely, a priming agent and a

primed agent for the experiment. They examined the

impact of priming on the perceived social presence of

the primed agent judged by human participants. Their

result indicated that the priming by the agent affected

the human perception of the social presence of the

primed agent (Daher et al. 2017).

Researchers in human-agent interactions and

human-robot interactions have reported people's

different behaviors and attitudes to collocated robots,

remote robots projected on the display, and virtual

agents that performed the same task. Some

researchers report that they evaluated robots to be

more social, reliable, and competent than a virtual

agent in a mutual task (Pan et al. 2016, Powers et al.

2007), while such difference was not found in a

simple greeting interaction (Bainbridge et al. 2008).

Another research showed that an embodied robot is

more reliable than a robot projected on the display

(Kiesler et al., 2008). Therefore, it is necessary to

verify Dahers' findings by using other embodiments,

i.e., collocated robots, instead of virtual agents.

The purpose of this study is to verify whether the

embodiment of the priming agent affects the social

presence of the primed agent. We will follow the

266

Koda, T. and Kikuzawa, K.

Effects of Agents’ Embodiment and Robot Anxiety Scale on Social Priming.

DOI: 10.5220/0009144202660271

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 1, pages 266-271

ISBN: 978-989-758-395-7; ISSN: 2184-433X

experimental procedure of Daher et al. 2017 and add

a humanoid robot to the priming side.

2 EXPERIMENT

2.1 Hypothesis and Experimental

Conditions

From the related research (de Greef et al. 2001, Daher

et al. 2017, Kiesler et al. 2008), we formed the

hypothesis that "the social presence of the primed

agent becomes higher when the embodied robot

primes than when the agent primes." The

experimental conditions are the robot condition

where the robot primes the participant about the

game-playing agent, the agent condition where the

virtual agent (different from the game-playing agent)

primes the participant about the game-playing agent,

and the control condition where there is no priming

before the game. We conducted the experiment as a

between-subject design.

The experiment in this study follows the

procedure in (Daher et al. 2017), where the

participants play two games of twenty questions with

the game-playing agent. Twenty questions is a game

in which one side asks the opponent a question that

should be answered with a yes or no, and then guesses

what the opponent has in mind. Fig. 1 shows the

experimental scene of (Daher et al. 2017). A life-

sized agent (right) gives priming on another agent

(left) that plays twenty questions with the

participants.

Figure 1: Experimental scene of (Daher et al. 2017).

2.2 Experimental Environment

We developed a female agent who plays the game

with the participants and a male agent who performs

priming before the game using Unity5.2.1f1 and C#.

Fig. 2(a) shows the male agent, and Fig. 2(c) shows

the female agent who plays the game. The robot used

for the experiment is CommU, a social conversational

robot made by Vstone Inc. Fig. 2(b) shows the robot's

appearance. We used Java to control the robot

motions and dialogues during the game.

In this experiment, the male agent (shown in Fig.

2(a), priming agent) or the humanoid robot (shown in

Fig. 2(b), priming robot) give priming information to

the participants about the female agent (shown in Fig.

2(c)). The female agent plays the game with the

participants.

The distance between the participant and the

primed agent or the robot was set to 90 cm, which is

defined as an individual distance between

acquaintances by Hall (Hall 1966). Fig. 3(a) shows

the experimental settings under the control condition,

Fig. 3(b) shows the agent condition, and Fig. 3(c)

shows the robot condition.

(a) (b) (c)

Figure 2: The agents and the robot's used in the experiment.

(a) priming agent, (b) priming robot, (c) primed agent

(game-playing agent).

(a) (b) (c)

Figure 3: Three experimental conditions. (a) control

condition, (b) agent condition, (c) robot condition.

2.3 Experimental Procedure

In total, 43 Japanese university students of age group

19 to 24 years (36 men and 7 women) participated in

the experiment. We should consider the participants’

anxiety level towards robots because we use a robot

in the experiment and the existence of the robot might

affect the effect of priming. Therefore, we asked the

participants to answer the questionnaire on robot

anxiety scale (RAS) (Nomura et al. 2006) before we

started the experiment. RAS consists of three

categories, namely, robot’s conversational ability

anxiety (i.e., the robot may not understand complex

stories), robot behavior characteristics anxiety (i.e.,

what kind of behavior the robot performs), robot

Effects of Agents’ Embodiment and Robot Anxiety Scale on Social Priming

267

dialogues anxiety (i.e., whether one's utterance is

understood by the robot). The questionnaire has 11

questions and answers with a 6-point Likert scale.

In cases of robot and agent conditions, the

participants hear a conversation between the robot or

the priming agent and primed agent before they start

a game. The intention of this conversation is to tell

the participants that the priming robot or the agent had

been playing the game with the primed agent just

before the participants' turn and they really enjoyed

the game. When the participant enters the room to

play the game, he sees the priming agent saying, “I

really enjoyed the twenty questions game with you. It

was really interesting and you are very good at

guessing what I had in mind." The priming

conversation lasts about 15 seconds. In the case of the

control condition, the game starts without priming

(thus there is no priming agent or robot situated in the

experimental room as shown in Fig. 3(a)) after the

instruction of the experiment. We conducted the

experiment in a between-subject design, thus each

participant experienced only one condition.

The twenty questions game proceeds as follows.

The participant and the agent select one out of nine

creatures (from a list given during the instruction)

without telling the others. In the first turn, the

participant asks the primed agent questions and the

agent answers with a yes or no, and the participants

guess the animal that the primed agent has in mind.

The roles are reversed in the second turn. The

participant answers the questions from the agent, and

the agent guesses the animal the participant has in

mind. Examples of dialogues with the agent include,

"Is it a four-legged animal?" "Does it swim?" and

"Does it run fast?"

After the twenty questions game, each participant

answers a questionnaire on the social presence of the

primed agent that they have finished playing the game

with. This questionnaire is composed of questions

about the social presence of humans defined in Harms

et al. 2004.

The experimental procedure is as follows:

1) The participants receive instructions about the

experiment and answer the RAS questionnaire.

2) They enter a separate room for the twenty

questions game.

3) Only one of the three conditions is assigned to

each of them: the robot condition, the agent

condition, or the control condition. In case of

conditions for receiving priming (the robot

condition and the agent condition), the

participants hear a conversation between the

priming robot or agent and the primed agent as

they enter the other room. In the control

condition, the participants immediately go to step

4) They play twenty questions game with the game-

playing agent two times.

5) They move to the original room where they were

instructed before the game, and they answer the

questionnaire on the social presence of the game-

playing agent.

2.4 Questionnaires

Answers to the RAS questionnaire were rated with a

6-point scale (1: I do not feel anxious at all – 6: I feel

very anxious). Table 1 shows the questions from the

RAS questionnaire. Because we assume that one’s

anxiety level towards robots would affect the priming

effect from the robot, we divided the participants into

two groups according to their RAS scores. The

average RAS score of the 43 participants was 39.1.

We categorized 20 participants whose scores were

higher than 39.1 as high RAS group and 23

participants whose scores were lower than 39.1 as low

RAS group.

Although we assigned a similar number of

participants to each condition, the number of

participants in the two RAS groups was not even. The

number of participants who were assigned the agent

condition was 14 (high RAS group: 6 participants,

and low RAS group: 8 participants), who were

assigned the robot condition was 14 (high RAS: 4,

low RAS: 10), and who were assigned the control

condition was 15 (high RAS: 10, low RAS: 5).

Table 1: RAS questionnaire (Nomura et al. 2006).

Anxiety toward

Communication

Capability of

Robots

Robots may talk about something irrelevant

during conversation.

Conversation with robots may be

inflexible.

Robots may be unable to understand

complex stories.

Anxiety toward

Behavioral

Characteristics

of Robots.

How robots will act.

What robots will do?

What power robots will have?

What speed robots will move at?

Anxiety toward

Discourse with

Robots

How I should talk with robots.

How I should reply to robots when they talk

to me.

Whether robots understand the contents of

my utterance to them.

I may be unable to understand the

contents of robots’ utterances to me.

The social presence questionnaire was rated on a

7-point scale (1: I do not agree at all – 7: I agree very

much). The questionnaire consists of 6 categories:

Co-Presence (CoP) that measures the sense of being

with the other party, Attentional Allocation (Atn) that

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

268

measures the degree of attention paid to the other

party, Perceived Message Understanding (MsgU) that

measures the degree of understanding of the other

party's thoughts, Perceived Affective Understanding

(Aff) that measures the degree of understanding of the

other party's feeling, Perceived Emotional

Interdependence (Emo) that measures the degree of

being influenced by the other party's feelings, and

Perceived Behavioral Interdependence (Behv) that

measures the degree of being influenced by the other

party's behaviors. Table 2 shows questions from the

social presence questionnaire adopted from Harms et

al. 2004.

Table 2: Excerpted questionnaire on social presence

(Harms et al.2004).

Co-presence

(CoP)

I noticed (my partner). (My partner)

noticed me.

(My partner’s) presence was obvious

to me.

Attentional

Allocation

(Atn)

I was easily distracted from (my

partner) when other things were going

on.

(My partner) was easily distracted

from me when other things were

going on.

Perceived Message

Understanding

(MsgU)

My thoughts were clear to (my

partner).

(My partner’s) thoughts were clear to

me.

Perceived Affective

Understanding

(Aff)

I could tell how (my partner) felt.

(My partner) could tell how I felt.

Perceived Emotional

Interdependence

(Emo)

I was sometimes influenced by (my

partner’s) moods.

(My partner) was sometimes

influenced by my moods.

Perceived

Behavioral

Interdependence

(Behv)

My behavior was often in direct

response to (my partner’s) behavior.

The behavior of (my partner) was

often in direct response to my

behavior.

3 RESULTS

We conducted a one-way ANOVA for the social

presence questionnaire with the priming factor.

Consequently, the main effect was seen in the priming

factor in MsgU (F=4.555, p=<.01). The control

condition was rated significantly higher than the robot

condition. In addition, the main effect was seen in the

priming factors in Emo (F=5.625, p=<.01). The

control and the agent conditions were rated

significantly higher than the robot condition. The

main effects of priming factors were not observed in

the other four social presence categories.

To investigate whether human anxiety against

robots affects the priming effect, we conducted a one-

way ANOVA with the RAS factor. Fig. 4(a) shows

the results of the social presence CoP ratings and Fig.

4(b) shows the results of the Atn ratings by the two

RAS groups. As a result, the main effect was seen in

the RAS factor in the categories CoP (F=4.182,

p=<.05) and Atn (F=9.281, p=<.01). The low RAS

group rated CoP and Atn significantly higher than the

high RAS group. For the other categories, there were

no main effects among the RAS factor.

(a) (b)

Figure 4: (a) CoP and (b) Atn scores made by the two RAS

groups.

We also conducted a two-factor ANOVA for the

priming factor and the RAS factor. Fig. 5, 6, and 7

show the questionnaire results of MsgU, Emo, and

Aff respectively. As a result, there were significant

interactions between the priming factor and the RAS

factor in MsgU (F=4.389, p=<.01), Aff (F=7.467,

p=<.01), and Emo (F=5.502, p=<.01).

In the low RAS group, they rated MsgU

significantly lower in the robot condition (F=4.553,

p=<.05) and the agent condition (F=4.553, p=<.01)

than in the control condition. They rated Aff

significantly lower in the robot condition than in the

agent condition (F=7.202, p =<.01) and the control

condition (F=7.202, p=<.01). They rated Emo

significantly lower in the robot condition than in the

control condition (F=3.801, p=<.01).

In the high RAS group, they rated MsgU

significantly lower in the robot condition than in the

agent condition (F=3.828, p=<.01) and the control

condition (F=3.828, p=<.05). They rated Emo

significantly lower in the robot condition (F=5.177,

p=<.01) and the control condition (F=5.177, p=<.01)

than in the agent condition. There was no significant

difference between the priming conditions in the

evaluation of the RAS high group in Aff.

In the Behv category, which indicates whether the

behaviors of the primed agent affect the participants,

there were no significant differences in the main

Effects of Agents’ Embodiment and Robot Anxiety Scale on Social Priming

269

effects or interactions on/between priming factor and

RAS factor.

Figure 5: MsgU scores made by the two RAS groups.

Figure 6: Emo scores made by the two RAS groups.

Figure 7: Aff scores made by the two RAS groups.

In summary, the main effects of the priming factor

were seen in MsgU and Emo. The embodiment on the

priming side affected a part of the social presence of

the primed agent. However, contrary to the

hypothesis, evaluations of the social presence of the

primed agent became significantly higher when there

was no priming and no physical embodiment on the

priming side. Moreover, CoP and Atn showed the main

effects of the RAS factor. The low RAS group rated the

above categories as significantly higher than the high

RAS group.

In terms of interactions between the priming factor

and the RAS factor, the low RAS group rated the robot

condition as the lowest and the control condition as the

highest in MsgU, Emo, and Aff categories, while the

high RAS group rated the control condition and the

agent condition as significantly higher than the robot

condition in the above categories.

4 DISCUSSION

The embodiment on the priming side affected two

categories in the social presence of the primed agent.

There were no significant differences in the other

social presence categories among the priming

conditions. The hypothesis that "the social presence of

the primed agent becomes higher when the embodied

robot primes than when the agent primes" was not

supported. On the contrary, the social presence of the

primed agent was perceived lower when the robot

primed than when the virtual agent did and when there

was no priming. We interpret the reason for this result

as follows: 1) the experimental environment where the

embodied robot interacted with the virtual agent

emphasized the different embodiments of the robot and

the agent, 2) the robot size is 307(H) ×180(W) ×130(D)

mm, and it is smaller than the agent, which is displayed

on the 40 inch screen. 3) Thus, it decreased the

believability and credibility of the robot and the game

environment itself where the robot and the agent play

the twenty questions game.

CoP and Atn ratings showed the main effects

between the RAS groups. The above scores were rated

significantly higher in the low RAS group than the high

RAS group. This result suggests that we can apply the

high RAS group's negative and anxious attitude toward

robots to the virtual agent.

The interactions between the priming factor and the

RAS factor in MsgU, Emo, and Aff indicate that there

is a tendency that the low RAS group rated the robot

condition as the lowest and the control condition as the

highest in the above categories, while the high RAS

group rated the control condition and the agent

condition as significantly higher than the robot

condition.

We interpret these results as follows: The low RAS

group can perceive high social presence of the game-

playing agent without any priming, because of their

nature of having low anxiety toward robots. They

perceived the unnaturalness of the experimental

Control

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

270

environment where the robot played the twenty

questions game with the virtual agent and have

opinions about it. Thus, their ratings in the categories

of MsgU, Emo, and Aff, which measure one's effort to

understand the other party's thoughts, feelings, and

moods, decreased. In other words, priming by the robot

resulted in decrease of the social presence of the game

playing agent for the low RAS group.

The high RAS group has high anxiety against

robots. This might have led to their low social presence

ratings in the robot condition. However, their social

presence ratings of the game-playing agent were as

high as the ones in the control condition or even higher.

Thus, they perceived higher social presence from the

game-playing agent when the other agent gave priming

information.

In summary, the overall results suggest that the

social presence of the primed agent becomes higher

when the other agent performs the priming for the high

RAS group, while the social presence of the primed

agent becomes lower when the robot performs the

priming for the low RAS group. One of the limitations

of this experiment is that the number of participants

assigned to each condition and RAS group is not even

and there are a limited number of participants. Further

experiment should gather a larger number of

participants.

In terms of other limitations, further research

should use a life-sized agent and robot. Daher et al.

2017 used a life-sized agent for the priming agent, and

they implemented an animation of it leaving the spot

soon after it gave the priming information to the

participant. The robot and agent used in this

experiment were not life-sized, and we did not

implement animations such as leaving the place;

therefore, the actual perceived presence of the robot or

agents was lower than the experimental setting in

Daher et al. 2017. Hence, there is a possibility that the

priming effect would be more noticeable if life-sized

robot and agents were used. Moreover, the

embodiment of the primed side should be considered

in a further experiment. This experiment used a virtual

agent for the primed side, but a further experiment

should switch the primed side from the agent to the

robot.

5 CONCLUSION

This research investigated the effect of embodiment of

the priming agent on the perception of social presence

of the primed agent. The preliminary results did not

support our hypothesis that "the social presence of the

primed agent becomes higher when the embodied

robot primes than when the virtual agent primes."

However, the results indicated that there is a dichotomy

in the perceived social presence between the

participants' groups when we divide them according to

their anxiety level toward robots. This indicates that the

priming effect on the social presence of the primed

agent is different depending on the embodiment of the

priming agent and people's anxiety toward robots.

Thus, we should consider the effects of the

embodiment of the agent and people's attitude toward

robots on the social presence of agents.

ACKNOWLEDGEMENTS

This research was partially supported by JSPS

KAKENHI, JP17K00287.

REFERENCES

Bainbridge, W. A., Hart, J., Kim, E. S., & Scassellati, B.,

2008. The effect of presence on human-robot interaction.

In Robot and Human Interactive Communication, (RO-

MAN2008), 701-706.

Bargh, J.A., Chen, M., and Burrows, L., 1996. Automaticity

of social behavior: Direct effects of trait construct and

stereotype activation on action. Journal of Personality

and Social Psychology 71(2), 230-244.

Daher, S., Kim, K., Lee, M., Schubert, R., Bruder, G.,

Bailenson, J., and Welch, G., 2017. Effects of Social

Priming on Social Presence with Intelligent Virtual

Agents. In Proc. of Intelligent Virtual Agents 2017, 87-

100.

de Greef, P., and IJsselsteijn, W. A., 2001. Social presence in

a home tele-application. Cyber Psychology and Behavior

4, 307-315.

Hall, E.T., 1966. The hidden dimension, Doubleday and

Company.

Harms, C., Biocca, F., 2004. Internal consistency and

reliability of the networked minds measure of social

presence: Seventh Annual International Workshop:

Presence 2004.

Kiesler, S., Powers, A., 2008. Anthropomorphic interactions

with a robot and robot-like agent. Social Cognition. 26,

169-181. doi:10.1521/soco.2008.26.2.169.

Nomura, T., Suzuki,T., Kanda, T., and Kato, K., 2006.

Measurement of anxiety toward robots. In Proc.

IEEE Int.

Symp. Robot and Human Interactive Communication,

pp.372–377.

Pan, Y. and Steed, A., 2016. A comparison of avatar, video,

and robot-mediated interaction on users’ trust in

expertise. Frontiers in Robotics and AI 3, 12.

Powers, A., Kiesler, S., Fussell, S., and Torrey, C., 2007.

Comparing a computer agent with a humanoid robot. In

Proceedings of Human-Robot Interaction (HRI2007),

145-152.

Effects of Agents’ Embodiment and Robot Anxiety Scale on Social Priming

271