Investigating the Social Influence of Different Humanoid Robots

Full text

(1)

Linköping University | Department of Computer Science

Bachelor thesis, 18 ECTS | Cognitive Science

Spring term 2017 | LIU-IDA/KOGVET-G--17/029--SE

Investigating the Social

Influ-ence of Different Humanoid

Robots

Sofia Thunberg

Supervisor : Shahram Moradi Examiner : Felix Koch

(2)

Upphovsrätt

Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare – under 25 år från publiceringsdatum under förutsättning att inga extraordinära omständigheter uppstår. Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för ickekommersiell forskning och för undervisning. Överföring av upphovsrätten vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av dokumentet kräver upphovsmannens medgivande. För att garantera äktheten, säkerheten och tillgängligheten finns lösningar av teknisk och admin-istrativ art. Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i den omfattning som god sed kräver vid användning av dokumentet på ovan beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan form eller i sådant sam-manhang som är kränkande för upphovsmannenslitterära eller konstnärliga anseende eller egenart. För ytterligare information om Linköping University Electronic Press se förlagets hemsida http://www.ep.liu.se/.

Copyright

The publishers will keep this document online on the Internet – or its possible replacement – for a period of 25 years starting from the date of publication barring exceptional circum-stances. The online availability of the document implies permanent permission for anyone to read, to download, or to print out single copies for his/hers own use and to use it unchanged for non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional upon the con-sent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security and accessibility. According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement. For additional information about the Linköping Uni-versity Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its www home page: http://www.ep.liu.se/.

c

(3)

Abstract

The aim with this thesis were to investigate social influence of the two humanoid robots, NAO and Pepper. The research questions were if there were a differ-ence in human social acceptance, in social infludiffer-ence and in infludiffer-ence on human decision making between NAO and Pepper.

To answer these questions, an experiment using the Wizard of Oz-method were used with 36 participant, 18 in each group, interacted with NAO or Pepper. Afterwards two questionnaires, NARS and GODSPEED, were answered and an additional interview were held with the participants.

The result showed a significant difference on GODSPEED, where NAO in-dicates to have a higher amount of social influence on the participants then Pepper. The result for NARS were not significant. The result from the de-cisions made during the experiment indicated that humans follow NAO more than Pepper, a result that got more explained and understandable during the interviews.

For future studies there would be interesting to test the scenario with a larger selection and also with a more natural Wizard of Oz-design.

(4)

Acknowledgments

I would like to thank Arne J¨onsson for guiding me in the right direction regarding my thesis subject when all I knew I wanted to do something with robots. Also a big thank you to my external supervisors, Tom Ziemke and Sam Thellman, for taking me in with open arms and helping me develop an exiting project. I would also like to direct a special thank you to Sam Thellman for all your help and enthusiasm during this project and for teaching me so much about statistics when I needed it the most.

(5)

Contents

1 Introduction 1 1.1 Aim . . . 1 1.2 Research questions . . . 1 2 Theory 2 2.1 Human-Robot Interaction . . . 2

2.2 Robots Effect on Humans . . . 2

2.2.1 Social Influence . . . 4

2.3 Method . . . 4

2.3.1 Wizard of Oz . . . 4

2.3.2 The NARS and GODSPEED Questionnaires . . . 5

2.3.3 Thematic Analysis . . . 5 3 The Robots 6 3.1 NAO . . . 6 3.2 Pepper . . . 7 4 Method 8 4.1 Procedure . . . 8 4.1.1 The Experiment . . . 8 4.1.2 The Questionnaires . . . 9 4.1.3 The Interview . . . 9 4.2 Material . . . 10 4.3 Measurements . . . 10 4.3.1 Decisions . . . 10 4.3.2 Interviews . . . 11 4.4 Participants . . . 11 4.5 Analysis . . . 12 5 Results 13 5.1 Decisions . . . 13

(6)

5.4.1 Appearance . . . 17 5.4.2 Interaction . . . 19 5.4.3 Mood . . . 19 5.4.4 Decision . . . 19 6 Discussion 20 6.1 Results . . . 20 6.2 Method . . . 21

6.3 The work in a wider context . . . 23

7 Conclusion 24 References 25 Appendix 27 A NARS . . . 27

(7)

List of Tables

5.1 Bar chart showing the percentage of the participants that followed what NAO and Pepper said, in part 1 and 2. . . 14 5.2 The median and quartile range for NAO and Pepper in the

ques-tionnaires NARS and GODSPEED. . . 15 5.3 Bar chart showing the median rated score for NARS and

GOD-SPEED, for NAO and Pepper. . . 16 5.4 The words that were used to describe NAO. . . 17 5.5 The words that were used to describe Pepper. . . 17

(8)

List of Figures

2.1 The Uncanny Valley. . . 3

3.1 NAO. . . 6

3.2 Pepper. . . 7

4.1 The experiment scene with NAO. . . 9

4.2 The experiment scene with Pepper. . . 10

4.3 The sentences NAO and Pepper said during the experiment and the different outcomes of the decisions that where made, divided into part 1 and part 2. . . 11

5.1 The words that was used to describe NAO and Pepper, divided in positive, negative and neutral words. . . 18

(9)

1

Introduction

Robots are becoming a bigger part of our every day life throughout the world. Humanoid robots are developed to be able to help patients and assist the sick in residential care homes (Dautenhahn, Campbell, & Syrdal, 2015), to educate children at home (Tanaka et al., 2015) and to assist in therapy and education of children with autism (Robins, Dautenhahn, Boekhorst, & Billard, 2005). For the humanoid robots to be able to help humans with these things, the developers are faced with a lot of challenges and one of the biggest problems to solve is how one build a robot that is socially accepted by humans. In the research field social robotics, the emergence of studies about robot height, closeness, voice, face and many other aspects about how the robot appears, behave and react have led to a greater understanding on what humans think about different robots.

In the recent years, several studies have focused on the social acceptance of robots (Nomura, Syrdal, & Dautenhahn, 2015). Many of these studies have focused on questionnaires and online surveys and there is an absence of exper-iments where participants interact with real robots in this research field (Rau, Li, & Li, 2010).

1.1

Aim

This thesis had the aim to examine human interaction with humanoid robots with a focus on social attitude and social influence. Further this work wanted to examine how humanoid robots could affect human decision making.

1.2

Research questions

The research questions are:

- Is there a difference in human social acceptance between NAO and Pepper? - Is there a difference in social influence between NAO and Pepper? - Is there a difference in influence on human decision making between NAO and Pepper?

(10)

2

Theory

2.1

Human-Robot Interaction

Human-Robot Interaction (HRI) research shows that robots which behave more naturally are more likely to become socially accepted by humans (Dautenhahn et al., 2005). Modeling the robots behavior to be acceptable by humans presents a great challenge since it is dependent on a combination of verbal and non-verbal expressions (Obaid et al., 2016).

2.2

Robots Effect on Humans

Hiroi and Ito (2009) conducted a study where they investigated the psychological effect on different ”robot sizes”, which in this case only were robot height and not width. They tested three different robot sizes, 600, 1200 and 1800 mm, which covered the standard robot size of 1200 mm and those that differ 600 mm, both taller and shorter. The result showed that a robot size of 1800 mm made the participants feel threatened and that there was no perceived psychological threat between the 1200- and 600-mm-tall robot when the participants were standing, but that the 1200-mm-tall robot was preferable when the participants were standing and the 600-mm-tall robot when they where seated

Huang, Olson, and Olson (2002) found that there was a significant difference between the individual influence on decision-making between ”tall” and ”short” people.

Walters et al. (2006) argues that robots that are aware of their social space increase the social acceptance by humans. Through a study, they found that the participants preferred a robot approaching from the left or the right side and that they disliked a frontal approach.

Sandoval, Brandstetter, and Bartneck (2016) argues that reciprocity is a cor-nerstone in human relationships and that it will play a principal role in HRI in the future. They studied this through an experiment where the robot tried to bribe a human in a decision making game. The result showed that humans tend to be reciprocal towards robots when they act for the benefit of the humans. However, the participants reciprocated less with a bribing robot than with a honest robot, though they tend to like the bribing robot and its unexpected

(11)

behaviour more (Sandoval et al., 2016). Mubin and Bartneck (2015) studied if unpredictable behaviour by robots could be used as a tool to increase the per-ceived anthropomorphism and the robot personality to increase the HRI experi-ence. Previous studies had been based on subjective questionnaires without real HRI that have reported that humans feel threatened by unpredictable robots. Mubin and Bartneck predicted that humans would be more tolerant against predictable humanoid robots compared to unpredictable humanoid robots.

Mori (2012) have noticed that in the goal of making robots appear like a human, one’s affinity for them increases until one come to a valley, which Mori calls the uncanny valley (see figure 2.1). He exemplifies this with that industry robots is clearly based on functionality and does not look like human beings, therefore people do not feel any affinity for them. In contrast, a toy robot that start to have a roughly human-looking external form with a face, two arms and two legs, can make children deeply attached to them. Mori describes that one of the objective with robotics is to create an artificial human. By covering metal cylinders with skin the robot can get a more human-like appearance, and as a result one naturally respond to it with a heightened sense of affinity. For example the development of prosthetic hands have come far where the hand can have wrinkles, veins and fingernails. One might not notice the difference between a prosthetic hand and a real hand before shaking it, and then one can be startled during a handshake with a boneless grip and cold skin. When this happens one lose the sense of affinity and the hand becomes uncanny. Before the uncanny valley lays the humanoid robot and the challenge is to develop a human-like robot that does not become uncanny. Therefore, Mori recommends to build a robot with its goal in the first peak, and not the second.

(12)

2.2.1

Social Influence

Milgram (1963) conducted a famous study of obedience in psychology where the participants were paired two and two, where one was the ”learner” and the other the ”teacher”. The teacher was the real participant and the learner was an actor. The teacher had to give the learner an electric shock if he or she failed to learn. The aim of the study was to see how far people would go in obeying an instruction if it involved harming another person, and compared the outcome if a experimenter stood beside the teacher, encouraging the teacher to harm the learner or a voice over the phone doing the same thing. The result showed that the experimenter had a higher level of authority and that the participants had a hard time to go against he’s demands (Milgram, 1963).

2.3

Method

In this section the methods Wizard of Oz, the NARS and GODSPEED ques-tionnaires and thematic analysis are presented.

2.3.1

Wizard of Oz

When the participants in a experiment situation think that they are talking to a computer, though they in fact are not, and the conversation is instead mediated by a human operator (the so-called wizard), it is called a Wizard of Oz-design (WoZ). According to Dahlb¨ack, J¨onsson, and Ahrenberg (1993) there are several reasons for wanting to conduct WoZ-experiments. One is that computers are rigid and that people are flexible. In an experiment when the participant interact with a computer, the situation might change quickly in an unexpected way. If the experiment relies on a static system that cannot change then the situation will not seem natural to the participant, and therefore it is better with a wizard that can adjust (Dahlb¨ack et al., 1993).

Riek (2012) reports that WoZ is a common employed technique in HRI, where the wizard are controlling the robot on a number of things, such as the robots movements, navigation, speech and gestures. Researchers mainly uses WoZ in HRI because robots are not sufficiently advanced to interact au-tonomously with people in socially appropriate ways, and this method make it possible for the participant to envision what future interaction could be like. Riek say that one methodological concern regarding WoZ is that one can argue that it is not really human-robot interaction so much as human-human interac-tion that is mediated through a robot. Another concern is the ethical problems that arise when the participant cannot tell whom they are interacting with, the robot or the test leader. Miller (2010) refers to this as the ”Turing Deceptions”, and since 2011 when EPSRC drafted ethical rules for robots it is stated that the users should always be able to ”lift the curtain” since robot intelligence is cur-rently an illusion (EPSRC, 2011). A problem from a engineering perspective is according to Breazeal, Kidd, Thomaz, Hoffman, and Berlin (2005) that WoZ can

(13)

make it more difficult to build robots capable of successfully mitigation errors on their own in the future. One way to mitigate these methodological, ethical and engineering concerns, according to Riek (2012), is to design experiments that employ WoZ in a rigorous and repeatable manner. This to make a smoother transition later to a more autonomous system in the future. He proposes a list with guidelines that researchers using WoZ in HRI studies should be able to answer. The list contains the categories robot, user, wizard and general and for example wants the answer to how many robots where used, was the simulation convincing to the user and how did the experimenter control for wizard error.

2.3.2

The NARS and GODSPEED Questionnaires

According to Turner (1993), Likert-scale questionnaires are often used to mea-sure characteristics, attitude and opinions. They often contain a five or seven graded ordinal scale that yields from strongly disagree (1) to strongly agree (5 or 7). Both the questionnaires used in this study contain a five graded Likert-scale. Negative Attitude toward Robots Scale (NARS) is a questionnaire that mea-sure the participants general attitude against robots (see Appendix A). It con-tains 14 questions where three questions are reversed. The score is divided into three subsections, 1) negative attitude toward situations of interaction with robots, 2) negative attitude toward social influence of robots, 3) negative at-titude toward emotions in interaction with robots. The version that was used was translated to Swedish by (Thellman, Silvervarg, Gulz, and Ziemke (2016).

GODSPEED is a questionnaire with five sections that contain anthropomor-phism (which means making something that is not human, more human-like), animacy (which means something that seems sentient or alive), likeability, per-ceived intelligence and perper-ceived safety (see Appendix B). The first four sections are filled out based on the participants impression of the robot and the last sec-tion are filled out based on the participants emosec-tional state when he/she met the robot. It was translated to Swedish by using the English version to make comparisons (Bartneck, 2008).

2.3.3

Thematic Analysis

Boyatzis (1998) describe thematic analysis as a method that is used to assist the researcher in the search for insight. More specific it is a process for encoding qualitative information, that usually require an explicit code, e.g. a list of themes. Boyatzis explain a theme as a pattern found in the data.

(14)

3

The Robots

The robots that were used were a NAO and a Pepper humaniod robot using NAOqi 2.0 OS.

3.1

NAO

NAO can be seen in figure 3.1 on the following page. NAO has a height of 574 mm (smaller than an average robot) and weights 4.3 kg. The robot is equipped with two HD cameras, four microphones, sonar range finder, two infrared emit-ters and receivers, inertial board, nine tactile sensors and eight pressure sensors.

(15)

3.2

Pepper

Pepper can be seen in figure 3.2 on the following page. Pepper is 1200 mm tall (average robot height) and weights 28 kg. The robot is equipped with a gyro sensor and a 10.1-inch touch display on its chest and 4 microphones, two RGB cameras, one 2D sensor and three touch sensors on its head. The robots hands have two touch sensors and its legs has two sonar sensors, six laser sensors, three bumper sensors and one gyro sensor.

(16)

4

Method

In this chapter the procedure, the material, the participants and analysis will be presented.

4.1

Procedure

The study was conducted in a three step way. In the first step, an experiment with NAO and Pepper was performed. Part 1 was that the test leader asked the participant to go inside a room and leave a book on the table. In the room, NAO stood on the table (see figure 4.1) or Pepper stood on the floor beside the table (see figure 4.2). The robot would ask the participant to not put the book on the table and to go back with it to the test leader. If the participant came back to the test leader, she would insist that the book should be on the table inside the room. Part 2 started if the participant returned. Then the robot would ask him or her to throw the book in the trash can. This experiment had the aim to show if the participants follow directives from humans or robots, or both, and if robots can influence decision making. It would further examine if there was any differences between the two robots.

In the second step, the participants got to fill in the NARS questionnaire to get an overview about the participant’s attitude towards the robot, and the GODSPEED questionnaire to measure social attitude towards the robot.

In the third step, a semistructured interview with open ended questions were held to explore additional information about how the participant perceived the robot. The main aim of the interview were to examine why the participants made the decisions they did and what they thought and felt about how the robot acted and talked.

4.1.1

The Experiment

The experiment was conducted with a WoZ-design, and during the experiment the test leader stood outside the experiment room with the curtains closed in a way that the participant did not see the test leader. The test leader had the robots head camera active to be able to follow what happened in the room. If the participant did not respond the robot it continued to force the conversation.

(17)

Figure 4.1: The experiment scene with NAO.

The robot had 12 fixed sentences in part 1 and 15 fixed sentences in part 2 (see figure 4.3).

These sentences were what the robot said during the experiment, but not all sentences was used every time if they did not fit the natural conversation.

4.1.2

The Questionnaires

The questionnaires (NARS and GODSPEED) were filled out by the participants after the experiment, with the aim that the participants would be primed by the robot they just met.

4.1.3

The Interview

The semistructured interview took place in the same room as the experiment but with no robot present. It was recorded and consisted of open ended questions about the general experience of the robot and why the participants took the decisions they did. The questions was open ended and a couple of examples are: ”What was your initial reaction when you saw the robot?” and ”What felt best,

(18)

Figure 4.2: The experiment scene with Pepper.

4.2

Material

The robot interaction were programmed in Python 2.7.12 and Monitor 2.1.4 was used during the experiment. The analysis was made i IBM SPSS Statistics Data Editor.

4.3

Measurements

The following measurements were used in relation to the experiment. They are described below alongside the interview questions.

4.3.1

Decisions

The decisions the participants made were logged manually by the experiment leader, and there were five possible outcomes (see figure 4.3). The first outcome was who stopped after part 1 and left the book with the robot during the first part. The second outcome was to stop after part 1 and leave the book with the test leader. The third outcome was those who chose to leave the book with the robot on the table during part 2 and the fourth outcome were those who left

(19)

Figure 4.3: The sentences NAO and Pepper said during the experiment and the different outcomes of the decisions that where made, divided into part 1 and part 2.

the book in the trash can during part 2. The last and fifth outcome included the participants that took the book with them at the end of part 2.

4.3.2

Interviews

The interview were recorded and were then analyzed by listening trough them while taking notes, and a thematic analysis were conducted.

4.4

Participants

Group 1, that did the experiment with NAO, had 18 participants (M =27 years old, SD=9,27, 67%=female) and group 2, that did the experiment with Pepper,

(20)

they met and basic English knowledge. The participants were recruited from Link¨oping University and before each session the participants got to fill a consent form and was informed that they could end the experiment session at any time. The participants where randomly placed in group 1 and 2.

4.5

Analysis

To answer if there is a difference in human social acceptance and/or social influence between NAO and Pepper, a Mann-Whitney U-test was conducted on the result from the GODSPEED and the NARS questionnaires, along with a Chronbach’s α test to estimate the reliability on the internal consistency on both surveys. The Mann-Whitney U-test were chosen over a Independent t-test, because it handles nonparametric data and are suitable for ordinal numeric data, like Likert scale McKnight and Najab (2010).

To answer if there is a difference in influence on human decision making between NAO and Pepper, a Fisher’s Exact test was conducted to compare the decisions during the first and the second part, reported along with the percentage of the decisions made. The percentage is relevant to report since binary data often need a larger sample size to give a significant result, and therefore the percentage can indicate a pattern in the result. Also a thematic analysis of the transcriptions from the interviews were made and coded in a way that focused on the describing words that where used about the robot and why the participants took the decisions they did during the experiment.

(21)

5

Results

This chapter will present the results from the experiment with the different deci-sions that were made, the date from the NARS and GODSPEED questionnaires, and also analysis from the interviews.

5.1

Decisions

83.33% of the participants in the NAO condition chose to follow what NAO commanded (to return the book to the human outside of the room) over what the test leader said (to put the book on the table) in the first part. In part 2, 60.00% of the remaining participants chose to follow what NAO said (to throw the book in the trash can) over what the test leader said (to put the book on the table). This can be seen in table 5.1.

50.00% of the participants in the Pepper condition chose to follow what Pepper commanded (to return the book to the human outside of the room) over what the test leader said (to put the book on the table) in the first part. In part two, 44.44% of the remaining participants chose to follow what Pepper said (to throw the book in the trash can) over what the test leader said (to put the book on the table). This can be seen in table 5.1.

This result suggests that a larger selection of the participants followed what NAO said over what Pepper said. It further shows that participants were less likely to follow what Pepper said in the first part than they were to NAO.

A Fisher’s Exact test was calculated comparing the decisions during the first part between NAO and Pepper, and a trend towards a significant difference was found p=.075. During the second part between NAO and Pepper no significant difference was found p=.675.

5.2

Negative Attitude Toward Robots Scale

A Mann-Whitney U-test was conducted on participants’ self-assessed negative attitude toward robots (see table 5.2). The result were no statistically signifi-cantly different in the two groups, with participants that met NAO rating

(22)

them-Table 5.1: Bar chart showing the percentage of the participants that followed what NAO and Pepper said, in part 1 and 2.

These results suggest that there is no difference between negative attitude to-ward NAO and Pepper (see table 5.3).

A Mann-Whitney-U-test was conducted on participants’ self-assessed nega-tive attitude toward situations of interaction with robots. The result showed a trend towards a statistically significantly difference on the two conditions, NAO and Pepper, U =103.5, z=-1.876, p=.061. These results suggest that there is a difference in negative attitude in situations of interaction between NAO and Pepper.

A Mann-Whitney-U-test was conducted on participants’ self-assessed nega-tive attitude toward social influence of robots. The result were not statistically significantly different on the two condition, NAO and Pepper, U =146, z=-.512, p=.609. These results suggest that there is no difference in negative attitude regarding social influence between NAO and Pepper.

A Mann-Whitney-U-test was conducted on participants’ self-assessed neg-ative attitude toward emotion in interaction with robots. The result were not statistically significantly different on the two condition, NAO and Pepper, U =150.5, z=-.370, p=.712. These results suggest that there is no difference in negative attitude regarding emotions in interaction between NAO and Pepper.

(23)

Table 5.2: The median and quartile range for NAO and Pepper in the question-naires NARS and GODSPEED.

NAO Mdn QR

NARS 2 2 - 3

Negative attitude toward situations of interaction with robots 2 1 - 2 Negative attitude toward social influence of robot 3 2 - 4 Negative attitude toward emotion in interaction with robot 3 2 - 3

GODSPEED 4 3 - 4 Anthropomorphism 3 2 - 3 Animacy 3.5 3 - 4 Likeability 4 3 - 4 Perceived Intelligence 4 3 - 4 Perceived Safety 4 3 - 4 Pepper Mdn QR NARS 3 2 - 3

Negative attitude toward situations of interaction with robots 2 1 - 2 Negative attitude toward social influence of robot 3 2 - 4 Negative attitude toward emotion in interaction with robot 3 2 - 3

GODSPEED 3 2 - 4 Anthropomorphism 3 2 - 3 Animacy 3 3 - 4 Likeability 3 2 - 4 Perceived Intelligence 3 3 - 4 Perceived Safety 3 2 - 4

5.3

GODSPEED

A Mann-Whitney U-test was conducted on participants’ self-assessed social ac-ceptance against robots (see table 5.2). The result were statistically significantly different in the two groups, with participants that met NAO rating higher than participants that met Pepper, U =67, z=3.009, p=.003. These results suggest that NAO has a higher level of social influence than Pepper (see table 5.3).

On the variable anthropomorphism, the result between NAO and Pepper were not statistically significantly different in the two conditions, NAO and Pepper, U =120.5, z=-1.324, p=.186. These results suggest that there is no difference regarding anthropomorphism between NAO and Pepper.

(24)

macy than Pepper.

On the variable likeability, the result between NAO and Pepper were statis-tically significantly different on the two conditions, NAO and Pepper, U =83.5, z=-2.506, p=.012. These results suggest that NAO has a higher level of like-ability than Pepper.

On the variable perceived intelligence, the result between NAO and Pepper were statistically significantly different on the two conditions, NAO and Pepper, U =95, z=-2.132, p=.033. These results suggest that NAO has a higher level of perceived intelligence than Pepper. the variable perceived safety, the result between NAO and Pepper were statistically significantly different on the two conditions, NAO and Pepper, U =100, z=-1.982, p=.047. These results suggest that NAO has a higher level of perceived safety than Pepper.

Cronbach’s α for GODSPEED was calculated to examine internal consis-tency, with a total score of .875, and for the subsections; anthropomorphism=.663, animacy=.729, likeability=.735, perceived intelligence=.826 and perceived safety=.612. This result is an acceptable high α score.

Table 5.3: Bar chart showing the median rated score for NARS and GOD-SPEED, for NAO and Pepper.

5.4

Interviews

The result from the thematic analysis resulted in three themes, appearance, interaction and mood. At a first stage, the result from the interviews with par-ticipants that met NAO, were the words that was used to describe the robot,

(25)

divided under the three themes (see table 5.4). All describing words and cita-tions have been freely translated from Swedish to English.

Table 5.4: The words that were used to describe NAO. Appearance Interaction Mood

Unnatural Good Determined

Friendly Pleasant Own will

Natural Prosody Nice

Lively eyes Awesome Stubborn Disarmed Human-like Kind

Nice Mean Unpleasant

Alive Calm Assertive

Small Clear Cocky

Cute Child-like

Nifty Fun

The words that was used to describe Pepper, divided under the three themes can be seen in table 5.5.

Table 5.5: The words that were used to describe Pepper. Appearance Interaction Mood

Human-like Weird Determined

Inorganic Uncomfortable Own will

Nice Good Unsure

Sweet Rigid Emotional

Shaky Scary Attitude

Robot-like Human-like Observant

Apathetic Foreign Unkind

Small Present Strained

Child-like Ridiculous Stubborn

Sleek Involved Sassy

Unnatural

The words that bring NAO and Pepper together (and apart) can be seen in figure 5.1. The words are divided into categories of positive, negative and neutral loaded words about the robots.

(26)

Figure 5.1: The words that was used to describe NAO and Pepper, divided in positive, negative and neutral words.

The participants that met NAO thought the robot looked like ”a small man”, with its arms and legs that copied a human. The general opinion was that NAO in some way was nifty, nice and looked friendly. These are all positive words that made the participants feel safe entering the experiment room. One participant further described that the robot had a disarming look, like ”it does not seem to want to harm me, instead it is quite cute”. As a result of NAO’s face recognition the participants thought the robot had ”lively eyes” that followed them around in a natural way.

The participants that met Pepper had more inconsistent opinions about the robots appearance. They both thought it looked human-like, robot-like and child-like, as well as inorganic and unnatural. Some participants thought Pepper was a bit shaky in its movements and some thought the robot appeared like it was in a apathetic state. One participant said ”it just stood there, starring at the roof”. Some more positive words that were used to describe Pepper was that the robot looked sweet and nice.

(27)

5.4.2

Interaction

Under the theme interaction, the participants talked about how the robot acted during their talk.

The participants that met NAO experienced the interaction as pleasant and fun, and that the robot gave a calm impression. Some of the participant found it pleasant that the robot could talk and respond in a natural way, also that it could follow faces and look the participant ”straight in the eyes”. One participant explained that ”it sounds like the robot is talking with prosody, which makes the interaction easier”.

The participants that met Pepper thought that the interaction was uncom-fortable and foreign, but also that the robot seemed present and involved in the conversation. Some participants expressed that they were unsure what the robot would do, that the robot did sometimes look them in the eyes and then suddenly looked in another direction in a unpredictable way. One participant explained that ”it was not much of an interaction. It observed its environment and just talked”.

5.4.3

Mood

Under the theme mood, the participants talked about how the robots frame of mind were perceived.

The participants that met NAO thought that the robot appeared to be determined and stubborn, like it had its own will. One participant explained further that ”the robot was so on to me (about the book), but humans are like that too. It must also have its own will”, which indicates that the participants afforded the robot human-like feelings. Some participants experienced NAO as cocky and assertive, but in a nice a way.

The participants that met Pepper also thought that the robot appeared to be determined and stubborn, like as it had its own will. They also thought that the robot appeared to be emotional, observant and strained. Some participants expressed that Pepper seemed to have attitude and that it acted in a sassy way.

5.4.4

Decision

Those who chose to follow what the test leader said over NAO, explained it by that a robot can be turned off, and that you will not be held as much morally accountable if you do not follow what the robot says than what you will be if you refuse the humans commands. The participants that followed what NAO said explained their behavior with that they did not want to hurt the robot and wanted to make it happy and pleased with them.

The participants that followed what the test leader said over Pepper ex-plained that the human had greater authority over the robots wishes, but some said that they would had followed the robot if it had showed more emotions.

(28)

6

Discussion

This chapter contains the discussion of results, methods and the work in a wider context.

6.1

Results

The result showed that there was a trend toward a significant difference between NAO and Pepper on the outcome of the decisions in part 1, but that there were no significant difference in part 2. There were also no significant difference between negative attitude toward NAO and Pepper, but that there were a sig-nificant difference in social influence between NAO and Pepper. The thematic analysis showed that there were a difference between NAO and Pepper.

The main result of this thesis were that NAO got a higher rating than Pep-per concerning almost all of the parts of the study. Regarding research question one, is there a difference in human social acceptance between NAO and Pepper, the main part of the answer were the NARS questionnaire. It measured neg-ative attitude towards robots in general, along with negneg-ative attitude towards situations of interaction with robots, negative attitude towards social influence of robots and negative attitude towards emotions in interaction with robots. There were not a significant result, which indicates that the participants did not got primed to have less negative attitude toward robots on either of the conditions.

Regarding research question two, is there a difference in human social influ-ence between NAO and Pepper, the main part of the answer were the GOD-SPEED questionnaire. It measured the general opinion about a robot, regarding anthropomorphism, animacy, likeability, perceived intelligence and perceived safety. There was a significant result where NAO (M =3.42) had a higher score than Pepper (M =2.95) had, which indicates that NAO had a greater social influence on the participants.

Regarding research question three, is there a difference in influence on hu-man decision making between NAO and Pepper, the two parts of the answer were the experiment and the following interviews. The decisions were reported by the proportions that the participants followed the robots or the test leaders commands. For the first part, 83% followed what NAO said and 50% followed what Pepper said. For the second part, 60% followed what NAO said and 44%

(29)

followed what Pepper said. This indicates that NAO had a higher influence on human decision making than Pepper has. One could see this tendency though the result were not significant, because one needs a larger sample size (at least 30 participants in each group) to get a significant result with bivariate vari-ables. The result from the experiment were confirmed in the interviews and the thematic analysis. As can be seen in figure 4.1 there are more negative words mentioned about Pepper than about NAO, and there are less positive words than neutral words stated. NAO on the other hand have more positive words stated about it and only two negative words. NAO and Pepper does not share any negative words and a few positive and neutral words together. Figure 5.1 gives a good overview over what the participants thought about NAO and Pepper and it clearly states what the experiment result indicated, that the par-ticipants followed NAO to a higher degree and it also explains why, because they had more positive feelings about NAO, which also is showed through NARS and GODSPEED.

This result opposes Hiroi and Ito (2009), where the 1200 mm tall robot was preferable when the participants were standing. In this study the 600 mm tall robot (NAO) was preferable over the 1200 mm tall robot (Pepper). The participants reported that NAO looked disarmed and that Pepper was more unpredictable and scary. This is in line with Mubin and Bartneck (2015) and one possible explanation could be the uncanny valley. Pepper is more alike humans with a more human face and bigger eyes, more advanced arms and hands, along with its height. This combination could have given the participants a lose sense of affinity and yielded closer to an uncanny feeling. Pepper is not as human-like to be in the uncanny valley, but more human-like than NAO to its appearance. NAO’s appearance tendency is closer to a toy robot.

The result could also show that the participants felt a greater reciprocity with NAO than with Pepper. They felt that NAO was friendly and they felt more safe than with Pepper, which Sandoval et al. (2016) had previously showed to be .

Lastly, the result shows that the participants had a higher level of obedi-ence against Pepper than NAO, where the test leader had more authority than Pepper, which is in line with Milgram (1963). In the case for NAO, more partic-ipants made the decision with the outcome to leave the book in the trash can, than they did for Pepper. This indicates that the participants were more obedi-ent to the test leader than to NAO, a result that make the Milgram experimobedi-ent compatible with humanoid robots.

6.2

Method

The method of this study is a between group design, because the participants should not be affected by another robot during the experiment and therefore it

(30)

affect the result in a negative way. A problem with the decision to use a between group design was the small sample size (18 in each group). It would had been preferable to have at least 30 participants in each group for a stronger statistical power.

A problem with the WoZ-method was the ethical difficulties that arose during the experiment. Many participants did not know if the robot was autonomous or if the test leader was a wizard and some expressed an uneasy feeling about that the test leader probably saw what they where doing in the room. This unease was regardless of which robot they met, but it was pointed out that it looked like Pepper had a camera in one of its eyes that blinked red. In other aspects, Riek (2012) guidelines was followed and the participants did not express worries about any other part regarding the experiments method.

A problem with the Likert-scale questionnaires NARS and GODSPEED is the validity, and the issues that arise when analyzing and interpreting the or-dinal data. These main problems with Likert-scales fall into three areas. The first is the limitations that appears when one translate the survey from several languages and the meaning of the different alternatives starts to differ. E.g. the NARS questionnaire was translated from Japanese, to English and finally to Swedish for a previous study by ((Thellman et al., 2016). According to Turner (1993) it is necessary to establish the statistical equivalence of the various parts of the questionnaire. Thellman et al. calculated the Chronbach’s α for NARS and got a result of .895, which is a high score, and for this reason no changes was made for this study.

The second area of concern arises from the time, effort and expertise it takes to develop a questionnaire, and that it not always is better to use an existing one. Turner explains that an existing questionnaire cannot be assumed to be valid and reliable for any other group other than the population for which it was established. This is always a problem with all questionnaires and both NARS and GODSPEED was chosen because of the limited amount of time that this thesis had. In a bigger study there would had been preferably to develop questionnaires more suited for the population in mind.

The third area of concern is that Likert-scale questionnaires yield ordinal data, were the points on the scales are not equidistant. This means that the distant from e.g. 1 (strongly disagree) to 2 (disagree) is not as big emotional jump as from 4 (agree) to 5 (strongly agree). To analyse this type of data one should use nonparametric statistics, like the Mann-Whitney U-test in this thesis, that does not assume a normal distribution of interval data. An additional problem to this is that participants tend to keep themselves in the middle (2-4), and avoid the extremes (1 and 5). This could lead to a misleading result.

A problem with thematic analysis is that the result is subjective, even if the analyst always tries to be as objective as possible. This is a problem with all qualitative methods, according to Boyatzis (1998).

The literature used were all scientific articles or books. The sources did not directly address the research questions stated in this thesis, but were relevant to different parts to explain the result of the experiment, and to the overall theme of the study.

(31)

6.3

The work in a wider context

This thesis is important for the continuing work of understanding how different humanoid robots affect humans and how robots should be designed to be able to live and work together with humans in the future. In the research fields cognitive science and social robotics it is essential to investigate which aspects of a robot is important for social acceptance and social influence, and what have an effect on human decision making. The result of this thesis is a step further. For future studies, it would be interesting to test a different kind of the Wizard of Oz- method that make the robot move and act more natural, to be able to test the limits of the human expectations on robot behavior. Possible research questions could be how does a more natural interaction affect social acceptance, social influence and human decision making? It would also be interesting to conduct experiments with other humanoid robots than NAO and Pepper.

(32)

7

Conclusion

The aim with this thesis were to investigate social influence of the two humanoid robots, NAO and Pepper. The research questions were if there were a difference in human social acceptance, in social influence and in influence on human deci-sion making between NAO and Pepper. To answer these questions, an experi-ment using the Wizard of Oz-method were used, with 36 participant, 18 in each group, interacted with NAO or Pepper. Afterwards two questionnaires, NARS and GODSPEED, were answered and an additional interview were held with the participants. The result showed a significant difference on GODSPEED, where NAO indicates to have a higher amount of social influence on the participants then Pepper. The result for NARS were not significant. The result from the decisions made during the experiment indicated that humans follow NAO more than Pepper, a result that got more explained and understandable during the interviews. For future studies there would be interesting to test the scenario with a larger selection and also with a more natural Wizard of Oz-design.

(33)

References

Bartneck, C. (2008). The Godspeed Questionnaire Series.

Boyatzis, R. (1998). Transforming Qualitative Information- Thematic Analysis and Code Development.

Breazeal, C., Kidd, C. D., Thomaz, A. L., Hoffman, G., & Berlin, M. (2005). Effects of nonverbal communication on efficiency and robustness in human-robot teamwor. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE .

Dahlb¨ack, N., J¨onsson, A., & Ahrenberg, L. (1993). Wizard of Oz Studies-Why and How. , 1–27.

Dautenhahn, K., Campbell, A., & Syrdal, D. S. (2015). Does anyone want to talk to me ? – Reflections on the use of assistance and companion robots in care homes. 4th International Symposium on New Frontiers in Human-Robot Interaction, 1–4.

Dautenhahn, K., Woods, S., Kaouri, C., Walters, M. L., Koay, K. L., & Werry, I. (2005). What is a robot companion - Friend, assistant or butler? 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS , 1488–1493.

EPSRC. (2011). Principles of robotics: Regulating Robots in the Real World. Engineering and Physical Science Research Council, EPSRC .

Hiroi, Y., & Ito, A. (2009). Effect of the Size Factor on Psychological Threat of a Mobile Robot. Kansei Engineering, 8 (March 2014), 39–46.

Huang, W., Olson, J. S., & Olson, G. M. (2002). Camera angle affects dom-inance in video-mediated communication. Proceedings of the Conference on Human factors in computer systems in Computing Systems, 716–717. McKnight, P., & Najab, J. (2010). Mann-Whitney U Test. Corsini Encyclopedia

of Pscychology.

Milgram, S. (1963). Behavioral Study of obedience. The Journal of Abnormal and Social Psychology, 67(4), 371–378.

Miller, K. W. (2010). It’s Not Nice to Fool Humans. IT Professional , 12(1), 51–52.

Mori, M. (2012). The uncanny valley. IEEE Robotics and Automation Maga-zine, 19 (2), 98–100.

(34)

Nomura, T., Syrdal, D. S., & Dautenhahn, K. (2015). Differences on Social Acceptance of Humanoid Robots between Japan and the UK. 4th Inter-national Symposium on New Frontiers in Human-Robot Interaction. Obaid, M., Sandoval, E. B., Z lotowski, J., Moltchanova, E., Basedow, C. A., &

Bartneck, C. (2016). Stop ! That is Close Enough . How Body Postures Influence Human-Robot Proximity. 25th IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN), 354–361. Rau, P. L. P., Li, Y., & Li, D. (2010). A cross-cultural study: Effect of robot

appearance and task. International Journal of Social Robotics, 2 (2), 175– 186.

Riek, L. (2012). Wizard of Oz Studies in HRI: A Systematic Review and New Reporting Guidelines. Journal of Human-Robot Interaction, 1 (1), 119–136.

Robins, B., Dautenhahn, K., Boekhorst, R. T., & Billard, A. (2005). Robotic assistants in therapy and education of children with autism: Can a small humanoid robot help encourage social interaction skills? Universal Access in the Information Society, 4 (2), 105–120.

Sandoval, E. B., Brandstetter, J., & Bartneck, C. (2016). Can a robot bribe a human? The measurement of the negative side of reciprocity in human robot interaction. ACM/IEEE International Conference on Human-Robot Interaction, 2016-April , 117–124.

Tanaka, F., Isshiki, K., Takahashi, F., Uekusa, M., Sei, R., & Hayashi, K. (2015). Pepper learns together with children: Development of an edu-cational application. IEEE-RAS International Conference on Humanoid Robots, 2015-Decem, 270–275.

Thellman, S., Silvervarg, A., Gulz, A., & Ziemke, T. (2016). Physical vs. Virtual agent embodiment and effects on social interaction. Lecture Notes in Com-puter Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10011 LNAI , 412–415.

Turner, J. (1993). Using Likert Scales in L2 Research . Another Researcher Comments. Teachers of English to Speakers of Other Languages , Inc . ( TESOL ), 27 (4), 736–739.

Walters, M., Dautenhahn, K., Woods, S., Koay, K., Te Boekhorst, R., & Lee, D. (2006). Exploratory studies on social spaces between humans and a mechanical-looking robot. Connection Science, 18 (4), 429–439.

(35)

Appendix

(36)

Figur

Updating...

Relaterade ämnen :